I try to keep track
of time. I take no responsibility of the contents here (expressed thoughts)
that sometimes are simply carbage.
2013
May 2 Documenting the BACKUP & preparing for Hyytiälä
The PDF.
The 2006, 2007, and 2012 UCD images have a 1:2.9 pan-sharpening
ratio.
It takes a lot of storage not use the TIFFs but the RAW-versions as there
are also the many levels of processing of digital aerial images.
90 CCW, 270 CCW - who decides how the raster images are rotated or flipped?
As if there is not enough trouble from
the use of omega, phi and kappa - rotations.
If al goes well, this area in Siikaneva will be scanned in the very
near future. To assure 67% forward overlap, flying height
was set at 250 m AGL.
What to add to the kuvamitt HDR-files?
Now, the hdr-files only have a code for the color model in the raw-image
file. 1 x uint8, 3 x uint8, 4 x unit8,
1 x uint16, 4 x uint16. There is no information on the histograms of the
images for a better display of the
data on the screen. This is an issue in particular with the > 8 int per
pixel images.
Another issue is to select, when there are more than three channels, those
three that are displayed on the screen,
and mapped to R, G and B.
The percentiles 1, 5, 95 and 99% per channel could be computed, and the
8-bit RGB-image for example streched
between the 1% and 99% points, linear mapping.
Students, who needs them?
It is nice to do research and to be a research-oriented professor,
but, at some point, someone somewhere may
ask you: "Ok, that's nice, but where are the people that you teach all this?".
The way recruitment works at Universities
results in tenure staff, who's primary interest may not be the production
of degrees - degrees that are useful in the
real life. The research can start to live its own life, 10 ft in the air,
just like admin most often does. O'boy....
April 26, measuring LAD with close-range photogrammetry
Realized that the image files untiled from TIFFs had a zero
pixel at locations
corresponding to ImageWidth+1, as well as one extra zero pixel at the end
of the file.
Had to resort to Hex editor, which I haven't done in a while. When writing
a pixel to
the BIP file, there was a leak of one pixel (W instead of W-1), in each
tile reaching and
extending the max width of the output, untiled image. Arghh!
If there is a way of applying photogrammetry, I'm the first to volunteer
and show up.
We would need to know how the leaves in some plants are oriented. Leaves
are never
perfectly planar, but those that interest us, are pretty much so. Thus,
if three points
on the leaf surface can be measured in 3D, one can fit a plane thru the
points. It's normal
then telss about the inclination angle (between it and the zenith, 0...90
degrees).
We would like to visit a stand of this rather low vegetation (0.6-1.5 m
high), and just take
some convergent photographs from different directions, say, in an azimuth
range of 100
degrees and pointing the camera down about 60 degrees.
By measuring conjugate points between the images, the relative orientation
is known
and the 3D data is of wrong scale, and in arbitrary 3D location and attitude.
For us, the
scale is irrelevant as we are only interested in the inclination angle.
Similarly, the location
is irrelevant, but the attitude is essential. We would need to know the
rotation of each camera
such that the whole model is turned to be alingned with the plumb line.
Now, the Q is:
what sort information must we have available? I.e. what sort of targets
are needed in the scene?
Are 3D distances needed? Points with the same Z? points aligned with the
Z-axis (same X and Y)?
points sitting on the same 3D plane? 3D points of a known 3D triangle?
etc.
And then there's the problem from wind. If the leaves are not immobile,
3D ray intersection
will fail, as we cannot use a single camera that quickly at different view
positions. Calm wind
is needed indeed. And how to see details on the leaves? We are planning
to use paint!
April 24, Preparations for the summer
* The call for the two tenders went out. Siikaneva and Hyytiälä
- and we accepted them / except that the AT May images will be done later.
LiDAR campaigns are now fixed for this summer.
* Maps etc. all done for the summer 2013 campaign.
* TBD: PP sent the coordinates of the plots.
* A review for a RS journal was completed.
* Working on the Level-1/2/3 data for the 2008, 2009, 2010 and 2012 images.
Programming needed for untiling and RAW
output of the 8, 16, 32, 48 and 64 bit TIFFs of DMC, UC-D and UC-Xp.
Storing the 2ADS40 L1 corrected images as well
as there is no guarantee that Xpro will be around in the future.
The 2009 DMC L2/L3 were finalized.
April 19, snow is melting from below the skis
We designed a new 'tree label', that has the uncertainty
of the positioning depicted by an error ellipse. The original
eigenvectors (Q-matrix) is missing, so the ellipse is drawn with sigma(x)
and sigma(y), lacking the right rotation. The local tree
map has a minimum of six trees and a maximum size of 14 x 14 m. It is
tested from 2 x 2, 4 x 4, up to 14 x 14 m,
when searching for the neighboring trees to be mapped..
Some 10-30 cm left in places; skiing continued ok till
Monday, and the last drops of snow made 'skiing' possible Wed, April 17.
Had to walk on skies at times, so I guess the season is over for the
most manic skiers too :) April, 18; was still on my skies.
April 18, signs of spring.
April 17, The right purchase for this winter. Pyry Hankisukset. The
spruce seeds and wings have broken up, mostly.
Planning the LiDAR and field surveys of summer 2013; (spesific leaf
area, LAI, LAD) - all new types of measurements awaiting, which
of course is exciting. Going to try measurements of the LAD, using close-range
photogrammetry! Aeroplan 70 in use for the planning
of the 2013 LiDAR surveys. AH now looks at the field campaign preparations.
An optional 'Siikaneva LiDAR thing' lift its nose.
A thing to remember: When you submit an article, it may be returned
to you, and it only shows (notification)
in the web-based system. Who checks them on daily basis?
The simulation manuscript went to RSE, and another to the RS Letters.
Looking at the image data of 2012 and 2010 for the frame-image-signal-anisotropy
study. .
The
2012 image blocks were taken over a period of 100 minutes. The later images
have 10-15% higher DN-values, on first quick-and-dirty look.
The haze in the 2010 images shows in the SE corner of the image block.
The images were taken to document
the snow breaks of winter '09-'10.
April 11, Att flata på skara
April
11. The 'skara-season' was almost 4 weeks this year.
2013 Hyytiälä field campaign got started - work on the plot
details remains. May 13 we start.
LMS-Q680i (LiteMapper 6800) again in Hyytiälä? The 'Zeppelin
campaign' is scheduled for April 27 -- May 13.
Final
planning of the LiDAR campaigns 2013 is on the way.
Hannover
(ISPRS May 24)
paper was prepared.
Simulator paper was revised one more last time and now the submission
awaits.
April 8 Field season is approaching
We are working on the WF-simulator-paper to finalize
it for the language check and submission.
The revised ISPRS paper was returned from the language check and
returned to the editor.
Summer 2013 workers are now enlisted. Work plan preparation remains.
Started.
Looking for the option of using the Litemapper 6800 for leaf-off
mapping in Hyytiälä in addition
to the WF sensor in the summer.
Hannover paper needs to be prepared
- yes - a full paper is required.
The skiing conditions remain great. Sunday April 7, the temperature
at sunrise was -8C, in Mickelsböle Borgå.
April 8, fresh snow and -3C by the highway 6/7 in Sipoo.
April 2, What a winter!
Snowboarding
as late as March 31, a pine at a Hakunila pine bog. 60 cm of snow on April 1 in Kulloo!
Spruce seeds have departed; so have those in a bog pine.
Hakunila
landscape and a skier, April 2.
Cirsium arvense. Alnus incana about to flower.
How to know where the antenna is pointed ? - use a surveillance
camera (OH6KZP admiring)! A lot of gadgets are needed to run a
radio contest these days.
You cannot do without an engineer, and having two is
beneficial (WPX SSB @ OH5Z).
March 27 Digging archives, archiving, hiring people, installing
old software that Win7 dislikes
'PAN-sharpening gadgets' provide virtual views, where the wide
spectral range image has (in this example, 1:4) smaller GSD compared
to the multispectral image.
This exmaple shows 2-km data with 20 and 80 cm pixels (DMC I camera).
TBD: figure out the IO/EO for the L-2 Pan and Lr4
images.
The submission was finally done for the directional paper.
Seeking workers for the 3-month field project in Hyytiälä.
Backup's done after the 2-month hectic period. Win7 would not
compile the old VisStudio 6.0 C/C++ project for the TIFF to
raw conversion (DMC images). Needed to turn on the old XP machine
that was last used two years ago, and which was actively used
in 2008-2010. Win7 would run the code, in XP SP3 compatibility
mode. The VB6 pascaldll.dll caused also hairloss. You need all
the linked dlls to be present when the regsvr32 registers the
'main dll'.
The 4 TByte box is being filled with data and the documentation
continues. DMC 2009 images were recovered, all the 8-bit versions.
With them, the total number of aerial images (frames) stored from
Hyytiälä is 1555! In some places, there are over 300 views
per
target, 1946-2012.
On May 31, 2009, the DMC was flown twice above Hyytiälä,
at 08 and 11 hours.
The left mosaic has the noon 2 km images (3 lines) and thhat on
the right is 3 km (two lines)
March 22 Another try
The 'well matured, teflon coated' manuscript was
shortened further and tuned for submission to a non-remote-sensing journal.
HDRF vs. BRF at 0-degree phase angle --- system waveform
-talks & looking at the reference values available.
Finding anyone with hard evidence on backscatter R seems tricky.
Timetables to be sorted out for 2013-2014.
The revision of the 'silhouettet article' went to language review.
Great skiing continues in Kulloo,with 60 cm of snow and daytime
temperatures staying below freezing.
Taught Viljo how to skate on skies. He's a quick learner and
with good physical coordination - something
taught by the wrestling exercises.
March 19 Sun 'entertaining' day and nite
The March weather is as good as it can get. 50+ cm
of snow, bright sunny. The best visible aurora in years on
the evening of March 18, unfortunately the 'best' geomagnetic
disturbances had occurred when it was still daylight here.
The k index, measuring the three-hour dynamics of the Earth's
magnetic field, reached the value of 6. A coronal mass 'blast'
had occurred a few days earlier.
March 15-16: With Hasselblad included now at 1421 frame images,
Metsämiesten säätiö, Exclusive seminar - lessons
learned
The IO and EO of the 2011 November Hasselblad images
that were taken during the Riegl LMS-Q680i campaign were
resolved and turned into HDR-files. The final values were in
mm 35.0683, x: 0.0942, y: 0.0630, for focal length and
ppa-values. The image array has over 8000 x 6000 RGB pixels and
pixel size is 6 um.
This CSV
file now has the IO and EO information of all 1421 airborne frame images
from Hyytiälä, 1946-2012.
On top of that there are the line sensor images from 2008.
Image pair from Nov 15, 2011 (Hasselblad, 750 m agl) and May
2, 2012 (Nikon, 2 m agl). The same car is seen parked, but not
quite in the same position :) Despite the noise in the Hasselblad
images, image matching works reasonably, as it did for
the AAT by Match-AT.
Exclusice event on Wednesday proved the old rule: 'Value of a
scientific meeting is often in inverse relationship to the distance travelled!'
Learned new terms: Phytoelement, Bi-Lambertian;
about real refl. anisotropy vs. modeled; leaf angles; animated
reporting of scientific activities; LAI and clumping, Polarzation
of light etc. All this in a single day, 500 m from the office!
We received '20 k€ good news' as Metsämiesten säätiö
decided to invest in the Hyytiälä experiment - we can re-measure
the
permanent forest plots; establish a few new plots; have the GCPs
maintained, set up some sites of field photography in seedling stands
and
below canopies of broadleaved trees; etc.
We are a bit puzzled as to how to deal with 'reflectance' in
the case of coherent
LiDAR signals, and backscattering. Lambertian
assumption does not seem
to hold even for a Spectralon® target,
when being illuminated by a laser signal at lambda of 1.06 um. There
most likely is a specular com-
ponent at the leaf-level, especially for wax-coated phytoelements.
Polarization
of light is there too, we learned, and the reflectance is not
a scalar term, but should be seen as a 4x4 matrix, if polarization
would be considered (sky is polarized).
Could a surface be made of micro-scale corner reflectors?
Now, how to know what is the expected W/sr intensity field towards
the receiver, if illumination of the target has some W/m2 instantaneous
irradiance across the footprint that meets the scatterers at
some angle of incidence? If the emitted pulse meets a corner reflector,
we for sure have
the photodiode choking on the signal, and the BRF-value is way
beyond 1. If it meets something made of black substance, the signal
is zero,
independent of how the surface looks like in any scale - is it,
perhaps? How about calibration of signals using standard surfaces, found
in the
scene. Then we'd be able to say, 'amplitude at value N corresponds
to an average from heathy canopies of Rubus idaeus, late summer phenology',
or 'N corresponds to signal obtained from the planar parts of
cars painted in white, that were detected from the scene using simoultaneous
RGB
imaging' (here we'd assume that car industry uses the same type
of white color, that ages at some pace, aging being a stationary process).
This may
sound strange, but there seems to be no guarantee that reflectance
calibration targets can be constructed Lambertian, nor measured for
the accurate
BRDF around the 0-degree phase angle at high angles (any) of
incidence. So, why not calibrate with standard vegetation, statistically,
over all the
noise (and trust that the response is linear to P received through
the aperture)? Ok, footprint size matters,....
So, could it be possible to estimate the system response using
flat surfaces, and then simply scale the amplitudes to match (real vs.
simulated)
in some vegetation type (say Rubus Idaeus), and test if the signals
match in the other vegetation types? If they do, as well as the WF feature
distribution metrics, should we not have the simulator properly
calibrated (for its receiver and the geom.-optical properties of simulated
vegetation)?
Now, here is a potential pit. Building on a long chain of assumptions
is like building a house of cards, without access to glue or tape. The
needed
redundancy (glue) will come from added information: measured
inclination angle distributions of leaves, improved knowledge of the
leaf-level reflectance
behaviour, use of multiple LiDAR data sets providing some degrees
of freedom (de-correlation) and adding more diverse vegetation types
to the game.
It seems that we are back at reading the '2006/2010 W-papers'.
March 12, Giving in
There's almost always much sense in the feedback
that you receive, and a teflon coated attitude won't help. This time,
once again, we realized that the comments that we received are
valid, although in the beginning I felt quite the
opposite (the usual, light, healthy, humane omnipotence :) ).
If a reader misses a point in your report, because
the report fails to be clear, it isn't his/her fault.
A geodetic GNSS receiver (antenna) that follows large-scale deformations
needs to sit on a firm rock, view a free horizon above 15 degrees,
and
have access to 230 VAC. Basically it needs an opening.
These three criteria hardly ever meet in one place. My few cents.
L-review by April 4, DL-extension April 8.
Giving in, trying anisotropy in an isotropic journal.
OH-ES meeting March 13, prepared a slide-rich talk (pdf).
March 7 Sneeking a peak
Have you sometimes seen something that makes no sense?
March 4, MCRT takes firm steps
Feb 28 How to remain objective?
Today I faced a situation, where clearly a conflict
of interest was observable. Whether it was due me being biased
in what I wrote, or the reviewer not liking what I wrote, and
how I used a commercial system in an example.
Subconsciusly, I guess that we can fail to remain objective.
Parroting 'I'm objective, I'm objective' and you may
start to believe in that no matter how you actually are. And,
it is never possible to reach a pure level of neutrality.
Science should however, strive for that, I suppose. But, when
research money is scarce, you may easily start
to retreat from the noble principless. This week the discussion
in Finland has been around the Pekka Himanen -case.
Well, seen that, been there. A great deal of the public funding
to R&D and science is political, artificial respiration
type , or hidden business subsidies ; under the veil of scientific
activities.
Well, again, taxpayer money has its roots in the business life,
so stop whining. Just play the game and close your
eyes from any possible inefficiency of public spending.
Worked on the rebuttal.
Finally some light shed on the RX. The (i)FOV might also be an
interesting parameter to look at in the simulations
of signals from the vegetation. And, yes indeed, if the sensor's
RX sees a large footprint, the 'wanted' irradiance field
is much smaller than the viewed patch that might receive solar
illumination at an E of 1000 W/m2 (VIS-NIR). Now,
depending on how much it is possible to spectrally filter from
this E(Sun), the E-ratio is affected, and so is the SNR.
We think (probably too optimistically) that we have caught our
tails that we were running after for so long, funny, how the
self-evident can escape you, as something being self-evident
is so hard to consider, in everyday life, not to mention
multidiscipnary research, where ultimately you face this issue
all the time as one dives (tries to grasp) the other
'personally unexplored' domains. Huh.
Our thoughts on 'the mirror effect' - we've been on the right
tracks here too. If the iFOV looking thru the mirror towards
the area being illuminated is circular, on the sensor side of
the mirror, it is circular once projected via the planar lens
that has to be used. However, while the photons take their 2-way
path, the mirror doesn't wait patiently, but makes a turn
and the iFOV is repostioned on the ground. For a scan freq of
30 Hz, and a PRF of 100 kHz, the angular movement, if
linear in time, for a scan angle of +/- 15 degrees, is 30 * 15
* 2 = 900 degrees / 100,000 = 0.144 mrad / 10 us. If the
scanning is from an altitude of 2 km, the 2-way distance is 4
km, the time it takes for the pulse to travel is 13 us. During
that time the mirror moves 0.17 mrad. From 5 km, it takes 33
us, and mirror departs 0.5 mrad. For low altitude scanning
at PRF of 100 kHz, and mirror freq of 75 Hz, multiple pulses
in the air, simoultaneously, from 1 km, travel time is 6.7 us and
mirror moves 0.36 mrad / 10 us. The within-iFOV weight function
is preferably even (but it probably
isn't) , or, the linear
dependency in the photon detector is distorted (think of acclerations).
Similarly, depending on the angle of incidence
(detector's line-of-sight vs. mirror normal), the elliptical
nature of the iFOV may change within the scan pattern (that's carbage).
Now, this effect is probably very small. All considered, the
iFOV cannot be very small, or it misses the footprint, but on
the other hand, it cannot be very large, unless some spectral
filtering is applied to get rid of daytime sun-photons..
Feb 27 Leaning towards number crunching, when analytic solution
escapes.
Peer reviewed for a RS journal, and struggled in
understanding the MCRT LiDAR simulator. We decided to
extend the debugging to cases (of vegetation, scene geometry/radiometry)
for which it is very difficult to derive
the analytic solution for the radiant intensity W/sr intersected
by the receiver aperture. This happens when
there are several surfaces in the scene of varying size and orientation.
The analytic solution can be approximated by a more computer
intensive version of forward ray tracing, where the intersection
splits to a large number (M) of rays sampling the Lambertian
reflectance and transmittance (accounting for
absortion). This assures that the hemisphere above the vegetation
will receive billions of rays, if we continue to
high-order scattering. Now, we can bring the aperture with area
A1, a bit closer distance, r, and sample with
it the radiant intensity (at distance r it converts to an irradiance
field W/m2), which is captured by the aperture
and the received power is P (W). If the area of the hemisphere
is 2 ⋅ π ⋅ r2. and r is 50 m, and the aperture
diameter is 0.5 m, the areas become 15700 m2 vs. 0.196 m2; 1:80000.
If the rays split to 1E9 rays eventually hitting the hemisphere,
the aperture will sample some 12500 of these, and
each of them carries a proportion of power, of the initial ray(s)
transmitted to the scene. The aperture converts to solid angle
and by defining a power for each ray transmitted, we can deduce
the radiant intensity in W/sr, and compare that with
the value obtained using the method that reduces the computational
complexity by using the sensor-rays & ray that
was either reflected or transmitted. This can be done separately
for 1st, 2nd, 3rd etc. order scattering.
Transmittance calls for careful implementation, we learned today.
We also learned that, yes, indeed the digitizer circuits take
turns in processing the WFs, and the I/O module limits the output
to 80 MByte/sec. Because the sample is 256+400 bytes, there's
a limit at 120 kHz in our sensors, restricted by the I/O
board feeding the BUS. Learned also a bit about photon counting.
This is a technique, where the WF noise is just positive, as there
cannot be any negative photon counts, whereas with our sensor,
the oscilloscope is fed a voltage from mVs to several Vs and noise
is made of oscillation around the offset+dark_current + (solar_scattering_through_aperture+electronic_noise_in_circuits+quantization_noise)
The NI-buddies taking turns are identical twins, and their bit-wise
response to the voltage pattern is similar.
Thus no worries of having some autocorrelation between the response
of every 2nd WF.
Feb 26
Feb 25 Laser 'Physics'
With AH we went thru the physics of light propagation
in LiDAR, to make me understand what goes in that process, for
us to report the simulator with the right terms and logics. From
the transmitter, thru the collimator, deflecting mechanism,
via the atmosphere, to the target, and back to the receiver and
its aperture. Photons are transmitted and received, ultimately.
In our case they are collimated, in phase, and energy of photons
is that of lambda at 1064 nm.
The transmitter sends a collimated pulse with a total energy
measured in [J]. It is of very short duration, the power [W] profile
over time t, has a raise
and a fall off pattern. Instantaneous power thus peaks at some
tpeak. In space, the pulse in a 'truncated
cone' filled with photons and
the density peaks in the middle somewhere, if the pulse shape
with respect to time is symmetric, and the irradiance field [W/m2]
across the cross-section is Gaussian.
Image source: http://www.olympusmicro.com, jolisfukyu.tokai-sc.jaea.go.jp
The irradiance field [W/m2] changes with
distance R. It reduces to
its second power. If a planar target is intersected, that is
rather smooth (notice that speckle is unavoidable at 1064 nm),
and the surface is larger than the footprint, and its reflectivity
and directional pattern of it, are stationary and homogenous,
we have a certain proportion of the photons hitting the illuminated
area reflected towards the sensor, which is, in spherical coordinates,
in direction (θ,
φ). The receiver has an aperture, circular,
quite far away, actually also at distance R, since the transmitter and receiver
are close to each other, even using the same deflecting
mirror in some cases. Now, if the reflectivity of the surface
is 0.5, it means that in total, 50% of the photons reflect towards the
directions
that are characterized by the BRDF of the surface. The direction
that matters (θ,
φ), is
the same for the illumination and the receiving. It is
the hot-spot
reflectance, which is due to the
monostatic configuration of LiDAR (if we omit forward scattering photons
paths to the aperture).
Now, the transmitted beam is very narrow and at each point in
the footprint, all incoming photons arrive from the same direction. This
is why the concept of
BRDF (although in reality not measurable), is useful here. In
addition, since the receiver is so far away, the interesting reflected
energy
cone is also very narrow. Thus BRDF, and in the hot-spot geometry
only, are important. Recall that we now assume a flat homogenous surface,
which
is larger than the footprint. The instantaneous footprint irradiance,
I, is measured in [W/m2].
Thus,
for a footprint area of A [m2],
the power received
is P = I x A, in [W]. The BRDF says that towards (θ,
φ)
we have (an instantaneous) radiant intensity [W/sr],
instantaneous power towards a solid
angle in the direction of the sensor. Now,
this intensity field is captured by the receiver's aperture at quite
some distance (R). The aperture
has an area, Aaperture. It corresponds to a solid
angle (A no the surface of a sphere, at the end of radius R), and we just integrate over the radiant
intensity field (W/sr), with the solid angle defined by the aperture
(the intensity field has somewhat attenuated because of the atmosphere).
This way we know the power P entering the aperture at time point
t, i.e. the instantaneous
power.
If we hit a surface larger than the footprint, which is perpendicular
to the pulse path, things are as depicted above.
If the surface is inclined, the returning pulse is extended in
time and the received peak power will be reduced, because
of the fact that the pulse hits an inclined surface, and because
the surface orientation now is different and the hot-spot
reflectivity may have changed. However, it does not matter what
the divergence of the transmitted pulse is, the same
energy [J] is ultimately received. independent of the beam
divergence.
Now, if the surface being illuminated is for example a wire,
or anything smaller than the footprint, the beam divergence
and footprint size start to matter and influence the total energy
received . And in vegetation, these kind of surfaces dominate;
and they are also distributed in a volume. The vegetation surfaces
have high NIR reflectance, but they also have high transmittance
(let light thru and redistribute it in space), as well as absorb
some NIR energy.
At some time point t,
the instantaneous power integrated by the receiver is contributed to
photon paths scattering in the
foliage and finally contributing to an irradiance field, entering
the aperture. The figure scale confuses as does the fact that the
receiver is not next to the light source. All this stuff, that
was 'new to us', has been explained in literature by e.g.
Wagner et al. 2006.
Started to work on the ISPRS resubmittal, agreed to review for
an IEEE journal.
Review notification
of Editor decision, where is publishing going? Is it going "Boink",
as ISBN: 0-8362-1878-7?
Recently I did a normal review of a paper, where I ultimately
did not propose if that paper should be accepted, revised or rejected.
I left
it for the Editor to decide. Just did what I usually do, a review
that I'd like to get. That points out the strengths and weaknesses or
potential
errors in the many parts of a typical article. I have had the
tendency of proposing lots of recommandations to reject/return for an
entire revision.
That does not feel good, but if a paper does not put forward
the state-of-the-art in a discipline, why would we want to add it to
the bulk of reports?
---------- letter ---------
...
Reject
The
author decision letter and reviewer reports can be found below.
I
must say that I am truly appreciative of the reviews the two of you
did. This is the best set of reviews
I have received in a long time. I'm sure that the authors will
benefit very greatly from them. I wish all
reviewers would be so conscientious---the quality of scientific
publishing (and science itself) would be
improved. We appreciate your time and effort in reviewing this
paper and greatly value your assistance as
a reviewer for ...
----------letter -------------
Now, if an editor says this, for a standard review that evaluates
introduction's validity, soundness of M&M and results, what is the
situation over all? Well, simply, all sorts of garbage is being
published these days (maybe not in all disciplines, say like in medical
and food sciences) and you have to review a paper again, when
reading it.
Could this be an option? Peerage of science.
February activities: School's skiing; Viljo at snow-board school,
Tassu follows you whereevr your skiis take you;
First beams of Sunlight after
a very dark Dec-Feb; White-tail-dear-hunters Tassu & Tiltu;
Lactuca muralis.
Feb 20 Resuming slowly
Three weeks has passed and things are starting to
calm down gradually. Thoughts of gratitude fill my mind. Many practical
things need to be arranged in remote mode, 160 km away from home.
I don't know how all this would be feasible without the
internet and e-mail; and help from friends.
Reading has been the main activity: AH send the manuscript on
the 11th, and have been revising & discussing it
An interesting peer review was done for the CEA journal. The
ISPRS submission for late Dec returned and we
need to do some revising to have it published. That was good
news. The aerial images from Nov 2011 were oriented
with Match-AT at Finnmap, and to my surprise, the AAT was succesful
despite the noisy images. 2.7 um RMSE
and cm-level residuals at some 50 GCPs. At times also contributing
to the 'Hyytiälä-experiment-pdf', because it can
be chopped into small tasks that can be done in rather short
time slots. The RSE revision is still pending, not because
I want to avoid doint it, but because I think that rushing things
is really not wise with it.
Even cleared my Webmail from all stored msgs 2003-2010, keeping
only those that might at some point be interesting
to look a. It is strange, in the past we'd file alot on paper,
incl.all correspondence. In that way some documentation
would be available, at least when you threw away all the old
folders. My Q is. What remains, in general, for the future
people to look at? Even this Weblog is just some 1.2 x 8 million
bits on some disk of the Unix cluster in Helsinki, and
surely not of much interest, but what about all news, official
documents etc.?
Feb 5..6 Change in the course of events
Lots of activity in SW Finland. Resuming (trying
to) work on Feb 4. Documenting the experiment (pdf).
This is the AOI of the 2012, repeated, fixed-exposure, UltraCAM-D
1.5-km photography with three parallel N-S-oriented strips and a perpendicular
one.
Two such blocks were repeated. Surprisingly, the Level2 images
have a different rotation compared to the Level3 images. Now the int./ext.
orientation parameters
and KUVAMITT headers were computed for the Level3 version. This
needs to be corrected.
A quick look at the Lapinkangas August 2007 & July 2012 images.
The lichen patches seem to be most stable, most of the litter of 2007
is not to be recognized anymore in 2012.
The perspectives do not match here, and no 3D intersection was
yet done. The views are larger in 2012, as 357 images covered the 32x30-m
area, while 1053 were needed in 2007.
I hope that we, or someone else can find use for this time-series
fotogrammetrc image data. The 2007 data was used for (Korpela, 2008;
RSE; lichen mapping with LiDAR), and the
2012 imaging was done 'just for fun', as it did not take more
than 2 days of field work.
Jan 31 ..Documenting
The G6 was calibrated in 2007 @ HUT: 2.3 um, f =
7.23 mm (7.3448) , xp = yp ~ 0.0; and (k1, k2, k3, p1, p2) by
the Brown (1976) model. iWitness uses a slight modification for the
model of the radial distortion
(the powers of r are
3, 5, 7), and everything is in mm. By OLS solved the iWitness parameters
and they were k1=3.4530E-03; k2=-6.1517E-05; k3=-7.0916E-07; p1=-8.9787E-05;
p2=1.9454E-04; which do not
differ much at all from the calibrations (the two curves of radial and
tangential distortion) done spring 2012. iWitness allows
the computation of images to pinhole camera images, and the epipolar
geometry was
very accurate in a few tests. The image pair shows leaves of
lingonberry from a distance of about 2 meters and two epipolar lines
for the pine twig. Now, the only remaining issue is the interpolation
of the 1053
images, as a batch-job cannot be used with iWitness.
Jan 27 Documenting
Going thru the Hyytiälä backups I found
this excellent illustration by AH from 2011:
The image has, overlapped images with and without the lens errors
in the Powershot G6.
Jan 23 Documenting
Working on the experiment:
pdf (20 MBytes)
Jan 16 Reading and writing
Completed a peer-review for a RS journal, and commented
on AH's manuscript. LK submitted the study on crown volume estimation.
Got the anisotropy paper parsed together so that it could be
presented to the co-authors for comments and improvement.
We were on Friday (Jan 11) to FGI in Masala, where AH gave a
talk (pdf).
Began the documentation of the
data backups to be put the 8 and 12 TByte disc-arrays.
Documentation files being updated: Word-docx,
pdf
- - - - - - - -
Deadline for the final
construction inspection approaches and we did final tuning of
the documents.
This is how planning began in February 2008, when the grass was
green; We wanted to extend our small house. Soon the old 'carage' was
down and the rock mapped in 3D.
Getting the new roof crest to match the old house was one of
the challenges. Luckily the photogrammetric 3D models (measurements)
proved to be more or less ok.
The only remaining worry are now the underdrain wells, they got
30 cm of dirt on top of them, and, I'm afraid, also some new constellations
on top of that, auts :)
Jan 8 ρ or R?
There's the reflectance, intrinsic, ρ = 0-1. Then there are Rs,
reflectance factors, of varying measurable and conceptual types. We
had some sort of Rs, probably
closest to RHDRF in "our operational reflectance products"
that we investigated.
Jan 4 Reading....shortening...rewriting
It appears that the complexity of the (air-, spaceborne)
image observations, the radiometry and the role of the atmosphere,
the coupled effects, are really difficult to grasp. From
the viewpoint of our application the sources of error, all the
causality, explanations, are secondary. But, in the long run, it will
hopefully be beneficial to try to augment one's
knowledge here, even if it takes some (a lot of) extra effort.
It appears that we have been taking lots of shortcuts on the racetrack
of reflectance calibration. One option
would be just to argument that we used the ADS40/XPro data 'as
is', being the first real photogrammetric ρ-product.
The manuscript is getting shorter at least as I'm less in love
with the text written already six months ago. I'm letting go of figures
and tables. We were not able to show
that anisotropty signatures constitute a substantial aid to species
classification. Having many Tables with all trials to show that is not
that essential - it did not help
and that's it!
Dec 28, 31; Jan 6. Back to the Species classification using anisotropy
-work
Returned to the paper that was idling for nearly
four months and that calls for rewriting. To begin with, I decided to
include a comprehensive
analysis (textbook review, again) of the reflectance anisotropy
and atmospheric modeling, to better understand and report "the forces"
behind the
observed anisotropy.
Back to basics. Reflectance (ρ) is a ratio of in- and outgoing
radiation. L is the outgoing radiance and E is the incoming irradiance
(W/m2). A camera is basically a photometer, and a pixel
measures
band-averaged (sensitivity over a range of spectrum) radiance
from a solid angle defined by the pixel's iFOV. How accurate these L(obs)
measurements are,
depends on the calibration and the stability of the pixel (photometer).
At close range, and in the presence of a single calibrated light source giving direct light only
(dark room),
"we have a goniospectrometer". A surface patch receives irradiance
and some of it exits in the iFOV of the sensor.
The radiance (W
⋅ sr−1⋅
m−2) observed at the sensor L(observed)
contains photons that did not reflect from the target receiving illumination
(irradiance) E. There are photons that scattered
from the atmosphere once towards the sensor, photons that reflected
from other targets and again from the atmosphere, towards the sensor,
and photons that
scattered to another direction from the target and re-scattered
towards the sensor. This constitutes L(unwanted) that needs
to be substracted from the observed L.
What is then the illumination E? A target receives direct light
from the small solid angle towards the Sun (∠0.5º). In addition,
top-of-the-canopy illumination
comprises diffuse light. It includes also photon reflected once
from the ground (anything) and scattered back downwards (coupling). Coniferous
forests and water
bodies are dark, so the coupling probably is of low importance.
When the solar zenith changes, E (W
⋅ m−2 ) changes, with the cosine of the
zenith angle (θs).
Scattering dominates (over absorption) in the atmosphere (VIS−NIR). Aerosol (Mie) scattering
is difficult to predict and has less spectral dependency than molecular
scattering.
Without aerosols, only 4 and 2% of RED and NIR irradiance is
diffuse. Add aerosols to gases and the proportion of diffuse light (downwelling
irradiance) increases being 12 and 8% in RED and
NIR for natural clear-sky conditions. This reduces e.g. the hot-spot
effect, towards more Lambertian behaviour. However, since the phase
function of aerosols shows
strong forward scattering, "most of the diffuse light comes from
directions near that of the Sun". E, the illumination, is thus hemispherical,
with spectral dependencies in the angular
distribution (consider spectral irradiance measurements to different
directions). The literature refers to hemispherical directional reflectance
factors ρHD.
Irradiance from all directions
is convolved and radiance to a certain solid view angle is used
to compute ρ. According to literature clear-sky
atmosphere dampens the hot-spot reflectance and increases the
reflectance in the forward scattering direction, up to 10−30%, in relative terms
(compared to gaseous case). With haze, anisotropy decreases. These observations
apply to a sensor
right at the target, and thus neglect the upwelling radiation
through the atmosphere and consider only the illumination changes.
Now, the thickness of the atmosphere (gases, aerosols) increases
with height. If the sensor is air-
or spaceborne, perturbations exist both for the down- and upwelling
radiation. It
is a situation with many disturbing factors compared to the 'simple'
goniospectrometer-in-a-dark-room
case. L(unwanted) enters the sensor and reflectance factors
increase, especially
towards high θv angles (we see a bowl-like
'polar anistropy cake'). If atmosphere is gaseous only, effects
are symmetric about the xy-plane. In NIR, effects are negligible, and
slightly asymmetric.
With gases and aerosols, a similar increase in L(unwanted)
causes a bowl-effect, but because of asymmetric Mie scattering, there is
more L(unwanted)
in the forward scattering side. There can
even be haze-induced additional hot spots at very large values
of θv (very minor backscatter
peak in aerosol particle phase function). In NIR, again, impacts are
small. In NIR effects of gaseous
and aerosol scattering counterbalance each other?
What is characteristic
of high θv?
The path length is longer. The surface contribution
→
0 at large view zenith angles, and high sensor altitude (VIS). In NIR
it never goes below 0.8. In NIR, the
contribution is almost constant across the xy plane (clear sky).
Asymmetric aerosol phase function primarily affects the forward scattering
side (VIS).
→
forward
radiance peak, smoothing of the natural BRDF, aerosols are low in the
atmosphere →
even
aerial images are impacted. Atmospheric effects increase with view
zenith
angle, optical depth, and are dependent on λ.
Principal plane is most affected by the atmosphere and cross-plane
the least. If surfaces are dark - path radiance dominates. Coniferous
forests have low albedo.
Thoughts on
camera and reflectance calibration
If
a light source used for calibrating a camera has an irradiance output
uncertainty of ± 2.5%, it means that the at-sensor radiance calibration
(and subsequent L measurements) will
have this uncertainty in L(observed). Consider a target
with ρDD at 0.05 (vegetation in VIS).
The ratio L / E is subject to errors (L + ε1) / (E + ε2). It becomes clear
that the maximum
attainable ρ accuracy is limited by the
±2.5% inaccuracy of L observations. On top of this, in a real
remote-sensing case, separation of the L(unwanted) from L(observed)
is
restricted and challenging, as is the estimation of illumination,
Eat-target. Within an imaging campaign, the same L(observed)
measurement (systematic) errors will be observed across all data,
provided that the sensor is stable (integration, shutter speed,
CCD temperature (thermal, shot, read-out noise)). So if the proportion
of L(unwanted) is estimated right, the resulting imagery will
be
free from atmospheric effects showing the real anisotropy due
to targets, but the ρDD will not be
correct in absolute terms. Thus, high demands are set on the estimation
of L(unwanted).
In a frame sensor, there are
millions of pixels in the focal plane, each individually measuring L(observed).
If the camera sees a constant radiance field, the individual pixels
should measure the same iFOV radiance, and we should see the
same DNs across the array of pixel detectors. If the shutter is released
another time, the DNs in the next
image should again be constant across the image and the same
as in the previous image. When the relative aperture is changed but
the shutter speed remains the same,
the DNs just show a global offset. If the dynamic range of the
CCD pixels is e.g. 12 bits, we can have 4096 observation values of
L. Preferably the DN response is linear to L.
Spectral separability and the use of
reflectance anisotropy
There are features that can be used for the species separation
(identification). With image data, it is done based on L or ρ observations that sample the
camera visible crowns.
Aggregated (crown level) L observations change across the focal
plane due to anisotropy of both the targets (shadow-casting + sub-surface
reflectance anisotropy) and the atmosphere.
L observations are thus not invariant to the view-illumination
geometry. That is not a preferable trait of classification features.
If the imaging campaign has short duration, aerosols and gases
are stable, and just E changes slightly with the cosine of solar elevation.
The atmospheric effects may show
asymmetry, but all of it (Lunwanted) is just on top
of Ldesired that shows "the true reflectance nature of trees"
disturbed by the trends in E. Removing the trends in E brings
about some of the 'true reflectance', especially if they
have the same impact in all components of Lobserved. The other
option is to strive for good-quality ρ images.
From previous research we
know that average ρpine ρspruce ρbirch in VIS range do not deviate but
10-20% at max. That is, at the level of single crowns and their sunlit
parts.
The differences, if the previous results hold, are somewhat dependent
on θview, θSun, φview, φSun, and λ. In NIR, the view-illumination
dependency is, if it exists, almost
entirely lost in the noise. The previous results (RSE
article) on anisotropy are namely, in all likelihood, somewhat
contaminated by remnant errors in the used ρ-images.
Thus, whatever analyses we do based on these 'BRDF' patterns,
the results are highly tentative, showing trends but with somewhat right
magnitude of effects.
The idea in our analyses is to show that in the presence of anisotropy
patterns, can we utilize these for an enhanced species discrimination
by considering
the (θview,
θSun,
φview,
φSun)
geometry. Actually we omit θSun, but that's not so relevant
here. We collect for each tree the mean spectra, at the bands available;
mean(red, grn, blue, nir). And we do it for both the sunlit part
of the crown as well as the self-shaded part. This involves a tricky
part: how can we separate between the
two, from altitudes of 3−4 km? And is there a meaning
to it as tree crowns do not form opaque surfaces as required when using
the concept of BRDF. Perhaps
the terms 'on average sunlit' and 'on average self shaded' would
be more appropriate:
Conical-round tree crowns map to the focal plane as illustrated
on the left. From this we see that our assumption on having both sunlit
and shaded part visible to the
camera is an optimistic one that will not hold true for large
values of θview, and/or for very conical crowns.
The figure on the right has a sidelit crown, viewed obliquely and
the relative solar azimuth suggests that the view is almost perpendicular
to the solar principal plane. The tree stands in an opening (what comes
to upper canopy) and occlusions are not
an issue for this tree - we see lots of sampled positions (3D
points mapped to a pixel) on the 'crown surface', down the crown .The
division between sunlit and self-shaded shows in
the color of the points and was determined from the 3D geometry
of the canopy neighborhood as measured by LiDAR. To achieve this illumination
characterization
in practice would be challenging because image matching fails
to provide an accurate canopy reconstruction, and LiDAR data normally
aren't as dense as we had it, at
10−12 pulses m−2. In this respect,
what we do is of more 'theoretical value', or shows upper bounds of
accuracy achievable for species discrimination.
The observed
signals from trees
Now consider the approximately 20 sunlit crown points in the
right sub-figure shown above. From these we can derive a mean(blu,
grn, red, nir) feature vector of ρ-values. They apply to
a certain (θview, φview−φSun) geometry.
That is, the tree is viewed at some view zenith and relative azimuth
angles. What are the four ρ's made of ?
To start, we can say that
ρ ≈ (Lobserved −
Lunwanted)
/ Eat-target
This says that we have converted the observed at-sensor radiance
to a pixelreflectance factor (0−1) by means of reflectance calibration.
In our case we did it with Leica Xpro. Similar
equations apply to the four bands, but for convinience, consider
now only one band. Xpro, or any other refl. calibr. software isn't
entirely error free (in the estimation of the
last two terms), so actually we have
ρ ≈ (Lobserved −
Lunwanted)
/ Eat-target + εrefl_calib
Because the atmospheric effects have their 'BRDF', the errors
(εrefl_calib) in most likelihood are dependent
on the
(θview, φview−φSun) geometry
as well, and strongly on λ,
for that matter. There probably is correlation of the errors
from band to band (across λ), because errors made in the
atmospheric status (gases, aerosols, water vapour) impact similarly
the estimation
(of Lunwanted and Eat-target) in the VIS
range. If we naively assume that εrefl_calib is zero, the ρ's measure real anisotropy in
trees. Or, almost, as we may have introduced
artifacts in the estimation of the illumination conditions (right
sub-figure above). Let's omit these now too.
The term (Lobserved
− Lunwanted) brings
about the radiance in the direction of the sensor, right at the target, as
if the sensor is brought down from the sky and moved about
to measure the whole crown and aggregate the mean L over the
sunlit crown patch. Such data exist, almost, collected in early
1990s in Joensuu by Jääskeläinen et al., who
measured 400−850 nm spectra of trees (pine,
spruce, birch) on the ground level. These data show that if a crown
is 'dark or bright' at 400 nm, with respect to other crowns of
the same species, it was similarly dark or bright across the
spectrum measured. Now this observation has very important implications.
To make use of small between-species differences of ρ or Lobserved (raw image pixel DNs), it would
be favorable that the variables (four image bands in our case) show
small within-species
variation, and would be as orthogonal as possible (i.e. non-correlated).
The observations made by Jääskeläinen et al. (1992),
do not suggest either property, unfortunately. In
general, trees are dark in VIS and bright in NIR, and the level
of brightness is strongly correlated at the tree level, which, I have
understood, is a physical fact and holds true for
opaque artificial and natural surfaces as well. Real spectral
data is noisy too showing large within-class variation.
Now, the image observation for a sunlit crown could also be expressed
as
ρ ≈ ρ(species, θview, φview−φSun, λ) + εrefl_calib(θview, φview−φSun, λ)
ρ(species, θview,
φview−φSun, λ) says that we should expect
some mean (expectance of) ρ for an observation made of
tree belonging to species,
and made in a certain
view-illumination
geometry, and band
(λmin−λmax). The reflectance calibration
errors have their dependencies on the latter two.
Now,
ρ(species, θview,
φview−φSun, λ) is of course just an expectance,
and real data shows large variation around that expectance, and we
can write
ρ ≈ ρ(species, θview, φview−φSun, λ) + noise + εrefl_calib(θview, φview−φSun, λ)
There could also be a term for L-noise (pixel-level, pixel rows,
entire focal plane array), and L-trends (remaining errors of flat fielding,
i.e. removal of lens vignetting effects), but at least the
L trends can be absorbed to the reflectance calibr. error, and
L-noise is of minor concern given the magnitude of the other sources.
The noise term
above is mainly due to trees and our inaccuracy of sampling the mean
spectra of crowns, which is retrieved from a small sample of possibly even
dislocated pixels. Pixels
may actually map to the wrong tree or capture background. And
each pixel certainly has some contribution from the background, even
if it captures the right shoots of the correct tree crown.
And the illumination incident has its variation too, due to shading
and multiple scattering.
The structural variation within a species is thought to be the
main driving force explaining noise. That is also why
noise(blue)
correlates with noise(grn),
and so on. Deviations from the
expectance show correlation in one image (over bands), and across
view directions, i.e. the deviations from the expected ρ of a tree are similar in many
views and bands. If deviation
is caused by neighbors (multiple scattering), we should see correlation
between views (images). Statistically, we can try to split the noise component. To do that,
let's assume that
the observations contain the expected reflectance for a species,
at a band, and in a perticular view-illumination observation geometry,
and noise 'in
one clump'. If we omit the division between sunlit
and shaded illumination classes that both give raise to spectral
mean vectors with four elements, and consider just the sunlit part,
we have, for each image observation of a tree,
a system of four equations:
ρblu ≈ ρ(species, θview, φview−φSun, λ=blu) + noiseblu
ρgrn
≈ ρ(species, θview, φview−φSun, λ=grn) + noisegrn
ρred ≈ ρ(species, θview, φview−φSun, λ=red) + noisered
ρnir ≈ ρ(species, θview, φview−φSun, λ=nir) + noisenir
Now, since each observation is from an image, im = 1...18 (18
views in our data), and ( θview, φview−φSun)
presents two
continuous variables describing the view-illumination
geometry of an observation (we can call them x and y). With four
bands (i), two illumination classes (j) and and three species {pine,
spruce, birch} (k), there are altogether
3⊗(2⊗4)
= 24 models. There are some 16000 trees, and >200000 observations
of a tree in the 18 images.
ρ(tree,
image) | band, species, illumination ≈ ρ(x, y | (species band, illumination)) + noisetree + noiseimage + noiserandom
The part "ρ(x, y | (species band, illumination))" is a polynomial regression
surface in x and y, which are the polar BRDF coordinates.
These are the fixed parts in our mixed affects analysis of variance
model system. The mixed effects include the image, tree and random term.
Their covariance structure
is estimated across the four bands and two illumination classes.
The split of variance tells about the contributions arising from the
four 'sources', anisotropy, image,
tree and the random effect.
The
term/part "ρ(x, y
| (species band, illumination))" says what proportion
of the observed reflectance (ρ) is explained by anisotropy,
and in what way, in terms of the xy coordinates.
If between species differences in directional reflectance are
to be used, the differences "reside" in these xy-surfaces.
An example ρ(x, y)-surface
showing one species, band,
and in direct illumination. x-direction shows forward-backsattering,
y is perpendicular, in degrees.
The noisetree term tells about the correlation
at the tree level, and the variation due to trees. That measures sort
of "biological noise", although it includes for sure some
effects from positioning errors of a tree. If this variance component
is high and correlated between bands, the chances of accurate classification
are reduced.
The noiseimage term tells about remnant reflectance
calibration errors, about offsets between images. No xy-dependency
was included as we could not formulate a justifiable
model. Maybe x (principal plane direction) could be used as a
covariate (in this direction atmospheric BRDF is strongest). This term
s.b. minimal, as an (partial) indication of
succesful reflectance calibration.
The term noiserandom includes the unexplained random
variation. This probably includes noise from pixels not sampling the
crown level reflectance very precisely (as it
is based on just tens of pixels, at max), and other factors such
as asymmetry of the crown (sampling off, in some view direction), etc.
The simulations (Jan 7)
Now, naivly, we think that at-sensor-radiance measurements can
be calibrated to some "invariant surface descriptor" called ρ.
By that we
mean that if we return with the
same sensor to the same site, be it after 5 or 50,000 minutes,
when the solar elevation is more or less the same, we have a means of
producing observations that are
the same for the same targets seen in that illumination, from
a specific view-illumination angle.
But ρtrue
= ρtrue + epsilon, always,
as it is impossible to derive L at-target unambiguosly (it appers,
this is something to find evidence for).
So, in our simulations we, from a statistical and a user point-of-view,assume
certain scenarious for epsilon, and that it entirely arises from the
refl. calibration process. We
will probably not be able to say what 'forces' are exactly behind
these deviations, just some textbook argumentation about the magnitude
and trends.
The ρtrue
is also something
very stochastic, when it comes to the reflectance of a crown belonging
to a tree species. The varying structure being the major source,
and the role of bkgnd in all likelihood is correlated with the
variation of structure at tree and canopy level, at least to some degree
(e.g. think of sparse pine crowns
on a pine bog).
So, we simulate realistic variation and covariance between bands,
in the
ρtrue
and epsilons,
varying the nature of the epsilons.
We assume further that multinormal deviations simulate the real
case.
Because the anisotropy differences
(that we observe in our non-optimal design) are marginal, our results
in species classification in real and simulated
data do not show great enhancement of classification performance,
if anisotropy signatures are considered. This of course is disappointing,
as we had hoped
that 'the BRDF-approach' will provide 'a major improvement'.
BRDF is given lots a 'bandwidth' in RS literature, and therefore the
disappointment is amplified...
Dec 21 The 18 month project came to some conclusion
We made the first canopy-photogrammetry trials in May 2011 with
LK in Hyytiälä. In June 2011 the Lapinkangas forest was photographed
and first lessons learned.
It took AH ten weeks to triangulate these images. May 2012 we
were back in Hyytiälä for leaf-off photography and a long arduous
orientation work followed that. Then,
on several occasions, we were to Hyytiälä to establish
systematic canopy image blocks, during June-July 2012, and weeks were
spent on these in August-September.
Some man-months went to figure out the issues with the LiDAR
data of 2011-2012, and Oct-Dec was the time for analyses and reporting.
Many things were learned
from this WF-LiDAR project that hopefully pay off when we tune
the LiDAR simulator and commence studying species identification in
WF LiDAR data.
Some
of the final phases visualized.
Dec 14 Slowly forward
The pseudoechoes are causing hairloss, as the
routines need to be tuned separately for the Riegl and ALS binary records
/ data structures.
The qualitative analyses, if we wish to include some examples,
isn't that straightforward either, for someone who's quantitative.
The leaf-off
images shows some 'nice phenomena', but are these worth mentioning?
Also, it could be possible to find two-echo pulses, where the two
(targets) are seen in an image pair, or dampening WF tails. But
all this is very difficult to "show". It seems also that the accuracy
does not
allow very accurate (cm-level) penetration (where was the range
detected?) analyses. Maybe it would be possible to find a wide
echo, at the
peak of which there are no scatterers. It would be nice to illustrate
the potential of the photogrammetric approach - maybe just do it qualitatively
although it feels unnatural.
More 'Powerpoint art' that illustrates the geometry of imaging
and capture of silhouettes.
Got the feedback done for LMa.
Realized that the CQ WW SSB QSL-labels are accurate to the weekend
but not to the exact date of the QSO :). PU! interview went to
QTC as well, and is edited for CQ, if all goes well.
This can happen if your receiving antenna is broken, a loose
joint in the 160M inv-L feed point started arcing when I was transmitting
on the 80M dipole.
ACOM 1000 is enough to produce this. Nice Xmas photo?
Lots of snow in Kulloo. Snow depth is 30-35 cm in our garden
and the heat-leaking roof has been cleared already twice as the temperatures
are low
and the snow on the roof melts and builds ice dams at the roof
eaves. Have to consider insulating the old chimney at the old attick.
Dec 13 What is the short-term geometric accuracy of LiDAR?
Extracting echoes from a 3-cable power line. Left
shows Riegl 750 m (Nov 2011) data from three strips and the right has
Leica ALS60 500-m data from 2012.
When the data are oriented in the direction of the cables, the
echoes spread in the perpendilar direction, which is the direction of
the scan pattern (mirror).
The pattern of intensities, as a function of distance from the
cable, would show a bell-shaped curve if everything were perfect: homogenous
cable reflectance,
perfect pulse geometry, and Gaussian PSF -based within footprint
irradiance distribution. But the reality deviates from it.
The ALS60 11 cm footprints show only a +/- 2.5 cm deviation,
but the intensities have almost no dynamics, as they are all either
0 or 1. Unfortunately, the
ALS60 1-3 km data (2010, 2011, 2012) gave raise to zero echoes
from the lines!
Dec 12 Mekrijärvi
seminar (yearly UEF Xmas seminar)
Some 10-12cm of snow allowing
also for some winterly activities at Mekri. Tried to enjoy Mekri as
it still exists in the 'old configuration'. Mekri is the field station
of
UEF, former U Joensuu. In Ilomantsi, 'far away from the civilization'.
Learned, while there, that in 1939, it was a scene of the Winter War.
FMI has a WXstation there (link to on-line data).
Typically, we who came from Helsinki were the ones travelling
the longest distance, but this year we (the UEF kärkihanke) got
visitors from Tarttu, Estonia. The talks, mostly
by PhD students were of high quality, mostly linkig with LiDAR.
Applications ranged from bird ecology to fire control and forestry (ABA,
ITD). Aarne explained his LiDAR simulator and I
got to talk about the silhouettes of individual LiDAR footprints.
Transmittance of needles is something to consider, we learned. Altogether,
we were some 25, foresters (inventory,
remote sensing, management), mathematicians, ecologists, and
computer scientists. Professors, post-docs, PhD students and undergraduates.
First evening we had some
group work around relevant RQs in LiDAR that we could propose
to one another that we then tackled and presented. A good way of meeting
once a year!
I like good quality seminars which are in the same
time zone +/- a few hours....
Dec 8 Mekrijärvi awaits
Prepared the talk for the annual xmas-seminar at UEF (pdf).
"Powerpoint Art".
Skiing season started Dec 8 in Porvoo with 20 cm of snow. Forecast
until Dec 18 says -8°C ... -1°C.
The QSL-labels were printed into pdf-format, prepared some simple
ad-hoc code: VB6_code.html , here's
a sample of the log file (txt). The last 0/1
variable
is 1 for the last QSO with that station (and label).
Dec 2-4 What are the
intensity observations, exactly?
The other day I was wondering as to why we do
not seem to find LiDAR returns with i1 == 0 in well-defined but dark
surfaces,
such as the bitumen or the SMEAR II roof that has the special
absorbing paint. But there are first-of-many ALS60 echoes in
vegetation that have i1 == 0. Is this because intensity somehow
considers the echo width?
The black squares have zero-value intensities (y) at low values
of relative silhouette (x) in the ALS60 data. It appears that in this
case the 750-m LMS-Q680i sensor was more sensitive, in terms
of relative silhouette, but owing to the 5:1 ratio in footprint areas,
the 1-km ALS60 was triggering echoes from smaller objects. The
transmitting power of the ALS60 was reduced to keep the SNR
constant at heights of 500-2750 m. This lowers the sensitivity.
Had to tune the tools for the pseudoecho analyses. Now, we measure
in multiple images the xyz of tip of a branch (in the middle of both
images). The green circles are pulses
retrieved to the elevation (z) of the branch tip, i.e. what we
call pseudoechoes. The number is a z-difference in meters (z_of_first_echo
- elevation_of_branch_tip). The other colors
are for real echoes (in the vicinity, at any height) and their
footprints, HSV-color denotes LiDAR intensity. The (dx, dy) between
the tip and the pseudoecho is stored, as well as the
silhouette of the pseudoecho. Now, the silhouettes s.b. zero
for well non-tangent pulses, while there s.b. more and more pulses with
silhouette as the distance is closer. In polar
representation of (dx, dy), the silhouette pattern s.b. symmetric
with respect to the direction, or, else, the strip matching still has
systematic errors. In the pseudoecho analyses,
we restrict to strips with low zenith angles so that the footprints
are more or less circular at image FOVs (and max zenith angles) of +-25
degrees.
Nov 28 AGC calibration for the 2011 and 2012 Leica ALS60 data
(same sensor) (rewritten Nov 29)
We observe the amplified intensity: I_obs == I_ori * (gain + delta_gain) + offset.
If we assume that offset == 0,
then the calibration for I_ori
is I_ori == I_obs/(gain + delta_gain).
We may assume that delta_gain is something from 0...to max. It is adjusted
by the AGC in the receiver.
When delta_gain == 0, I_ori = I_obs/gain. What we have in the
ALS60 data is a just gain value. We assume now that it is actually
(gain + delta_gain), and
at some small number of registered gain (agc), delta_gain ==>
0. So gain == 1, and when delta_gain == 0 ==> signal in == signal
out.
I_ori is the original signal before amplification, the one that
we wish to know, or calibrate to.
We set the calibration equation to be I_cal == I_obs/ ((a-gain)*b+1)
"(a-gain)*b" is the delta_gain part, and it goes
to zero when (a-gain) =>
0. b says how much the
amplification increases for an increase of gain.
In Hyytiälä, there are four targets with varying NIR
reflectance. Bitumen ~0.05, asphalt ~0.16, fine sand ~0.3, and grass
~0.45. These values
were measured in 2008 with spectrofotometer at 900 nm. In 2012, the footprint irradiance
was kept as constant as possible from all flying heights,
From 500 m the transmitted power was slightly too high. Thus
we separate 500 m from the other heights.
(a) Grass surface. (b) Leica 500, 1000, 2000 and 2750 m data
combined, intensity x maximum amplitude of the waveform. There
is less dynamics in the max_amplitude data.
(c) shows raw intensity
(I_obs), 500-2750 m, as a function of gain (delta_gain) that takes values
from 124 to 148. 20 units in AGC lifts the
intensity signal 250/140 ~1.75 times higher.
Old dry asphalt.
In grass, the gain was ~1.7 per 20 units of AGC, and it seems to hold
in asphalt as well (40..65, gain from 124 to 148, 65/40 ~1.7).
There is something darker even, let's look at the bitumen roofs
that were cleared from all the gravel in the summer of 2008.
The bitumen is
dark in 1064 nm. Here, the raw intensities (I_obs) range from ~7 to
~12 over 20 units of gain, again an increase of about 1.7-fold.
The volley ball field: Fine sand shows an average I_obs intensity
increase from 90 to about 150 in 20 units of AGC. Again about 1.7-fold.
In RSE, 2008, I proposed a model I(cal) = I(raw) + b x I(Raw)
x (c-gain). If we take the grass data, and
fit b and c by constraining with I(cal) > mean(lowest
gain I(raw) data), we get
(the constrain fixes the outputted I(cal) to the level obtained
with the lowest gain):
Now, x-axis has the gain, and y-axis is the intensity, I(cal)
and I(raw) (from grass), and
the minimized parameter was the within-class coefficient of variation,
that reduced from 18% to 9%. Now, clearly the
response isn't quite optimal, as the red I(cal) still show a
trend across the gain range. We need to improve the model. As if a
(-d*gain^2) -kind of a component is missing?
The new calibration equation (proposed by AH) became I_cal == I_obs/ ((a-gain)*b+1)
Now, in the data, the gain minimum was 124, in Hyytiälä,
where the targets are. We may assume that a == 120, that is where the
AGC is at minimum, and no additional gain is
applied, just the basic gain of 1, i.e. signal in == signal out.,
thru the agc-circuit.
Then we solve b by maximizing
the n"et decrease" of the sum of the four per target CVs (%) of I_cal,
decreasing from the CVs of I_raw. We can maximize this total CV reduction
by requiring for example a certain
I_cal level for the darkest class bitumen. The four classes here
are bitumen, asphalt, sand and grass.
(a) shows the 1000-2750-m data for grass. Calibration sets the
average to about 130, which is the value at no-extra-gain (gain ==
120), for grass.
(b) shows "an absolute" calibration, where the reflectance factors
are on the x-axis, and the corresponding mean I_cal intensities
on the y-axis. The lines intersect y-axis at reflectance == 0
below the y == 0. The intensity observations seem to be linear, but
the offset is unknwon, somewhere below the y-axis, a negative
value (intensities are recorded 0...255).
2012, dataset, a,
b
1-3 km 0.029822,
120
500 m 0.073987,
120
2011, dataset, a, b
900 m 0.058593, 122
Some interesting papers:
http://ww.gim-international.com/files/productsurvey_v_pdfdocument_11.pdf
http://www.isprs.org/proceedings/XXXV/congress/comm3/papers/267.pdf
Nov 26 The most exciting
CQ WW CW in years!
An amazing 7200Qs, 12.2M points, The new European
record @ CR2X (Azores) by Kim, OH6KZP.
ÅIF säh.. ups floor hockey, boys '01. P in the middle
happy over the victory.
Nov 22 Attempting to grasp the signals
If we have two well-defined surfaces, of clearly
different reflectance, and a Gaussian emitted pulse shape, the signals
in the
receiver can be understood to comprise of noise and the real
signal. Noise constitutes backscattering from the atmosphere,
noise by the photon detector, amplifiers etc. The shape of the
returning signals (photon surge from the target, two curves) are Gaussian,
and depend on the reflectance. Now, consider the y-axis to be
e.g. in uV, very low voltage from the photon detector and amplifier,
just prior to the A/D conversion. The x-axis is the time
axis.
From the figure above, It is clear that the sensor cannot react
to any positive uV signal.
So, we can mask out the noise by applying a threshold. In the
above image, it blocks the noise (the light blue curve is a goner). The
first peak blue is
still noise. It is too short to be true. But the blue signal
is well above the threshold long enough - it must be a real reflection.
What we see from the dark surface (green) is ugly. Although it
is a well-defined surface, it has multiple peaks. The cure of course
is
to lift the threshold.
This is a Riegl LMS-Q680 example, very little scatterer produces
the first echo, and starts (triggers) waveform recording.
Now, the problem is that we could actually see the noise here.
But, there is no guarantee that the amplitude digitization
is linear at low values. Or, in all, it is just a lot of guesswork
at this point.
We don't know how it really is, but the example shows that if
the gain (uV in vs. uV out) is non-linear, the shape of the returning
pulse (Image) gets to be distorted (from Real).
So, what is the image of the emitted pulse, if we cant trust
its image in received data?
What about the system response, to a new set of photons (or silency),
when the state of the receiver is {"just received five photons", "haven't
received any in a while",
"I'm saturated because of a high surge"}?
Range estimation in vegetation?
If the vegetation has two subsurfaces, and we apply "constant
fraction discrimination", i.e. take the 50% value from maximum apmplitude,
project this to the time axis, and take the same time offset
from the transmitted pulse (time offset between the 50% point and the
maximum),
we can determine range. Now, in vegetation it goes wrong, and
causes 'penetration'?!
Ok, it seems that we still have a few lessons to learn.
In LMS (first strip, 9 degrees off nadir) some 40% of the amplitude
variation was explained by the silhouette (here corrected for dX = -0.5
m , dY = 0.10 m)
Nov 20, Calculating silhouette areas of LiDAR pulse footprints
in canopy images
Started selecting appropriate pulses/images for silhouette computation;
letting the computer to shift the pulses and analyse
the silhouettes using different threshold values, with and without
Gaussian PSF weighting.
Two examples of silhouette images in a 60-yr-old pine forest.
(a) The sphere is 150 mm in diameter, but not necessary
at the Z of the echo (anyhow, we rejected this observation).
(b) The other footprint image shows a few scatterers (pine needles/shoots),
which can be "highlighted" by thresholding the image. (c)
and (d) illustrate the use of two different threshold values.
The result
depends on the binarization.
The idea now is to compute the footprints for a series of threshold
(sky vs. scatterer) values by shifting the pulses (dX, dY) in an offset
grid.
The silhouettes are then computed with and without the Gaussian
PSF of the within-footprint irrandiance trends.
Then, we examine the correlations of relative silhouette and
waveform max amplitude values, and search for the
maximum over the parameters {threshold, dX, dY}. This is done
separately for each strip in every campaign.
The optimal threshold varies, because the pulses from a strip
all come from the "same direction" in the hemisphere,
which had a certain sky-brightness in the images (with some temporal
trend as the photography took 45 minutes in the mng)..
The figure illustrates correlation (y-axis) at different shifts
(+/- 20 cm) in two strips (1 km LiDAR data of 2012). The surfaces show
an x-offset of
approximately 15 cm between the two strips. The weighting with
the Gaussian PSF is improving the correlation, but not significantly,
and in some
cases actually not at all. Maybe the small scale noise in the
mirror (scan angle) position, or actual fluctuations of the PSF are causing
this.
At best, 50-80% of the variation in the maximal amplitude (of
the first echo) is explained by the optimally shifted silhouettes. The
1/e2 footprints are 23 cm wide
in the ALS60 from 1km, 46 cm in 2km data. Correlation analysis
could be carried out for the 2 km data of 2012, and the 1 and 2 km
data of 2011. Too few pulses
per strip were available in the 2.75-km data of 2012.
When the offsets have been found and corrected for (in a strip),
it will be possible to verify the results using pseudoechoes, that
measure the empty space around targets at a certain elevation.
These pseudoecho-footprints were drawn at the
elevation of the pine shoot in the middle of the stereo pair.
Pseudoechoes are drawn using pulses that have penetrated the upper
canopy without producing an observation . They shoud intersect
only very little, if any, as they represent
1/e2 footprints.
How many photos is 24 fJ and do we see the solar background
noise fluctuations in the noise of waveforms?
In (daytime) laser scanning, the photon detector of the receiver
sees the world below thru a FOV, or iFOV. There is also a bandpass filter
(BPF),
that filters unwanted photons from being detected by the photon
counter. In a LiDAR that uses monochromatic light, e.g. at 1064 nanometers,
it is preferrable that the BPF is narrow, and attenuates as little
as possible the signals at 1064 nm. On a sunny midday, the solar
irrandiance (300..2500 nm), integrated, is apprx. 1000 W/m2.
At 1064 nm, the spectral irradiance is ~0.7 W/m2/nm. If the BPF in a
LiDAR
sees 50 nm (box-filter), we have ~35 W/m2 of background irradiance,
in a 1 m2 footprint. If that footprint views reflective material, the
signals at the
sensor can be significant. If the FOV sees 100 m2, we have 3500
W of spectral at-target spectral irradiance "in the FOV footprint". How
much is
the at-target irradiance of the LIDAR pulse? If the sensors uses
10A at 28VDC, and transmits photons 0.1% of the time, with an efficiency
of 10%,
the average power during a pulse is (10A x 28V x 0.1)/0.001,
which is 28 kW. All this power is concentrated in the footprint. And
the ratio
of solar irradiance seen by the FOV vs. pulse irradiance in the
footprint would be 3500W : 28000W. It is only a difference of 10dB!!
Cant be! Seemingly,
the FOV has to be smaller, or the BPF is narrower, or the pulse
peak power is larger than 28 kW. We would like to get the big picture
right here,
so there remains some studying.
Somewhere we read that the signal at the LiDAR has an energy
of 24 femtoJoules (some case in literature). How many photons is this,
at 1064 nm?
Nov 16, Scientific Writing Process
The documentation of scientific activities aims at
dissemination of information, and a description of the activities that
led to the
possible conclusions made. The reader is convinced with the text
and other representation. A typical article is 6-10 pages, which
means that you need to balance between details and length, condense
and present only the essential. That is painfull at times,
as your readers have so different background and missing details
may annoy as well as 'narration' of, to him/her, self-evident facts.
Citings are another thing. Should you cite the work that essentially
has affected the work done, or, cite people?
Telling the same thing in so different manner. I'm easily tempted
to text-book style of writing (left), when the other option would
be to cite text books and write in a more general and easy-to-read
manner.
Well, the motivation for selecting the style of representation
in an article, that I hear about from colleagues here and there (my
opinions included, amidst), are
- Cite the potential reviewers (to please.
Show that you care about the research done by the community)
- Restrict
to an easy style, do not irritate the reader with too complex associations
- Maintain
to 'Tabloid-approach' - one theme per article - provide the holistic
approach and links in the Introduction & Discussion sections
Too complex papers (even if superb, classic) do not get citings
that eventually help you climb the academic ladders.
- Select
a 'reader' to whom you report, helping him to easily grasp the essential
- Provide
basis for a quick-view-style of reader with condensing figures, lists
and tables (Well, some journals have introduced a separate "Higlights
of the article", a few bullet points!)
- Restrict
to the essential, do not bother the reader with any unnecessary, but
provide all the necessary (This could be called the "keyboard contradiction"?)
- Discuss
within your work. Here, I notice that in some papers that hold little
contents, the authors are tempted to discuss even ongoing work that
in no way was reported in that paper. Well, you may do it, if
you can get away with it?
- Exercise
self-criticism. If the work has some experimentation, exlpain the confines
of it that might affect the generalization of the results.
Don't be afraid of the fact that you are helping the reviewers
to identify weaknesses. If they are up to their task - they should appreaciate
the sincerity.
- Aim at 'good results',
and if the results aren't that positive for the community (for example you
show that there is no potential in some approach), you have
two options: 1) make the results appear "promising" and highlight
the need for future studies, or, 2) report the results as they are and
be aware of the consecuences.
<>
Nov
13, Harvesting is over, is it?
Oat and wheat in Kulloo fields in November 2012. This is the
first time I ever saw germinating seeds in the ears of standing grain
plants.
The weather of summer 2012 only shows that the climate in Finland
does not give a 100% guarantee for grain crops.
Nov 12, Film scanning
project finalized
Photography is as easy as wearing out your pant legs - these
days. Cameras are inexpensive to manufacture, buy and use. A lot of
photos
are taken and shared quickly. Digital film is really revolutionary.
And the thousands of images occupy, in physical terms, a lot less storage
space compared to the analogue film. The industry still remembers
36 x 24 mm film size in particular, as it was the most popular in consumer
grade cameras. If you had good quality BW film, sensitivity at
ASA 50, the film could store up to 100 line pairs per mm. That sums
up to
3600 x 2400 pixels, roughly. In scanning, the 2D grey level function
stored on the film is sampled at some spatial resolution, and the grey
levels
are turned into discrete (integer) values. If 256 grey levels
suffice, 8 bits (00000000...11111111) represent the tone of each sample.
3600x2400 is
8.6 million samples, or pixels in a digital image. And 8 bits
constitutes one byte (eight-bit byte). If the film had colors, the resolution
suffered,
and not as many details are available in the film photo, maybe
50 line pairs per mm at best, which is 6.5 million byte samples of red,
green and blue.
What is RED, then? When the photography took place, the objects
in the scene were illuminated by a flash, lamps of some sort, or the
Sun.
The light sources determine the light available at the targets,
the lens, or what hits the film through the lens. With a flash, the spectrum
of light is highly
different from that of an evening "after-the-sunset diffuse blue
light conditions". The flash is often right next to the camera objective,
and the objects
in the scene are nearly shadow-free. Our perception of color
is of course subjective and we cannot define "what red is". We may have
our
own feeling about the "colors" at the moment of exposure. This
is captured by the film chemically, or by the photon detectors on the
focal plane.
Film development could ruin things for us, even if the exposure
settings were perfect. That's why those, who could afford, would take
multiple exposures
with varying settings on the film camera. Each exposure was something
like 1€, if you had good, expensive film. Now, luckily, the film development
is now longer needed, and we can view the image instantly on
the device's output. And the colors depend on the device as they always
do. Having
the devices (the screens) calibrated is a must to assure that
the colors are adjusted right (if we adjust them) and retained. We can
always fool the eye.
Storage of one's precious photographs has been, and remains to
be, the bottleneck. You have slow and quick noise that can destroy
the images.
Slow noise prevails with analogue film. The material simply changes,
even under perfect storage conditions. The colors wear out. Thus, digitization
is an option to stop the slow noise. In digital data, noise can
can be much higher before it finally prevents getting the 0's and 1's
right. But its different.
If you scratch an analogue film, it may well be possible to enhance
it back to what is was, visually. But, if you drop your hard-disk drive
on the floor,
or it gets an electromagnetig impulse (EMP) of some
sort, the 0s and 1s become impossible to recover. A scratch on an optical
disk has the same
consequences. So be it analogue film or digital storage on a
magnetic or optical device, ultimately, your images aren't entirely
safe.
Well, how long do we wish to have the photographs around? I guess
the answer is 3 or 4 generations. And to whom is a photo important?
Here's
a link to really
famous photos. Are these important to us, yes, for sure, but at some
entirely other level compared to the family albums. If you
have your family photo album at Google's server, will it be around
after 20 years?
Technically, it is enought to have the 0's and 1's stored, and
the rules available for converting the sequences of 0's and 1's back
into meaningful image content.
That's how you have the digital photo displayed. How about the
other information? In the past, the rule was to write instantly at
the backside the needed
information: date, event, people in the photo (the reason for
the exposure, back then) etc. Without these information, the photo
is almost useless but
to the very people in it. Now, if you did not pay attention to
this side right after storing the films or digital images, you are in
trouble. Indexing is needed
if the images number thousands. But who has time for that? The
same few people who have their family albums in good shape and up-to-date.
How to index? Sometimes a photograph that was initially taken
for an entirely different reason can serve a need for something else.
For example (a) might
do well when you wish to display life in Finland for
a foreigner. (b) was initially
just taken to document activities at the workplace I was at, but maybe
it
could be indexed with the word "journeyman - a senior and
a junior". (c) documents
teaching / learning activities at Hyytiälä that haven't changed
much over the
years, "Joy
of Sphagna" But the thing is that the associations that you have
when indexing need to be checked with others as well, because the photographer
has
a biased subjective view influenced by his own memories.
The scanning project is now finalized and I owe a big one to
my uncle Markku Korhonen (Vantaa/Ristiina), who scanned and tuned some
5000
photos.
Nov 10, All things must come down
Weather and the skills of Salon VPK were excellent on Fri. The
pine tree was planted 1984, and was now ig, 40 cm in stem diameter at
the breast height.
The open-grown pine and two other Picea omorika were planted
on arriving to Salo. The omorika spruce had a specific branching - starting
from the stem
the first 20-60 cm of the branch would descent right downwards
to became ascescent. This is required to have a slender conical crown
structure to survive
in the snowloads seen at Balkan mountains, where P. omorika
is domestic.
Nov 7 "Old" photos: How things and meaning of words
change
The scanning of old films has cntd (by MK) and I
came across this one, from August 1993. The text on the slide says "Kouluammunta
meni hyvin!". You have to know
Finnish to get the point here. "Kouluammunta" is what the precision
stage of the 25-metre center fire pistol is (or at least was, widely)
called for. The term
has had wide coverage later and the sport, in which Finns have
lots of Olympic medals, has become almost obsolete. In 1993, perforating
the paperboard targets was
still rather popular (and sufficient), but the new winds in shooting
sports were already blowing - people were more interested if the targets
would move, or at least
fall or get scattered. Since then we have seen the development
towards crazy phenomena such as the Paintball, where, if
the paint is red, it spills on someones
chest. Perhaps more realistic even than with the game consoles.
On the other hand, owing to laser, we have also the safe option of ecoshooting,
bullet-free, or
with NIR optical bullets. I have probably sowed tens of kgs of
lead during my active career in 1982-1994. But, the lead was always
collected from the dirt of the
bank behind the targets. That was feasible on rifle and pistol
tracks, but not at shotgun or practical shooting tracks, where the
lead is effectively dispersed. Precision
sports are nice (e.g. shooting, darts, snooker, golf), and train
one's concentration and control. But what if your darts would become
optical, or you would only
be able to use a golf simulator, would that be as cool?
Nov 4 "Your best friends..."
...follow (Tassu) you wherever you go and give comfort (Tiltu)
at all times. Halloween and Pumpkin Rotator 2000. One way of finding
use for microwave owens that have stopped heating the contents.
Caution (230 VAC): Do not try the same unless you know exactly
what you are up to!
Oct 28 - Nov 8 Getting
the footprints in the images (Project description)
The focal plane of the D300 is 24 x 16 mm, roughly. (a) shows the 1st echo image positions of
2-echo pulses in the pine stand (in the 'film' plane). The camera was pointed
up, and rotated about the Z-axis such
that the 24 mm "film" (CMOS) side was S-N-oriented, more or less.
There were 495 2-echo pulses that passed the camera closer than 30 cm
in XY. 36 from 2.9 km, 45 from 2.15 km, 78 from 1.15 km
and 336 from 0.67 km (a.s.l). The pulses in a strip come from
nearly the same direction, which explains the clumps (camera rotation
changed a little between exposures). Scan zenith angles are from 6 to
15,
while the FOV with the 18 mm lens was 64° × 42°. Thus the little
variation in rotation was not alarming. (b) The spatial pattern shows, where these
pulses are on the plot, constrained at the Z of each
camera that sees them. In the south, there was barren rock with
fewer trees. The 1st echoes were 2.03 - 17.54 m above the cameras.
Next
follows extraction of the binarized image patches that represent the
footprints of the pulses in the from-below-images. Now, if the waveform
suggests that the backscattering is over
a short depth, the depicted footprint area in the vertical image
has an approximately ok size, as the ellipse (circle) has infinately
many sizes, depending on the scale, which depends on the
distance from the camera.
(a) is a clear case,
in which the 1/e2 footprint (green dots) captures a sole branch 9.3
m above the ground. (b) shows
a pulse that hit something 12.1 m above ground and produced a backscattering
in the lower branches, over a depth. These kind of cases we will
not use in the beginning. (c)
shows a weak case, 7 m above the ground. The few shoots at the same
height have given raise to a low peak.
(d) has the returning
pulse width coarsely measured - suggesting 3.2 m and shallow depth of
reflection. The footprint with targets is 8.18 m above ground, and the
laser was shot 1158.4 m agl. The maximum
amplitude is 38. Thresholding is done with the estimated noise
level (blue line) + 2 DNs. The light blue dots indicate the end points.
This guarantees a more objective determination of the "ok pulses",
which are verified visually, not to have obstruction between
the camera and the target.
(a) The near circular
image area of a footprint. RGB = (0,0,0) denotes pixels that are outside
the 1/e2 footprint. (b) The thresholding
with a single value on the BLUE band gives this.
It is obvious that the threshold needs to adjusted for the background
brightness that varies across the 60 x 42 degrees FOV of the hemisphere,
with sun zenith at 75-80 degrees
during the photography. A gaussian weighting is needed for the
pixels doomed as silhouette to weight the footprint irradiance (W/m2).
(c) and (d) The correlation surfaces r(dX,dY)
of two 1 km strips in a grid of 20 x 20 cm, dX = -0.1..+0.1 m,
dY -0.1..+0.1 m, where correlation is between the Gaussian irradiance
-weighted silhouette area (%) in the footprint and
the maximum amplitude in some 50 + 50 pulses from two overlapping
strips. An offset of about 10 cm (same 1 km strips a here) is obvious in the other, while the other matches
the photogrammetric data
without a need to translate the cameras or the pulses, in XY.
About 45% of the amplitude variation is explained by the coarse estimate
of silhouette area (theresholding with BLU = 187).
This is really a fine
thing to see after all the trouble in getting the experiment geometrically
right (photogrammetry and LiDAR).
(a) This is the response
in one 500 m strip, now it is 9 x 9 grid, spanning +/- 20 cm at 5 cm
intervals. Peaks at dX ~0.15 m, dY ~ -0.15 m. In 500 m data, there are
many hits
near the terrestrial camera positions (low in Z), and it is necessary
to restrict the geometry such that the pulses are seen collinear: camera-echo-LiDAR.
In 2 km data,
there were just 10-15 pulses per strip available, but the offsets
are there tooas with 500 m and 1 km. 3 km will have too few 2-echo pulses
per strip available, because
the area is just 120 m2 and the per strip densities are
low, 0.5 p/m2. (b) Using all
50 + 50 2-echo pulses available, the two 1-km strips were searched in
40 x 40 cm offset grid,
and the peaks are inside it.
The 2 km data has four overlapping strips: dX,dY for these were
(+0.20, +0.15), (+0.05, +0.00), (+0.15, +0.05) and (-0.05, +0.00).
When applying the offset correction to the pulses in the first
strip, the footprints look like this with LiDAR intensities at values
of 10, 15, 25 and 40. The R2 in this strip is 0.75,
as if the silhouette area estimated from the ground images explains
75% of the waveform max amplitude variation. Perhaps the chosen thereshold
for sky (BLU band)
is most appropriate here. This can be examined by trying out
different theresholds and the optimal values per strip. In a strip,
all pulses come from the same direction
in the hemisphere that has a brightness trend. Before doing that,
the spruce stand will need some attention. It will take a while to remove
the lens errors with XYrectify,
that has no batch functionality in the economy version that we
have :)
Spruce stand
The Z at the tacheometer (gnd) was 188.20 m, based on 2006 &
2007 LiDAR hits. The 42 camera positions need to be adjusted for the
rotation of the camera frame,
about the Z axis, which was aligned with the optical axis. In
the iWitness EO solution that was obtained by matching the tacheometer
points of the 42 camera positions (3D offset, rotation),
the omega angles were [-177.1,-179.8] degrees, the phi angles
[-0.1, -2.4] degrees, and kappa angles [-11, -46.2] degrees. The camera
was first aligned to zenith, then
rotated such that the bottom (and the camera x-axis) was aligned
70..250 degrees (range of 35 degrees, apprx.) in azimuth. Thus, the
focal point was 9 cm off XY in
azimuth 160 degrees (aón average) and 18 mm above
the 1.55 m leveling arm. This gives dX = +0.031 m, dY = -0.085 m, dZ
= +0.018 m. Thus the total offset to the
tacheometer measured ground points (at the moonpod base) are
dX
= +0.031 m, dY = -0.085 m, dZ = 1.55+0.018 m. Because the offset applies
everywhere +/- similar (35 degree
variation in kappa), we can just offset the camera (X0,Y0,Z0)s
without the need to recompute the triangulation by iWitness.
The 1 km strips have ("silhouette-detected") offsets that match
those that were observed 100 m away at Silmäpäänlammi
using the shoreline.
The difference of the two strips was +12.5 cm in X and -12.5
cm in Y with the shoreline. It was +10 cm and -10 cm with the silhouettes,
yet the
responses are rather flat in the spruce stand, as it has so high
LAI. It becomes difficult to find images of branches with clear background
and no obstruction between
them and the camera. Lower R2s may also be explained by the varying
geometry of the branches (shoot orientation, branch type)
LK3
We have the 284++ Powershot images, vertical and oblique from
2011 from LK3 taken from places, where the pulse had intersected 1.5
m height above ground.
The same tree, south in the image up direction, imaged right
from below, four times, from the same camera location (last image is
a bit rotated about the plumb line).
Exposures at 08.45, 11.55, 18.20, and 22.45 DST, June 10, 2011.
Canon powershot has small 2.3 um pixels, and we used a light
JPEG compression. The images capture apprx. 55 x 55 cm in 80 x 80 pixels.
It is evident that the thresholding needs to adapt to
the illumination differences.
In LK3, both the 500 m and 1 km data contain a strip which 'saw'
the plot at the strip edge, which is where the oscillating mirror quickly
pauses and accelerates. The low number of
up-looking images (observations of slhouettes) has made it harder
to detect realiably the strip level offsets. (a) shows the pseudoechoes at Z = Z (tip
of one branch) They are pulses
that have their last echo lower than this Z and which have been
projected, as if they reflected from the given Z. The non-favorable point
pattern shows. It is due to the location in
the edge of the strip. (b)
The real echoes shows slight deviations, max. 20-30 cm. The question
now is whether the interior orientation of the near-edges LiDAR data
is sufficient.
LiDAR re-processing
status
AH has worked hard, recovering the lost echoes of the 2011 LiDAR
from raw SCN files. Well, in applying range (and various other) calibration
to these raw data, we discovered the original issue, a blunder
calibration value entered in ALS_PP for the 2nd return circuit
(MpiA calibration), a lost
minus sign. Well, we will see, if it is the right cure. Quite
a hastle for one minus sign if so.
We will be getting the LAS 1.2 point clouds re-matched to be
linked by ourselves with the waveforms in the Nov 10, 2011 LAS 1.3 files
that missed most 2nd - 4th returns.
The 2012 additional six 500-m flight lines were added to the
BINs. And these can now be analyzed also for the silhouette-footprint
association.
----------- Taking the camera with me in Oct ---------
(a) Teppo gave his "docent talk" - time flies - it was only in
1997 when we met first in Hyytiälä. (b) Teachers and pupils
at Kulloo school organized a "hobby evening" with children demoing
their own activities to other pupils and their parents. A very
good idea! (c) Kulloo boys playing in autumnly Kulloo forests.
Oct 27 "Legendary" OH5SM
in CQ WW SSB Contest - celebrating the many years of multi-op contesting
in Mustila
Light snow and -4°C in Mustila for the CQ WW SSB contest
2012 (QSL front side). Multi-op contesting has an over 50-yr tradition
there. I was there the first time in 1991. We were delighted to use
the
call of the early days - Carola's OH5SM. (a) View at the Mustila road on Saturday,
(b) The in 1984 erected 42-m
tower of OH5NQ (also OH5Z) with the original antennas by OH8QD, and
(c) OH2BX operating @
OH5SM in 1972.
QSL back side made for the occasion..
Oct 23, Verifying LiDAR 2012 xy-accuracy
Spruce stand
Near the spruce plot, Silmäpäänlammi was used
for strip matching. The shoreline was at 181.6 m and digitized from
the .175 m GSD images of 2012.
The offsets to the shoreline were (minimizing echoes inside the
"pond"), interpolated to the nearest 2.5 cm:
1 km strips
Strip 15 dX = -0.050 m, dY = -0.25 m (the left HSV-colored response
surface above)
Strip 16 dX = -0.175 m, dY = -0.125 m
2 km strips
Strip 8 dX = -0.150 m, dY = -0.10
m
Strip
9 dX = +0.05 m, dY = -0.10 m
Strip 10 dX = -0.10 m, dY = -0.05 m
Strip 11 dX = -0.10 m, dY = -0.175 m (the right HSV-colored response
surface above)
Because the average dX and dY were negative, the shoreline was
probably off too. The matching is thus relative.
It seems that the final tuning of the strip offsets will be based
on using the correlation between the footprint silhuette (image) and LiDAR amplitude.
We will let the pulses to shift a little in XY, then compute
the projection of the footprint in the below canopy images - binarize
(teaching phase precedes),
compute the silhouette (%) for the (dx,dy) case, weighting the
footprint with the Gaussian assumption of irradiance field within footprint.
These
silhouette estimates are then contrasted with the waveform traits
(max amplitude, integration of pulse, DR intensity), and we search for
correlation
maxima in the dx,dy space for each strip.
Pine stand
The 130-m tall, triangular SMEAR mast has some 90 mm thick Al-booms
attached to the NE side, 120 mm (center line) away from the mast's 50
mm thick "vertical corner tubing",
which have a spacing of 1200 mm (triangle, middle-middle). The
heights are 4.2, 8.4, 16.8, 23, 33.6, 50.4, 67.2, 101 and 125 m, AGL. The devices on the booms are
small, and the booms
look more to NW than SE (asymmetry). The mast is close
to our pine stand, seen by the same overlapping strips. 500 m data has
just one strip,
while there are 2, 5 and 6 strips in 1, 2 and 3 km strips. There
is a machine box up a top the tower (600 mm down from top) on the SW
side, 1250 x 600 mm in size. Smaller boxes
are located at different heights on the NE side (350 x 5/600
mmin size).(Tnx TP). The figs above show LiDAR points along a vertical
distance of 100 meters, starting from 25 m AGL.
The between-strip offsets are within 15-20 cm. In 3 km data,
one strip is a bit strange, but probably range accuracy suffers from
the type of complex object in question, in large footprints.
Trying to compute correlation of near-by point pairs between
two strips (A, B) on an open mire. The idea is to look for points <
15 cm apart. These should have correlation of intensities,
and the correlation should peak/change if B is translated slightly.
If the strip xy offsets remain tolerable, less than 15-20 cm, they may
be very difficult to be detected using large trees.
Then, instead, we can try computing the silhouette area in the
up-looking images for the footprint, and translate the pulses for maximal
correlation between the silhouette area of scatterers
and the waveform traits.
d_roll, d_phi and d_Z by strips were carried out for the Riegl
2011 Nov data. Now remains the recomputation of the per hectare BIN files
that combine the DR and WF data.
Oct 22, KKK 2012
There'll be a talk on Friday afternoon in Kumpula. The pdf-pptx-slides.
The FWF LAS data of 2011, that misses some 2nd-4th echoes, were
decided to be updated from the raw files using a RAW_2_Ascii decoder.
We will skip the
strip adjustment on these data as the campaign HRP were done
accurate in an urban site during the same flight.
Contd the local strip adjustement wit trees planning (anchor)
Oct 19, Finalizing the
spruce stand triangulation - some residual lens errors revealed.
Keeping
the leveling bubble at 1 mm / 10 m, one-handed
Out of the 11 x 13 camera positions, 42 had the XYZ measured
with a total station. A small plate was on the ground, on which the
monopod with the leveling bubble and the
camera were mounted during photography. Using iWitness for mere
relative orientation (tie points) of the image block and a 7-parameter
rigid 3D-transformation, which gave the
camera locations in the coordinate system used, the residuals
(42 x 3) had sdevs of 18, 11 and 22 mm for X, Y and Z, respectively.
The ranges of residuals were 86, 46 and 87 mm.
The figure on the right
shows that, in the solution, the images on the egdes were pushed up,
while the images in the centre were 2-3 cm down from the expected (total
station) height.
No control points (knowledge of the object space XYZ) were used
for these data, thus the +/- 2-cm-sdev trend seen in d_Z figure is an
indication of remnant lens erros. Very
modest though, because the mean distance to tie points was ~
10 meters. The d_Z deviations increase towards the block corners, and
this indicates that it was not only a camera
constant issue. The RMS at solution was 2.8 pixels, and the mean
sdevs of tie point coords were 4, 4, and 13 mm, for X, Y and Z. The
camera grid is just 10 x 12 m in size (see below, right).
Getting the RMS < 2.5 pixels is difficult owing to the movement
of the upper shoots (very mild wind), and the fact that the natural
targets and the 60 mm polystyre balls were
quite large points to target, on the image.
(Left) X = E-W axis,
Y = height axis. Side view of some tie points 5.5 - 8 m above the ground
(tacheometer was 1.55 m agl). These points comprise mainly of the polystyrene
balls attached to ropes, that had some sag. Getting the end points
at 6-8 m height with ladders was exciting. (Right) A levelling bubble was used for
levelling the camera, and
the mean of the elevation angle was 87.95 degrees with a sdev of 0.008 degrees, i.e. 140 microradians.
If this is true, we were good (stable) with the bubble :).
But we mounted the camera 2 degrees off the true vertical on
the monopod (the bubble was in the monopod, not on top of the lens).
The figure shows the negative and positive
deviations of the camera elevation angle, from the mean, 87.95
degr. Now, it would be better if these deviations were entirely random
and would not show the obvious trend.
N.B. the camera back was always pointed in 160 degrees azimuth
+/- a few degrees (i.e. the camera azimuth was pretty well fixed).
The trend is perhaps +/- 100 microradians,
and it is likely due to lens errors as well. Well,100 microradians
over a distance of 10 meters is 1.0 mm, so it is not so bad, but, yes
an indication of remnant lens errors.
All this indicates that the
ray block is accurate, in relative terms. The XYZ orientation of the block
was obtained by the total station, which was XY-oriented with the help
of 32 trees that
their stem coords measured in aerial images (assuming no tree
slant). The tacheometer was standing on flat open rock and the Z was
obtained from 1 km LiDAR of 2006 and 2007, which
have been calibrated with GNSS to be within +/- 5 cm in Z. With
the XY sdev of 0.2 m for the trunks, we can expect 0.2/sqrt(32) accuracy
for the image block X and Y, which is 3.5 cm.
Chromatic aberration - there again
http://nikonblog.cz/wp-content/uploads/nikkor_18-300mm_produkt_1_small.jpg,
http://nikonblog.cz/wp-content/uploads/Nikkor_18-300mm_CA_small.png
Peter Lindner's pictures from his blog show the 18-300 m lens
(adjusted at 18 mm as we had in the forest) with a D5100 (we used a D300),
and he talks about chromatic aberration. It was clear in our
images too, suggesting that it might be better to acquire a proper
12 mm or 18 mm fixed camera constant lens for the future activities.
CA makes it even more harder to
accurately point tie targets, and explains in part why thin objects
against bright bkgnd show funny constrast (colors), and are hard to discern
(distant needles and shoots of our trees).
The resolution of our D300 is limited not only by the lens MTF,
but also by the RGB pixel patterns on the CCD.
FWF status
LiDAR 2012 campaign HRP-calibration was re-done, and pitch was
affected according to our expectations. Strip adjustment with d_roll
and d_Z were further done, and we got a
first set of DR LAS 1.2 data, which were then sorted by time
and linked with the previous fwf's in the affected LAS 1.3s (by AH).
LiDAR 2011b data set's 2nd and 3rd echoes still remain in the
raw observation files. LiDAR 2011a will be strip-adjusted.
Trying local LiDAR strip adjustment with tree
apexes (work planning)
It seems now, that since direct sensor orientation is limited
to an accuracy (LiDAR pulse path) about 5-10 cm / 1 km of flying height,
and strip matching cannot remove the offsets
entirely (in a forest scene with very little man-made well-defined
surfaces of varying orientation), we need to, at least to verify, to
be within the 10 cm limits with all overlapping strips.
I.e. we need to move each overlapping strip a little for a perfect
XYZ match, locally.
The data we have are from 500, 1000, 2000 and 2750 m, and the
1/e divergence of the beams is 0.15 mrad, which means that the 1/e footprints
are 7.5, 15, 30 and 41 cm wide.
The strip overlaps (image) were
20, 50, 80 and 85% to achieve a minimum, combined, pulse density of
2 per m2. This means that the spacing between pulses can be quite large
in the 1, 2 and 2.75 km data. Down to 0.4 p / m2 - 1.5 meters
between samples.
Monuments on the ground are needed here, that are viewed by the
different strips. If these are 2D and planar, they should posses some
features that give raise to strong/weak backscattering.
In cities, road markings are such - dark asphalt is dark in NIR,
and the paintings show high signals. If one is able to map the polygons
(e.g. rectangles) of the bright patches,
it is possible to apply offsets to the LiDAR data that maximize
the mean intensity inside an area cut by the rectangle (we search for
a match). Consider zebra
crossings. In Hyytiälä
there unfortunately aren't that many road paintings - what is
available are some fresh asphalt patches, or water pools on the asphalt
(sometimes, if it was rainy, water absorbs!). The borders of
gravel (low refl.) - vegetation (high NIR refl.) aren't clear
by the forest roads, but exist to some degree, yes. There are a few small
lakes that have white moss (Spaghnum) shores,
(see terrestrialization, USDA),
rather complex and the water often has absorbed the pulse entirely,
while white moss, even wet, is seen. Thus, the shoreline becomes traceable
in the LiDAR data.
In some earlier papers of mine, we've used both these shorelines
(pdf)
and understory lichen mats (abstract)
for analyses of LiDAR XY-precision, but never at the strip level.
Then there are potential 3D signals. We have used roof structures
(eaves, ducts, ridge/crests), and small trees once (RSE article, pdf abstract),
again at the campaign level,
without assessing the strips individually for strip-level errors,
that actually should dominate, unless flying is always in the same direction,
which is when any constant roll/pitch/heading
errors, bias, would affect always in the same direction.
If we have the 2D polyline of the pond's shoreline - and a buffer
of +/- 1 m (streching noth ways into the water and mosses) and it is
possible to move it in XY such that the number of echoes from the
water (and their mean intensity) are minimized - this is done
separately for each strip, for example in an XY grid of 50 x 50 cm. First
sampling at 5 cm spacing and then densified, where
we seem to have the hot spot to be able estimate more accurately
(or you interpolate/fit the sparse grid with a parametric surface).
See the fig. below.
An example from the 2008 RSE paper, in which Lichen (Cladina)
mats were moved in a 80 x 80 cm grid and the t-test showing
classification accuracy using intensity distributions (lichen
mats vs. the rest of the forest floor flora), showed a "peak" at an offset
of about 15 cm. The offsets removed here were many (explaining the
flatness of the response surface). The field observations possibly
also had a small offset, as we used photogrammetry to establish the
lichen mat XY maps. So this analysis and the one with
the understory trees (pdf) removed
many offsets for a local match. With the understory we had two LiDAR
data sets to be matched with the field, and slightly different offsets
were
estimated for these, suggesting that some chances might exist
even for strip-level estimation. Of course, if the response surfaces
(fig. above) remain lame in their peakiness,
the offset estimates are not precise.
Another option would be to look for the autocorrelation of
intensity surfaces, that are interpolated from different strips. Consider
1st echo data on the ground. Fix the positions from
strip A, and interpolate a grid of intensities. Then, let the
points of strip B have small offsets and compute intensity surfaces IB(dx,dy).
Then compute the correlation of these (IA vs. all IB) and
interpolate the maximum of the resulting correlation surface
at sub-pixel accuracy. In this way, it is not only the shoreline (no
data - data border) that is being used. And cross-
correlation is rather invariant to slight offsets (between strips)
in intensity. The only open areas that are suitable for this are clear
cuts (low vegetation), open mires and roads.
Man-made 3D monuments would be nice to have - roof crests and
edges, roof planes, all oriented in differently, and clear from vegetation
(trees next to buildings). But these we don't
have for the strips that view our trees. The pine stand is close
to the SMEAR station, but for some reason the roof of the building absorbs
everything, its a black thing. There is now the
132 m tall SMEAR tower, triangular in cross-section, with many
layers of guy wires and some instrument booms at different heights.
We hope to be able to use it, although the guy wires
aren't visible but in the 500 m data (too thin in the higher
scans).
Now, the 3D monuments available in the forest for doing
the strip adjustment aren't many, or actually they are. Namely, we intend
to try this with trees, tree apexes. In Hyytiälä, 40-60-yr-old
pine and spruce are many. These Pinaceae trees normally have very
straight stems (butt-top sdev in XY < 20 cm) and rather symmetric
crowns. The heights range from 12 to 25 meters and
the spacing of trees is from 2 to 6 meters, usually. More or
less random spatial point patterns are seen unless planting in rows
took place once. If we have the treetop XY positions
measured for hundreds of trees in a small area that is seen by
overlapping trees, and the anticipated strip-level XY offsets are small,
within +/- 50 cm, it is possible to look for LiDAR
points inside cylinders that are centered at the treetops, maybe
2 m in width. We are in the pull-in-range.
Then, we can look for the highest LiDAR points. They are end
points of vectors that stretch from 500-2750 m agl down to the tree
crowns, 12-25 m agl. The scan zenith angles are very similar within
one strip, but may vary 0-15 degrees between overlapping strips.
Some LiDAR strip viewed the trees from the left, some above, and some
"from the right" - loosely speaking. Thus, the deviations from the tree
top are XYZ vectors (Treetop -- LiDAR point), conditioned by
the direction (i,j,k) of the pulse. If tree crowns were solid opaque
surfaces that do not let light thru, we'd be better off. The pulses
would
stop at the symmetric surface, but this isn't the case (See fig.
below).
Crown volumes comprise mostly of gaps, and it is loose talk to
speak of crown envelopes. LiDAR pulses reflect from volume, where
the they intersect a surface ("stuff") large (dense) enough and
properly oriented. However, for slanted LiDAR pulses, the point patterns
of a rotationally symmetric crown will not be symmetric anymore
owing to occlusions along the pulse path. We have the "LiDAR
front-lit and back-lit sides of crowns". This we need to consider not
to introduce any artifacts in the XY offsets.
Now, if we can measure 500 treetops with a precision of 20 cm
in XY, and 50 cm in Z, using photogrammetry, and the XY bias remains
below 10 cm, we should be able to adjust the strips down to an
accuracy of the bias, which of course will remain. Now, if the crowns
are not symmetric (same amount of
needles in any given direction), that bias will remain too. There
seems to be very little research into the symmetry of upper 40-60-yr-old
pine/spruce crowns. Some asymmetry
is reported in Rautiainen et al. 2008 (pdf), for entire
crowns (maximal extensions).
Work flow - 1) map the tree tops in aerial images
that have the Sun in SE and SW. 2) Compare the point patterns for possible
bias (mismatch). 3) Extract LiDAR data in the vicinity
of the treetops, store the echo with max Z, and store the offset
(dX, dY, dZ) to the top. 4) Let the LiDAR points move about in a grid
and compute the average dZ for each set of 300-500
trees. 5) Do this for all strips and analyze the surfaces for
the maxima, i.e. the XYoffsets. 6) If they all show similar mismatch
despite the flying direction, the errors are likely in the
treetop XY-positions (bias), but if the pattern is associated
with the strip flying direction, we have more likely found remaining
HRP offsets.
Oct 16 Luxemboug
We decided to leave behind a larger ecological
footprint and travelled all the way to Frankfurt by air and from there
another 400 km to Luxembourg by car to see what the other
country looks like that is seeking UN's security council seat
this coming week. Renting a car was 90 EUR for 54 hours, so at
least that was pro-larger-eco-footprint pricing.
With KÖ, we ended up in a place called Quatre-Vents, or
Ershdorf , a kind of a village very high in altitude, in terms of LX,
500 m a.s.l.. We were at a farm that had lots of
self-standing masts filled with ham antennas. The land parcels
in the village were very narrow (land re-division has not taken place
there as it has in Finland) and therefore
the self-standing towers - simply no room for the guy wires.
Autumn was there too, just some corn left unharvested, probably a late
sowing for forage. Oh, and we saw the three tall towers
of Radio Luxemburg (you remember? 1440 kHz) on the way. That
brought some nice memories from early 1980s, me and my Geloso, Nena
and the 99 balloons. The missing
100th balloon was being hoisted in the air on Sunday evening
as we flew back. People watched on the TV at Frankfurt, as the crazy
austrian wnt up in his to the stratosphere, while Austrians took off
from
the strips. Autumn weather in LX, wet that is, and sadly too
little light for proper photography. I saw a single pedestrian, so probably
Luxemburgish people stay indoors the whole of
October? The language situation there is interesting. There are
namely three, and we tried all but the third :) LX7I broke all
existing records in the contest. That 2000 km
makes a big difference - its as if DXing (HF radio) in Finland
compares to fishing from the shore while others take themselves out
there in the boats. And then there is the
operator, and LX7I was manned by the best available at the moment.
We made a contest trophy for ourselves beforehand - reflects
our trust in ourselves or the contest organizers? The local bier greets
in French. Land use in LX is the reciprocal of that in Finland.
Swap the percentages of cultivated land and forests. View downhill
to Finland from Ershdorf (Google maps).
AH examined the raw observation files of the DR campaigns and
the lost echoes (range data) are there. The intensity anomaly was also
traced to the raw files of 2012, and
it is there too, 2-5% of pulses have contradicting I2/I3 intensities
compared to the waveform data. This is then something to live with
and consider when using the DR data.
Now remains the task to get the lost echoes of 2011 to LAS files.
We yet don't know what causes them to vapour in the midst.
Trangulation of the spruce
stand phtos is progressing at a pace of 4-5 images per hour. Some tree
movement during the photography as the upper canopy tie points show higher
rms.
The ropes with polystyrene balls, with a spacing of 1-1.5 m,
6-7 m above the cameras (grey-blue objects) show some sag in the solution,
which is good.
Vice
members were sought and found last week for admin bodies. TBD.
UH is considering making one of our docents a person-non-grata.
It's funny. The only comment from 2007 at UEF, when I was granted my
docentship, that I well remember, was that of the Dean,
"You are now entitled to comment things on behalf of our University,
as a docent. Be aware of the responsibility".
Wrote a little story to the yearly publication of the UH forestry
students. With a pseudo-pseudonym this time. I gues the last contribution
was from late 1990s, so it was time for another.
Oct 10, Update on the LiDAR status: "intensity - amplitude anomaly"
and "Raiders of the lost echoes"
The 2010 LiDAR (sensor #1) was checked and found
to deviate from the 2011 and 2012 data sets (sensor #2). Thus, what we
observed (2011-12 intensity-waveform anomaly) is NOT sensor-model -specific.
That is now ruled out and excavation of the potential sources
of this anomaly will continue. We will also do enhanced strip adjustment
to the 2011-12 data, and separate output of DR
data and waveform data - combined with a post-processsing step
of merging the two with the time stamps available. This allows for a
beter definition of the 3D path of the pulse.
Strip adjustment will bring the pulses to a better match with
the field geometry and over time - with other data sets. We are back
on the tracks!!
Mom turned 75!
Oct 5, 2012 Learning about the LiDAR measurements
Now, while looking at the waveforms (pulse paths)
in the below canopy photographs, we realized that a complete examination
of the LiDAR data integrity is needed and this we did Wed-Fri,
for the 2011-2012 data sets. AH did computational detection of
outliers by applying known dependencies between observations. Some
interesting findings, yet provisional, about the 2nd
and 3rd return intensity data in the ALS sensor. These
findings, if true, may partly explain as to why our earlier attempts
to normalize I2 data for transmission losses have been less
succesfull and anyone trying it out with wf-data will have more
success!
The figures below illustrate some of the phenomena we encountered.
Left: A waveform
of a pulse that triggered three discrete-return echoes (green dots depict
the time difference to the echo from the beginning of the waveform
sampling, which
is 256 samples at 1 nanoseconds). We predict the amplitude, blue
dots, using the recorded intensity of the DR echo (see #s listed for
echoes 1-4). The middle and right pulses
have triggered four echoes. The green and blue dots seem to match
less accurately for the second and third echo (I2..I3) data. And, it
seems that the sensor really prefers
the last echo (middle). When there are 5-6 potential echoes
in the wf, the last one is seen captured much more often than being
dropped out. This is logical, the sensor is
primarily intended for topographic applications Oddly, the intensity
can be zero for I2 and I3 (right). And the estimated amplitude
(blue) is at value 28 for such an echo.
The amplitude data are recorded by one byte each, and the decimal
values for noise are at values of 11-13, and the recording of the waveform
starts when a discrete-return echo
is captured by the system - seemingly a buffer stores some amplitude
samples and the final recording always covers 22-23 nanoseconds that
preceded the echo. Sometimes
you see above-the-noise-values in that 22-ns segment too - and
there can be echoes omitted by the DR system, only visible in the waveform
data (left, middle).
A certain percentage of two-echo pulses show inconsistent second-echo
intensity (I2) vs. amplitude value dependency (left). The middle
waveform amplitude x (I1 or I2)
is as it should be, as well as the in the 3-echo waveform on
the right.
The
I1 x amplitude values show strong linear dependency (left),
which is partly broken in the I2 x amplitude joint distribution
(right). The I2 == 0 show on the y-axis.
As to what happens in the recording of I2 and I3 in a certain
proportion of pulses, is still unknown, and we must go for the raw data
before it was post-processed,
to see, what really goes on.
The Q now is, if the waveform
data are actually a better measurement of the backscatter strength
(max) than the discrete return intensity data.
In terms of the first echo (intersection), there seems to be hardly any difference
(issues), but for the subsequent backscattering it may well be advisable
to trust the waveform.
Until now we haven't really had any redundant control over I2
and I3 data, apart from some vegetation reference targets that are unstable.
Yet, some of the
noise in I2 and I3, we may have earlier attributed to a wrong
source, by saying the noise is attributed to vegetation variation only.
Now, the redundancy is
there in the form of the waveform. It is kinda interesting..
October 2 desperate.iwp
"Mannikko_aamu_143_8_old1_2_2_1_1_6_desperate
(5).iwp" - this is the name of the current observation file,
reflecting the stability of the software (and the user input, :))
used for collecting the image observation data.
Measuring
trunk tie points and applying the tacheometer XY-data for these (corrected
for trunk diameter), resulted in 4 cm sdev for the now 28 GCPs. Changing
the problem images
for
those from another image set made no difference, suggesting that it
is not a interior orientation issue (vibration).
Well,
trying out how the pulses look like, when viewed from the tail hitting
the ground near the camera that looks up. The number is the amplitude
at the distance of the
echo.
And the circle (rotated in 3D) is the cross section of a 40 cm footprint
(1/e^3). There seems to be some something in there.
September 28 Slicing centimeters, still
iWitness seems to crash when trying to
input the 72nd image observation for a tie point. Hair loss still from
the image orientation at the pine stand. The Q/C on the delivery
of the airborne
data
got underway, and we noticed, when monitoring the XY-accuracy, that the
applied boresight calibration might have been 300-400 umrad off. The
photo has data from three strips (colors).
Black
is 500 m (flown S->N), while the other two are 1 km data (S->N,
N->S). The Y-offset is 20 cm at 500 m and 40 cm from 1 km (1 m from
2.75 km). The left view direction is almost from
West
to East. We need to adjust the dPicth correction in these LiDAR data.
AH produced
these maps with FWF-sample density per m2 (0-6) for the four flying
heights of 500, 1000, 2000 and 2750 m. The DR data is more dense in 500
m
acquisitions
owing to reduced FWF sampling (every second pair quantized).
Vertical
canopy photos
There
are 152 images, of which 48 have cam coords observed with a tacheometer.
When letting these to have stdev of 2.5 cm in X, Y and Z, the solutions
show deviations with maximums at 8 cm (posteriori sdevs 2.3 cm).
The
bubble charts show both positive deviations in blue and negative in
black. The image RMS is stuck at 18 um (3.2 pixels), and a few images
have clearly higher RMS (average 6-8 pixels). The seven GCPs
in the
canopy were estimated as tie points (sdevs 100 m), and the 21 xyz coords
were from -9 to +7 cm off, which is good. The theodolite accuracy is the
limiting factor here.
Now,
I need to replace those problematic views with images from another image
set taken was in late evening (morning photos) the same day to see if
it was some mechanical issue (vibration) with the lens.
Also,
will try to use the trunks as 20 cm XY-GCPs to provide more control.
The are some 30 trees seen in the images and the trunk XYs are known at
20 cm sdev. It may well be that this accuracy is not
sufficient.
If not, we need to return to the site to measure some intertree distances
in order to increase the redundancy.
If the
RMS is 3 pixels, and one pixel is 5.5 um, c = 18 mm, the image observations
will be off 1 cm / 10 m of distance, which is tolerable, but the fact
that 4 images have locally high RMS worries me.
I had
not expected this work to be this tedious!
September 21, Trying hard, harder
Data
delivery
We got
the airborne data delivered for our LiDAR & photogrammetry
experiments. 300 GBytes of which the imagery is 200. Just copying it
around is 3 hours - minimum.
The
November 2011 Riegl data now seems to be in order with all waveform
packets nicely addressed - apart from those at the very tails of strips
(few tens of meters),
which
is fanny, because the discrete-return data is there (based on waveforms).
Waveform data aren't really used but in research until now. What awaits
is the Q/C.
E.g.
direct sensor orientation was applied with the images and that needs
always a bit of control (the 2008-2009 built signals are still seen mostly).
TIFF to RAW conversion
is needed
so that the images work with KUVAMITT, as is the conversion of MATCH-AT
style exterior orientation (EO) to KUVAMITT.
Triangulation
activities
Main
activity of the week was the orientation of the 143 ground (vertical)
images of the 60-yr-old pine stand with iWitness (some 5000 image
points). Iwitness kept crashing
and
the diagnostics for blunders in this (economic, non-pro) version that
we have - isn't entirely satisfactory (if program crash implies arithmetic
issues, caused blunders). Or then it is just
the
OS (win7) misbehaving. The images have a spacing of 1 m, and the FOV
(27 mm equivalent lens f) is roughly 40 x 60 degrees so that at 6-7
m above, the area seen is
4 x
6 m, and at the top of the canopy, at the height of 15-20 m, it is 10
x 15 m. For triangulation, one needs to find conjugate points, and the
lower they are, the fewer images
they
are seen in, and the larger are the angles. Since the camera position
was known in a 6 x 8 points (2 m spacing) from tacheometry, it would have
been better to avoid
very
high tie points, because they lead to weak intersection, and many observations.
However, in canopy openings, they were the only option. Some images
were taken under
understory
trees, and they were particularly challenging owing to the occlusions.
Some odd increase in the pixel RMS in a few images - as if the lens parameters
were not
stable
for these (after remeasuring everything many times the issues remain).
When the triangulation was done, it was possible to compare the obtained
XYZ coords for
seven
control points (150 mm polystyrene balls). "Tension in X-direction"
was 5 - 7 cm. This could indicate that lens effects weren't entirely
removed. The total RMS was 3.3 pixels.
When
the residuals were observed over the focal plane, no large systematics
were seen, however. The targets (twigs, branches, buds, etc.) in the
canopy aren't very precise
(focus
blur, large targets, perpective distortions, f8, 1/100-1/160s) - so some
2-3 pixel RMS is to be expected. A pixel is 5 um in Nikon, and the lens
calibrations RMS itself was 5 um (bottom noise).
View
from the above. The grey-blue squares denote the 143 up-looking cameras
in a 10 x 12 m grid, 48 of these were fixed (2.5 cm sigma) to known (tacheometer)
XYZ.
The
green dots with numbers are the corresponding points in the canopy.
Seven of these are illustrated with the image rays - these are the GCPs. The placement of the GCPs
could
have been better (now just the eastern part is covered), but it was
hard to get the ropes on tree crowns to install anything (stable) up
there. What
remains still, are
some
oblique images that were taken from the sides, adiing them to the ray-block
will hopefully reveal how the geometry really is. I don't like
the 5-7 cm
offset
in X (photogrammetry vs. theodolite positioning). Slicing the cm's is
tedious, but we have to have everything right. The 48 camera XYZ positions
and the seven GCPs,
the
RMS for these is 0.033 m (after rigid 6-parameter 3D transformation),
so in all, it may be hard to improve.
The
same 143 image triangulation awaits in the 60-yr-old spruce stand. It
will be even harder, because of the poor illumination and contrast. The
network of polystyrene balls
(1.2
x 2 m, 6-7 m from the ground) hopefully provides a dense enough and geometrically
appropriate tie point set. The fact that the balls are attached to the
lines means that
they
all sit on a '3D line', which isn't entirely optimal. Once the epipolar
geometry is established between images (needs a minimum of 5 and later
4 conjugate points), natural
points
can hopefully be used as a complement. The area covered in these two
plots is 2 x 10 x 12m, and the per-pulse costs of observation are becoming
very expensive.
Backups
Near
future activites also include the back-up of all LiDAR and imagery
from Hyytiälä (1946-2012) on some 4-5 TByte media. It's
becoming more and more problematic to maintain
the
Hyytiälä data. It would be great if it all were somehow nicely
indexed (spatially, temporally) and harmonized, such that any retrieval
would be easy. Airborne frame images
are
rather easy to handle - the EO tells what is covered by the image frame
and the HDRs point to all image versions (levels of processing). The same
basically applies
to ground
photographs. A database can handle these. The line-sensor data are
more complex, because of the excessive amount of metadata (EO), which
aren't harmonized.
The
field data are complex. The airborne LiDAR is mostly converted to our
own format, yet a few versions of it exist already, that adjust to changes
in the observables.
Well,
the first step will be just to list everything and store it in one
place (media).
Ham
Radio changed with digital technology
The
3-el steppIR went up (tnx OH3BU
& OH6KZP), and OH1WZ is on the
air again. The control box of the steppIR was damaged though. Watch for
the wirings!
Any
short circuit and you need to replace a whole board of electronics.
The solar cycle is peaking now, although it is very lame activity. Yet,
for a radio amateur
it is
fun times. The annual Scandinavian activity radio event took place
too. Was active for a few hours giving morse code with a straight key
and an iambic paddle.
Morse
code is fun too and learned it quite well as a youngster. It is
a digital mode with dits and dahs (0/1) that are combined into sequences,
synchronized (speed).
Well,
from this it is no surprise that morse code can also be coded and decoded
by machines. Coding by machines used by radio amateurs started already
a while
back.
The introduction of PCs in the radio shacks was the turning point.
Coding was easy in DOS machines, which could be set the LPT pins to 0/1
by software, +5V
1's
were then used for keying the radio (just a transistor needed). You
no longer needed to use the straight key or an electronic keyer with
a paddle. Decoding of the Morse signal calls for a systen that
turns
the audio signal into a digital signal and an algorithm that finds
the speed (synchronization of digital data), and recognizes the dits,dahs
and spaces. It is not
very
complex, but noise (overlapping signals, low signals) and synchronization
issues (hand-made vs.machine coding) constitute challenges. Well, talking
about digital
systems
(signal processing), the speed of them is so high nowadays that it
is possible sample simoultaneously a wide spectrum of the radio frequencies.
Consider
a radio
receiver that feeds the audio to the Morse Code decoder at radio frequencies
sampled between 200 Hz, e.g. over a range of 100 kHz of radio spectrum
(e.g. 7000-7100 kHz).
That
is 500 'channels' being read simoultaneously. If the audio of each channel
is sampled at 16 bit, 5 kHz, that is 10 kiloBytes of data per second
and channel. For 500
channels,
it is a data flow of 5 MBytes, which is easily moved around in a USB
bus. Now, if the computer being fed this information has many processors
that are fast
enough,
these 500 channels are decoded for possible Morse Code sequencies.
The computer may even be connected to the interned and feed these information
to a server.
Now,
if such receiving stations are all around the world, there is no way of
getting on the air unnoticed, using unciphered Morse code. This is implemented
by radio amateurs
who
use a software called Skimmer (decoding, feeding the server) and digital
receivers (software defined radios). The server is called reverse
beacon network, RBN.
It is
fun to use, if you wish to learn where you are heard, but it has also
killed all my interest in radio contesting, because we no longer need
to use hearing aid and
human
decoding of Morse code, to get a contact initiated (for finding people).
An aid kills a skill.
Autumn
Autumn
arriving in Kulloo. Apple trees are very productive. No pumpkins or sun
flowers this year in our garden, which reflects the weather. Wheat fields
still not harvested in Kulloo, and
with
the rains, its increasingly challenging. Finland is 60N and up, and it
is easily marginal with cereals if the weather isn't average. Well, lots
of frogs around in the garden!
September
10, Trying to get the show going
Sun-Tue
we visited SLU at Umeå in Sweden and met with colleagues there
to discuss ongoing and future activities. We went over from Vaasa to
Umeå, which was
the
full experience. Clear skies at 01 am Monday morning and Aurora borealis
right above our heads welcomed us to Umeå harbour. Umeå looked
as a nice mid-sized
town
with nice surroundings, although we had no time on our hands to learn
to know it any better. It is 415 km from Vaasa Helsinki, and during the
summer it is 5 hours
by car.
Thus the whole trip to Umeå takes one about 10 hours considering
the time spent loading and unloading the ferry.
The ferry, RG I, is operated by
a "konkursbo", that we heard is nowadays been subsidiced by governments
in both Finland and Sweden. The upper deck has an elevator, and
the
time it takes to unload your car can be substantial. We drove by Koskenkorva
to see Koskenkorva in Umeå too. That's Finnish. Well, we learned
that almost anyplace that we went
to,
there was someone who'd speak Finnish. I guess the same applies anywhere
in Sweden. Half a million Finns out there.
At SLU,
there were posters on the wall and these amusing wallpapers. Aarne
explaining to doctors Olofsson and Holmgren. At SLU, a renovation was
ongoing and the RS people
were
occupying a multistory office building made out of containers. We also
learned that annually, 85 students start forestry at SLU. The students
had their own building.
A whole
day of admin-meetings on Thursday. Let me phrase it like this - admin
is nicer when the resources are on their way up. I could do nicely without
the extra admin now.
Namely,
there seems to be enough admin with the projects, given the continuous
change of practices and limited (actually you get an overwhelming amount
of disinformation
through
bulk message-sending that uses the same channels as the official important
matters) information transfer. A sudden worry over the finances was brought
alive by some
dark
admin clouds. Must check. UH is currently under hard economic times (according
to Helsingin Sanomat), it is difficult to hire anyone, even as a replacement.
And the effects are
random
- maternal leave, grant, retirement - any reason to free the resources
is grabbed. No one has been sacked, yet. It is a huge (expensive) and
unpleasant process and thus
sensible
to evade.
The
blocktriangulation of the Hyytiälä 2012 image blocks, as
well as the (re-) post-processing of the Nov 2011 and July 2012 LiDAR
data are soon finalizing. A meeting
around
these on Friday at Malmi.
When
considering the energy distribution inside a LiDAR footprint (simulation),
it is obvious that it is not even. Gaussian spread of irradiance - we
have learned. Now, the manufacturers
and
researchers often describe the divergence of the pulse through an angle.
And they define an 1/e divergence or footprint. If irradiance drops
to 1/e or 1/e2, then what is
the
energy inside that 'circle'? I have always assumed that it is (proportionally)
1-1/e or 1-1/e2. But I cannot remember how I got these values. The graphs
display normalized
Gaussians
with sigma at 0.3 (red) and 0.5 (green). The integrals are on the right,
and two horizontal lines are 1-1/e and 1-1/e2. Both curves intersect
these a x = sigma*sqrt(2) and x = sigma*2.
Gaussian
functions have interesting features it seems.
Sat--Mon
prepared a lecture in Swedish for Bosse's course at UH. Spent time looking
for the right terms. At SLU, we learned that English terms seem to dominate
the
Swedish
language, as what comes to modern remote sensing. Powerpoint slides(pdf),
terms in Swedish (pdf).
The
12+4 m Al tower went up on Sunday - the wooden mast lasted operation.
Now comes the tricky part of climbing up with the beam on the shoulder.
Autumn
activates the radio genes in one. It is the peak of the solar cycle,
and thus time to get the antennas back in the air for the winter.
We are
trying a wooden mast to assist in hauling up the cranked tower. It'll
be exciting. Arttu was in his first ever orienteering event and that
was fun too.
tbd: get the coords to hyde-ref
Hyytiälä - In search for interesting
research sites
We had planned for a trip to look
for specific 2-layered stands, or plots to look at the joint distributions
of intensity (i1 x i2) data in pulses with at least two returns.
Wed-Sat
we searched the 700 ha study area for pine, or birch dominant sites,
with homogenous understory tree layer or bottom layer. The idea was that
the plots
would
be at least 0.02 ha in area. A Trimble GeoXm (2008 series) was there
to help (It was validated at forest GCPs).
Soon
we realized that it would be nice to have a PC with one, to produce,
on-the-fly LiDAR feature maps, to look at, in the field. This constitues
a TBD task for the
future.
It was
rare to find a non-mixed understory layer, or a homogenous bottom layer.
Of course, its subjective to determine the homogenuity of the bottom layer,
but it
wasn't
easy. Well, some sites were found at least, and we learned a lot about
the study area, as we this time did not use the roads and paths available.
We even
found
a contorta pine stand, 0.5 ha, that did not seem to be well known.
We also
learned that the shallow waters of lake Kuivajärvi are truly good
pike waters! Tue, we checked the status of B. multifidum in Siikakangas
and found them
at the
old sites. These two, and the two other found earlier in Hirvensalmi,
share one aspect - they all are at sites once subject to humus (2 x saw
mills, tree hauling area; tree bark,
and
the Siikakangas opening had been a storage of peat in the 60s and 1970s,
probably link between the construction of road 66 in the late 1960s).
Could B. multifidum
favor
such sites?
B. multifidum
at Isosilmän kämppä (where wood piles used to be loaded).
Vuorijärvi had some Warnstorfia type of moss growing at the East
end. An aerial image signal from 2001 greets one by the Forstie road.
Spruce
understory at the Jyrki-Pekko stand. Downy birch was often dense by the
distches at barren drained pine bogs. Calla at Vuorijärvi terrestrialized
pond (narrow Segde-fen that is scary to walk on).
The
old and current generation. Scheutzeria was ripen at Vuorijärvi.
Ledum palustre (R. tomentosum) "ruska" at Ketunsuo.
Sensors airborne in Cloudy Hyytiälä
in July
Despite
extremely poor weather in the summer of 2012 (photogrammetry weather),
the two airborne campaigns were carried out in Hyytiälä by
FM-International (Finnmap).
July 5-6 (night-time) and in the evening
of July 6 (patching up) the LiDAR
campaign from four altitudes with constant at-target
footprint irradiance (W/m2) were done, and the
long-duration
imaging campaign (repeated blocks) with constant f-number & exposure
settings was carried out July 25.
The sensor was a regular "PAN-sharpening contraption",
i.e.
a common large-format, topographic RGBN+PAN frame sensor. The LiDAR campaign
extended the time series of Hyytiälä LiDAR to include '04,
'06, '07, '08, '10, '11a, '11b, '12,
i.e.
eight data sets in over 8 years. The aerial images completes the imagery
of 1946-2010. I'm extremely happy for the imaging campaign as has required
precision from the
navigators
and weather experts. I wonder what the summer was like for those in need
of optical satellite images - the flexibility is lesser in when being
spaceborne.
Two
peer reviews on the start list after the holidays.
Holidays wrapped up
Schools
in Porvoo started (Tue 14) with excitement in the air. 1026
hPa and +22Cº.
Taivassalo-Paimio
OH1-excursion to see big amplifiers, antennas and cool, old radios.
In the
eyes of the young ones - big capacitors were whitewashed by the mobile
stuff, which did not have safety belts.
Signs of summer?
In Finland the summers are
short, and, on account of our challenged weather, some years you need
observe the calender to be sure that it is 'summer'. Summers can begin
by
taking
slow steps, but I guess its only because we are so anxiously waiting
for them. And when summer has arrived, you are in hurry to do all the
fun stuff of summer.
Some
can even develop a stress for all the 'needed things to be accomplished'
during ones summer holiday (if you are lucky to have a job / be at
school). And the older
you
grow, the shorter summers tend to became. Time flies, as they say. The
signs of summer used to be a more common thing in the past during the
times when most
of
us were involved with basic production. Letting the calves out to craze,
sowing the potatoes by June 6, seeing the first birds etc., you know. Old
stuff. The signs in
an
urban environment are perhaps more linked with the calender that specifies
the seasons. They sell ice cream on strees during summer, schools are
closed during
summer,
outdoor concerts are held, sand on streets from the winter disappers
for the summer, Sun is up and you need curtains to see your favorate DVD
etc. I hate
summer
for the fact that it ends. Maybe I'm one of those with mid-latitude
genes. Migrating with the birds to Andalucia, Canary Islands, or Florida
would be a nice option. But
I
guess you cannot reach Northern summer year around - it is not some vegetable
or fruit that you can buy from the local grocery store where they nowadays
have everything
possible
available year around - no wonder we get confused with seasons.. :)
Ok, so much for whining about the summer's exitus. A few (first) days at the saltmine
will be the cure.
Fishing in Nikuviken has been
good in the summer of 2012 (Arttu). Cynoglossum officinale in our
garden (sowed in 2010) is starting to ripen. The seeds are aggressively
attached
on anyone tangent with them. Verbascum nigrum has spread across the
garden in the last 10 years that we've been in Kulloo. This summer it
was in flower for
weeks
owing to the weather (and still is in places). V. nigrum in Kulloo was
from Hyytiälä, from where it came from Veitakkala in Salo.
C. officinale has unknown origin, it
grew
on neighbor's doorstep in 2010, and now, two years later (bi-annual
plant), it's ready for seed-translocation.
.
July 27 - August 5: Savo tour
Aug 2, Savonlinna: In hope for sighting Phoca hispida saimensis.
(norppa)
Aug 3: A few norppas in Punkaharju? Kerimäki has Herttua with
Salpalinja. Saimaa water level was 50 cm above normal owing to heavy rains.
Aug 4: Leaving Savonlinna via Highway 6 takes you to Kirjavala in (Saari, map
link) Parikkala. The cows had been there before us. We saw B. lunaria and a milk lorry!
Aug 4: Highway 6 offers continuous barren roadside meadows - a
test stop 163 km from Joensuu proved it: B. lunaria (map).
Highway 6 is right next to WW II airport in Immola Imatra. I remembered
having read from Lutukka that it has interesting
flora (probably also insect fauna) - so there we went to complete the
WW II airport touring of 2012 (Jämi, Siikaneva, Vesivehmaa).
Aug 4: Immola airport was (far more than) exciting for moonworth
maniacs. All four species were there, abundant! Or, we think that we
saw both B. lanceolata
and B. matricarifolium.
The latter seems to be highly variable in appearance (Immola
map link).
Aug 4: Immola's only spotted B.
multifidum. In Luumäki, highway 6 offers Salpa bunkers
for bunker enthousiasts (map
link). We met also with the landowner.
The bunkers are state-owned and probably the 800 Salpa-bunkers
will be sold in the future to cover public expenses across the €-zone.
Archaeophytes and newcomers (neophytes)
Exotic
species are not always warmly welcome. Indian balsam (Impatiens glandulifera), Garden Lupin
(Lupinus polyphyllus), and
Giant Hogweed
(Heracleum mantegazzianum) are examples
of not-so-wanted-nephyte-exotics in Finland. But these are not the
only ones. So, why are these somehow
different
from the ample rest? There are exotics that came as 'stowaways' and
then there are those that are a result of our vanity and 'greediness'.
Some
archaeophytes
reflect changing climate, but most relate to changes in land-use. Some
exotics were fast in their invation - (Matricaria matricarioides)
pineappleweed
(North America) spread across Scandinavia in just 100 years starting
in 1850s. Someone brought it to a botanical garden, and now it
is everywhere
man has walked to. Ground-Elder (Aegopodium
podagraria) is also an archeophyte, and is, together with Campanula ranunculoides the
worst
weed to control in one's garden. Indeed, what are the aboriginals in
Finland? This question will remain unanswered unless one specifies a
reference
point
in time, before which everything was 'domestic', original. So, why all
the hastle about vanishing dry meadows and 'cultural impacts', that
need protection?
Modern,
effective agriculture and farming have brought in new types of landscapes
and ecosystems that take over the old ones. Why shoud the old ones
be any
better...
July
31: Savoninna's Patterinmäki (map
link) has vegetation that perfectly matches the definitions for
lush sites.
Actaea spicata, Elymys caninus, and A. spicata,
Stachys sylvatica &
Scrophularia nodosa.
July
29, Mikkeli Parkkola (map):
Epipactis helleborine,
Satureja vulgaris, and
Daphne mezereum.
Wild
basil is probably an archeophyte as it only grows on dry meadows in Finland,
which are all man-made and exist since 400-500 years only.
July
29 Mikkeli: The basic components of Finnish summer holiday - lake, territory
and some earthmoving capacity.
A in
Haapalahti (Rantasalmi municipality, map)
meadow/pasture in search for B. matricariifolium,
but Nardus stricta was only
sighted.
What is it about moonworths?
Some
old Flora had an introduction in which the author wrote that despite
being of no use at all, botany is something positive that for example
keeps
the common people from doing more harmful things such as heavy drinking.
Botany is not really a year-round hobby in the boreal Finland.
The
peak is just 8 or so weeks. Up in Lappland it can be much shorter even.
It competes time-wise with fishing, for example. So if you do both,
your
partner is likely to notice it! Bird-watching, which I don't do, seems
to be seasonal as well. Migration up and down, is a must in the spring
and
autumn,
I guess. But bird-watching must be nicer, as the observables are mobile,
unlike plants, 95% of which are perenials in Finland. After having
done
the routine walk-route 10 times, there's nothing to be seen in the Flora,
but birdfauna can surprise you on each occasion. You can watch birds
while
standing, whereas botany makes you bend and crawl at times. I suppose
that bird-watchers and amateur botanist share in common one aspect
-
you
never get bored in a new location. But how do you flee the company
of your hosts without offending them :) ?
Hirvensalmi-Suomenniemi-Mikkeli: Botrychium multifidum in a new site in
Hirvensalmi (map).
In the 60's, the location had been a part of a small "field-saw-mill",
logs to be sawn were piled,
where
the Botrychia now grow, since late 60's, it had been an 'abandoned
meadow'. The Syväsmäki site had tens of B. multifidum and more B. lunaria than in 2010, when we found
it (map,
map).
Dry
meadows are changing character: many sites with B. lunaria were again
found by highway 15 (Ristiina map,
map,
Suomenniemi map),
"the Botrychium road", while none were found
"tradional
sites" such as in Punkka and Hujala, in Suomenniemi. (map,
map).
"Parkkila
20" -meadow by road 15. While searching for Botrychia, you can come
across Centaurea phrygia,
and Equisetum hiemale.
July 25 Virolahti - Salpalinja
Looking
inside a WWII bunker in Ylä-Pihlaja, Virolahti SE Finland.
We did the first trip to Salpalinja in 2006, and have returned each summer
since then. Now, we had the
courage
to leave the museums and see the other sites. Ravijoki is where it starts.
I returned to Vahtivuori (Ravijoki village) after 30 years to show A
& P the exciting landscape.
It is
70+ years from the construction of Salpalinja, and a full forest grows,
where it was open in 1940. Roof of the Bunker (the rock is the 'sighting
tower') in Vahtivuori. Rapakivi is one of the
wonders
of Kaakonkulma (SE Finland).
Jämijärvi
Jämijärvi has eskers that have
been exploited and then there's Jämi. B. lunaria is abundant at Jämi (by runways) and B. multifidum was there
too.B. lunaria was found at
Jylli
village as well.
Diphasiastrum complanatum, Jämi Hotel
& esker (view from the southern runway), Deschampia flexuosa & Solidago virgaurea on the barren Jämi
soil.
Dianthus
arenaria was in flower at Jämi. It is really rarely seen
in Finland. Dianthus deltoides is more common.
Good
hobbies last long, Jämijärvi 2012 and 1995, the horse
has changed, though.
Våtskär - Pernå
Hesa Cup.
2012, July 3-6. Hyytiälä was multifootprint
scanned and documented
Friday: The FWF LiDAR campaign
was at 23-01 local time, so we documented some Hyytiälä scenes
in the mng, took another 268 photos at the spruce site (against the
blue skies)
and
finished the 2007 lichen site by some 60 oblique and vertical photos
of the pine crowns. AH calibrated the camera again, looking up and looking
down, and there was a slight
difference
(0.1% in focal length) suggesting mechanical instability in the Nikkor
18-200 mm lens. Well, no wonder. Finally at 5 pm it was time to go home.
Thu: Our home-brew camera head
for the 4-m-long monopod. The needed stuff. In the evening, we got
a call that the multi-footprint LiDAR campaign is being tried tonite!
Tue-Wed: More balls - returning
back to the spruce site to set up a tie point network some 6-7 m above the
ground. About 150 were hauled up there. The wind was calm, allowing
photography
(the polystyrene balls remain immobile), but it was very difficult
to have both the uppermost shoots visible in the images along with details
in the low canopy (CMOS dynamics).
The
photographs are taken (33% of them) from accurately positioned points
(marked with a 6" nail + piece of plywood), making the block geometry
more stable. The interior camera orientation
needs
to be established accurately, because there are no XYZ control in the
object space (canopy), just the corresponding tie points.
Tue:
22 years (summers) of search payed off finally: spotted one Botrychium
lunaria in Hyytiälä!
2012 seems to be a good Botrychium year.
Lessons
learned at Hyytiälä this time:
- Clear
sky, solar elevation at 15-20 degrees makes good conditions for from
below canopy photography (FOV +/- 30 degr.). The wind needs to remain
below 3 m/s.
- Check
the motion compensation in the lens, off is preferrable. Otherwise
the principle point might be immobile.
-
June 29 - July 2 Moonworth Hunting
Highway 15: This one is a couple of kilometers
south of Tuohikotti, where Viljo spotted another Botrychium lunaria.
On the highway #15, we decided to
examine the crossroads,
two outta three had moonworths; B.
multifidum and B. lunaria.
July 2: Kuomiokoski in Ristiina, we could
not spot any moonworths, but Trifolium spadiceum was abundant by the
old road. Road to home took us to highway #15, which is
nice,
because the roadsides lack Lupinus entirely and for example Hypochoeris
maculata delighted us in many places.
July 1: North Hartola: 207 m a.s.l. at
Purnuvuori. Botrychium matricariifolium,
finally, at Harju estate, where we saw (were taken to) four Botrychia
spp.
July 1, At Harju, the workers. A pair
of Botrychium lanceolata. Carex
buxbaumii at Purnujärvi shore. Last seen in Viitapohja Tampere,
1993.
June 30, Hartola.
Moonworth -hunt resulted in a couple of spots: first one, with some
20 specimens was an old (abandoned 1940s) sand pit and the second
by a bicycle road.
Botany
and museums need to be combined by some exercise to keep the 7-yr-old
interested.
June 29 Heinola railway station:
Railroads used to be one route for exotic species. Trifoliun arvense
& Euphorbia esula.
Tragopogon
pratensis, Vicia sativa (angustifolia) at the raiway stations. Moonworth
-hunters north of Heinola (a fine sand
pit for Oxytropis campestris near Suomäki-crosroads, road
140)
June 28, Svartså:
Kvarnkärret, Svartså,
Porvoo.
Bryggars,Svartså, Porvoo.
June 27 And back to Hyytiälä again
before holidays can commence.
The
"directional reflectance - an aid to tree
species classification" -paper was finally submitted. Then,
began testing if the vertical photos from last week can be oriented.
The
15-m tall, rather barren site was feasible to orientate (find conjugate
points, two bottom views). With 18 mm focal length, the FOV is 64 by
40 degrees. The first 25 images had a 4 cm fit to the GCPs and camera
positions. Top view.
Now,
in the spruce stand, that is quite dense, it was desperate to try and
find points from the (dark9 noise. It is feasible to do a pair, or a triplet,
with a maximum 2-m baseline, but after that
it turned
out impossible. We need to install artificial targets at about 6-7
m from the ground, to enable 'relative orientation' of the block, from
the known camera positions.
June 18-21 Hyytiälä again
In Hyytiälä Viljo &
boys got to experience real forester activity: Prof. Markus Holopainen
demonstrated three styles of 'leuanveto', 10 pulls each - continuously.
Forester Jyri Makkonen
on the
stock. And one of his pupils.
The
moonworth season began again, and we went to calibrate our eyes at
the Siitama standard. Darn difficult to spot, they are, says Perttu.
Earlier,
Mon, we saw some B. multifidums near Ollinkivi south of Siikaneva, at
an old forester hut. For that site we got a hint some years back.
Siitama:
How many do you see here? It's better to crawl if you wish to spot the
Botrychia. The famous and mysterious Siitama meadow - a site for Moonworth
pilgrims, but for how long?
If you
take the road from Siitama to Kangasala, you can visit 'Harala-hill,
with fantastic lake views. The site inspired Z. Topelius. There's now
a new tower - thus far surviving pyromanics.
in Hyytiälä.... Establishing
two photogrammetric plots - in a 50-yr-old pine forest and a in 55-yr-old
spruce forest. 12 x 10 m x 1 m grid of (vertical (up) photo) positions
measured with a total
station.
11 x 13 in total. Tue was a bit rainy - now trying photography Wed-Thu.
The plots were oriented by taking positions to trees with the total station
- these trees have aerial image
positions,
from 2009, and 2011. Wed was cloudy and windy, so just calibrated the
camera, practiced and waited for the evening, when the sky was promised
to clear.
Wed
evening 11 x 13 we took photos in the pine stand at 21.41 - 22:08.
A bit blurred with 1/100 sec exposure, and ISO 1000 sesitivity (F8).
Camera rotated perpendicular to the 12-m
long
side of the plot (offset towards S-SE). There were 42 control points.
Thu
mng, in the pine stand, at 5:30-06:05 clear skyes, no wind.11 x 13 photos
+ 4 + 6 oblique photos (later, at 08:37 - 08:42) from control points 43-48.
The polystyrene balls were measured as well.
Thu
mng, in the spruce stand, at 06:11-06:46, clear skyes, mild wind, 11
x 13 photos. Difficult to tune the exposure to get the dark stems and
the upper shoots (against bright skye) right.
Offset
towards E, perpendicular to the 12-m side.
Calibration
#1
Camera
number: 1, name: NIKON CORPORATION NIKON D300, ID: Nikon_D300_June20_2012
Calibration
Date: 20/06/2012 11:46am
Resolution:
4288 x 2848 pixels, Pixel Size = 0.0055 mm
c
= 18.684494 mm
xp
= 0.034094 mm
yp
= 0.054202 mm
K1
= 4.7617e-004
K2
= -4.8210e-007
K3
= -2.3048e-009
P1
= 5.0747e-005
P2
= 2.0063e-005
B1
= 0.0000e+000
B2 = 0.0000e+000
Calibration
#2
Camera
number: 1, name: NIKON CORPORATION NIKON D300, ID: Nikon_D300_June20_2012
Calibration Date: 21/06/2012 10:27am
Resolution: 4288 x 2848 pixels, Pixel Size = 0.0055 mm
c = 18.634550 mm
xp = 0.023148 mm
yp = 0.029217 mm
K1 = 4.9098e-004
K2 = -8.2476e-007
K3 = -3.8028e-010
P1 = 4.4218e-005
P2 = 1.8163e-005
B1 = 0.0000e+000
B2 = 0.0000e+000
June 15 Getting out of the Anistropy Labyrinth..
Tue-Thu all the simulations
were calculated on two computers and re-checked. The results make sense now,
and final reporting will commence.
A slight disappointment is the fact that directional signatures
do not seem to produce classification accuracies of above 90%, and the
gain from
using them is rather low, 1-3% in accuracy (and owing to occlusions,
true multiangular data will ony be available for the tallest trees).
Well, this
option is now investigated, except for the use of block-level
relative normalization, in which the increasing or decreasing at-target
irradiance is normalized
in the images, which will then retain the signals from the atmosphere
and the targets. If the first ones are symmetric, azimuthally, they'd
cancel out and
it might be possible to use the directional signatures anyhow.
Did a quick 6h peer review. I wonder if editors have time to
pre-check the papers, before they are sent to peer reviewers, i.e.
do they do any
high-pass filtering? Why do we have so many papers these days
- nobody hastime to read them, and it is darn difficult to spot the god
ones.
I personally don't believe that I could (with a small group)
produce more than 1.5 "worth-while-to-read papers" per year. Anything
above that, I'd need polishing
up, I mean, covering up, you know. I now start to better
understand the concept of "good-quality-journals".
The news on € that we heard this week were kind of absurd. The
Q remains - how is this possible? The answers of course are obvious
to anyone who perceives
the 'big picture'. You are likely not to get it by being exposed
to Seiska alone. That's like global news. Well, the local news (Thu)
from Salo weren't too good either.
What goes up must come down, or must it? Kinda funny - we are
pumping borrowed future resources, yes, future, to save the banks (who
are banks?) in Europe, and
there's not a penny left to support areas like Salo to survive
the devastating changes. And I pinch myself again, but the sad truth
is that I'm awake. Well,
it is all relative. People are surely pinching themselves in
Damascus, Aleppo and Tadmur too.
June 12 Nostalgic
With kids on summer holiday -
I started to remember how time was passed in the 70s, when I was a
kid. One very popular way of escaping the 'there's-nothing-to-do' -boredom
were encampments. That's where kids were transferred in big loads.
Another nice memory is shown in the picture on the right, that has
lots of men, 20 or so years old (in Hamina). The one holding
the pistol won some years later Olympic medal. In the 1970s shooting
as a hobby was very popular and there weren't too many people with
tensions or listening to the sound of hysteria.
In Hamina, people back then would only say, when hearing the
bangs, 'it makes me feel safe that they practise'. It's a bit different
now during our information age...
Hamina, maybe 1967 or so. At
the baseball statue. Old (now demolished) MarketHall is in the background.
I recommend growing up in a small township with both a sea-harbor and
garnison. The kid
admiring the baseball hero is my brother. Summa paper mill was
quite new those days, and brought jobs. Now it is filled with Google's
servers. Kinda funny...
Some more old photos by my father Eero showing Hamina in the
1970s. The walls were entirely restored by prisoners. I don't know how
long it took entirely, but nowadays the
landscape looks good in Hamina, with buildings and the fortress
in good shape. The middle photo has the reserve officers school, which
was downtown - soldiers were literally on the streets always.
Hamina, like Kotka and Lappeenranta, was a part of Russia 1743-1917,
which is displaeyd in many ways.
June 10 Doing the tango - back and forth
Tango dancing reports work activities
100%. Started simulations from the beginnig by adding 100% more flight
lines to handle issues when overlaps are high.
The roadside from Viikki to Kulloo was reddish, from Lychnis
viscaria in full blossom this week. Stopped on the way in Sipoo to see
how the summer is taking slow steps.
A big day for MV on Friday. Kustos on the left and opponent to
the right. This is pre-defense smiling. In Tampere to learn that 12
months pass by quickly.
In Elimäki to enjoy,
tnx.
June 3 - Week of real and simulated events
'Struggling' with the simulator
to study the effects (on tree species classification) induced by image
calibration errors. Realized that to simulate properly the QDA approach
(discriminant analysis),
which involves a training set and a validation data set, I need
to have 10000+ trees in the population (9 km2 forest), and different
training sets with 0.5-5% (25-200 trees per species) of the
population sampled (spatially) in either systematic or random
manner. Four flight lines result in 104 images (4 x 26) - each with some
offset and trends in the observations. The
representativeness of the field sample (trees, plots) with respect
to the images (anisotropy, image-level calibration issues) needs to
be considered.
With 10-fold number of trees in the 'forest' - things started
to slow down again. The anisotropy-match algorithm is slower and s.b.
supplied with smaller populations. So work continues
though would like to give up for the time being. Radiometrically
quantitative photogrammetry suits better post-holiday brain.
The use of anisotropy signatures does not require training data
(if we trust the 'brdf' models, and accurate refl. data are in use),
but for possible calibration purposes. Thus,
for comparison of the approaches, a varying training scenario
set is needed for the tradional QDA approach, to assess the influence
of sampling ratio (and method) - but
there is no definate (single) way of comparing the two methods.
Summer holidays began on Saturday (lasts for 10 weeks in Finland)
and kids at Kulloo school were given grades. Excitement in the air as
always. Viljo presenting his first ever.
Oh, those were the days - when €-countries still had
manageable loans and Iceland was extremely expensive to visit. A
look at old scans from 1998.
MK payed a visit and we uploaded the scanned jpegs of the last
set with Iceland photos from July 1998. Coutryside was empty and people
swayed in Reykjavik.
First thing that I remember from Iceland is the price level,
as it was skyhigh, and we could not really afford even B&B places.
The nature is interesting: you realize that the human
activity has resulted in an eco-catastrophe,with wooded areas
entirely missing, peatlands drained, peat/humus blown in the air etc.
Ophioglossum azoricum, when found Central Notrh Icelad,
was
one of the many highlights of the trip. It is a really small
fern, which in Iceland grows near the hot springs / volcanic soil.
The partners of the catastrophe (The tree is the biggest tree,
downy birch, in Iceland. The pictured forester is of course counter-attacking
the catastrophe).
Larch planted on 'humus-free' soil. Lupinus nootkatensis
(exotic from Alaska, a problem now) was widely sowed as a "nitrogen-to-soil-plant".
Iceland's largest forest, fenced, planted with exotic species.
May 30 Oh Bug! - Again
Returning back to the directional
reflectance signatures -simulation in R... While reading the 2 month-old
code I came across a good explanation as to why I got stuck 5 weeks ago.
One of the for () -loops in R had a length(*) function call.
The data structure * was 2-dimensional, and the return value corresponded
to the number of columns rather than the
number of rows (observations), which was expected. All computer
intensive simulations need to be re-run, and now the code is also much
slower.
May 29 Summer is late
The current temperature sum
in Hyytiälä is ~160 dd5-degrees. Just when the birch leaves are
starting to be big enough (@200 dd5) for aerial photography the weather
is turning unfavorable.
Well, less watering in the garden and good news for cereals in
Uusimaa.
Desperate final tuning on the anisotropy paper. Well 'final'
is a relative term. Still haven't entirely understood what the analyses
are/were about,quite many details in radiometry & statistics
that need careful documentation and reporting - not to mention
the tens of R-scripts. Anyway, main results are according to
hypothesis and in that sense writing is confortable.
I just hope to get rid of it and enjoy some summer vacation if
possible. It's getting boring, tedius, dull, blunt, and gloomy. A researcher
quite can't have it that way.
May 26 Aviation fruitcases gathering in Vesivehmaa
Vesivehmaa, finally! A fine event was organized on
Saturday at Vesivehmaa, by the aviation amateurs, who displayed all
possible ways of getting airborne. And the museum was open
as well. To our surprise we met many hams there too, pictured
are OH2BP & OH2BU.
May 23 - Two week orientation procedure is over.
There were 86 images, with maybe 15-20 image observations
in each. iWitness is fast, yet manual triangulation of from-below canopy
images isn't very fast. Now, these
images and the 250 from summer 2011 wait for the analyses. In
the object space, the accuracy is 5-15 cm in X,Y and Z (forward ray intersection)
for reasonable distances.
Documentation (html)
The dry Picea engelmannii, with three pulses and five echoes.
Distant birch branch with four pulses, intensity color coded.
Birches, pines and a douglas fir. Size of the dot = f(distance
to camera).
May 20 - OH NO! THERE'S A BUG!!
AH is making steady progress in developing the
synthetic vegetation for the simulator. Definate Graminaceae specimen
to my eye, and red laser, this time.
An unfortunate bug was found in the triangulation software. Filling
the design (A) matrix with contributions
(partial derivates "1") for the tie points had a column (cols link to
unknowns)
index issue. This is why the convergence wasn't always reached...
Now, it is as fast as it theoretically should, as 2-3 iterations suffece
with good initial approximations.
The most annouying thing always is that you realize that you've
probably spent a lot of extra hours by being unaware of the (this kind
of ugly) bug, trying to find the solution from every other
possible place...The block adjustment code was changed in Oct
2011, to support full use of weights, and built upon existing code, which
has its dangers. It is easy to copy-paste
seemingly working code and then you think that 'it's all right',
as the results seem to support it. In hurry, you easily omit proper debugging
, testing with known input/output scenarios.
This photo shows intensity (discrete-return) data
of Riegl scanner (all kinds of pulses/echoes). With a good imagination,
low values (dark blue) are associated with
thin/sparse targets, whereas stronger intensities (red) have
been produced from the stem, or thicker branches. The data is from two
strips. Now working on getting the
individual 10-30 photo blocks to match one another (where there
is overlap) and the aerial images. Restriction is the 10 cm XYZ accuracy
of GCPs. A close
GCP can turn the block locally even if the weights are set right.
Or can it actually, if the weights are really set right? Now, I used
10 um for image observations (2 pixels)
to go with the motion (wind) of trees, 0.1-0.2 m for the aerial
photo XYZ points (man-made), 0.2-0.5 m for treetops, 0.05 m for Z in
flat areas.
Reviewed another manuscript for a RS journal. It was my
42nd, and of these I believe 37 have had the recommendation 'reject',
while only five were with 'revise'.
That is 12% only.
May 15 More orientations
iWitness shows in 3D (in a way) the ray bundles.
The pic shows the 6 camera positions on the field west of Muistokuusikko
and the Riegl LiDAR of November 2011 is overlayed in the two photos.
Good (X)Y and also Z match with the features in images. The Riegl
scanning took place 750 m AGL and
A 2 cm aspen brach has given raise to two first of many echoes.
The branch was positioned in 3D in the images, and multi-echo pulses
with first echo XYZ vloser than 0.5 m being the criterion for drawing.
More old analogue photos scanned... The Pulsatilla
was pictured round about this time of the year in 1994, in Hämeenlinna.
The photo in the middle is from Iceland in 1998 and the rightmost
photo was taken in August 1996, late in the evening on the base of
Fjäll Saana in Kilpisjärvi (Epipogium). The excitement
of those days does not show in good qualiyt of photography. Too much
excitement is
never the optimum.
Mother's
Day 2012.
May 7
IIHF Worlf championships in Finland and Viljo
was lucky to witness the FIN-SLK game. "Awesome" said Viljo. The 7-week-old
puppies are making a mess, but paying for it by being so cute.
Spring is taking slow and wet steps in Kulloo. Hoping for leaf-on
condx soon, as this week we agreed on some LiDAR and photography campaigns.
The rest of the week was spent in meetings
and getting the last week's photographs oriented. It takes time
even though iWitness is a good tool. Finding the correspondence in below
canopy photographs having elevation angles between
0 and 90 degrees (perspective changes) is not as straightforward
as I thought. RMS of 1-2 pixels remained in the blocks, which is understandable
given the 0.1 m accuracy of the GCPs and the
distances of 10-50 m to targets.
Next xmas card?
May 2-4, Photogrammetry in Hyytiälä
to go with the November leaf-off LiDAR
Some pictures were taken by just walking along a line and taking
convergent photography towards the buildings which have details seen
in aerial images. Given the RMSE of 0.3 pixels
of camera calibration, and the 0.1 m accuracy of the GCPs - it
should be possible to triangulate the block reasonably accurate with
the iWitness software. The other pictures of trees
were taken from known (XYZ ~0.01 m) camera positions, which should
improve block accuracy. The polystyrene balls (~0.05 m) help in identifying
scale issue due to focal length-
We decided to include four
scenes: i) aspens, ii) some exotic trees, iii) Muistokuusikko
from distance and iv) some birch & pine crowns. Little placeholders
were built from
plywood
and four-inch nails. These were 'anchored' to ground. Ropes with balls
were cast on trees and positioned with a thedolite. Wednesday was planning,
testing and
calibration day. On Thu, all geodesy was done. On Fri geodetic calculations and actual
photography. The idea was just to get some samples of leaf-off crowns
with waveforms.
The theodolite etc. computations were done (accuracy < 0.1
m) and the photography took place from on Fri early morning thanks to
AH.
Wind was milder on Fri, but the sky was somewhat overcast or
clear. Nikon D300 was used with the 18-200 mm lens at f = 18 mm ('glued'
there), and Canon powershot
similarly at shortest focal length (7 mm). Both were calibrated
at the beginning and Canon was re-calibrated at the end. The pixels of
Nikon are
5.4 um while in Canon, just 2.3 um. Report.
As we did not have a Network-RTK
GPS, and would be taking pictures partly under canopy, the positioning
relied on details seen from the Tacheometer, that
needed
to have visibility to camera locations and the crowns (ball GCPs, seen
in photos above). The details were from buildings and other man-made structures
seen in 1:6000-1:10000
aerial
images. It worked out, given the ~0.1m accuracy of the airborne observations
and the 0.5 mgon accuracy of angular measurements in the Geodimeter.
In
a test image pair, the GCP RMSEs (forward ray intersection) were 0.1
m at triangulation distances of 15-40 m. AH calibrated Powershot at
the start, and the
principal
point had travelled to a new location since last calibration (poor estimation
accuracy of those interior orientation parameters). The resection was
done
across distances of over 100 m, which is feasible only given the high
angular accuracy. Tacheometer unknowns were the XY and kappa, the rotation
of the
instrumenet
with respect to the used 3D coord system (KKJ/N60). SDevs were reported
for X and Y only, when more than 3 observations were made.
Vappu.
April 30, Stuck with simulations
Some issue came up with the
spectral classifications utilising synthetic, simulated image observations.
Using frame sensor
data, the results are insensitive to radiometric offsets and
trends. Probably a programming error owing to my poor R skills.
The UCD campaign was laid out, but the 0.5-3-km multifootprint
LiDAR campaign is still unfinished. It is difficult to attain constant
SNR
for large targets and canopy targets (with LiDAR). In fact, it
cannot be done at the same time as the R-dependence in the radar equation
is not
unambiguos.
We decided to stay near Hyytiälä forest stn, to photograph
the leaf-off vegetation as it was Nov 15, when the fwf-scanning took
place.
We will try locating the camera positions using man made structures
seen in the images (back-projection, resection), or with the tacheometer,
which was tested again. Another option is to climb to the roofs
of buildings, and place the camera at points that are seen in the high-res
aerial images giving sub-decimeter positioning accuracy.
What are foresters needed for?
Univ of Helsinki (UH) has had 'academic forester' -training for
over 100 years now - Hyytiälä turns 102 this year - it has
been the place to learn things in practice.
In the early 1980s, forester training was increased by establishing
the Joensuu faculty, now with Univ Eastern Finland (UEF). The early foresters
got jobs with
Metsähallitus (State forests), but also in the private sector
that was organized in geographical districts (now under threat), with
the intention of improving the silvicultural
status of private forests, that had been subject to slash and
burn (some areas still 100 yrs ago), and selective cuttings. Forests
then did not have other value but
that of the source of construction timber (log buildings), and
pasture (grazing). "Zero-limit" was scattered across the country, and
excessive use of fuel wood
had reduced the growing stocks of forests near the few cities.
Foresters were needed to organize a better use of the resource that
the country had and
played important role in getting the economy of Finland up. Still,
in the 70s, kids at schools were taught that 'our country stands on
wooden legs -
the legs of forest industry'. Foresters and their organizations
saw to it that the zero-line disappeared in most places (wood procurement
became economically
sensible), and the growing stock and growth of Finnish forests
went up. Forest ameliaration was the (now forgotten) word, and in Finland
it was mainly about getting rid
of the excessive water - drainage that is, but also fertilization,
selective intermediate fellings, tree breeding efforts etc.
The foresters, back then, knew what they were trained for, and
why they seeked the training at UH. And the profession had a good reputation.
And I guess that
the teachers also knew what their target was: training of foresters
with knowledge in technology, economy and biology of forerts and forestry.
Foresters
were also taught to respect the nature.
The change came with the digitalization, mm. Information systems
as tools reduced the need of experts to run forestry: management, planning,
silviculture and
wood procurement. Machines took over the harvesting work quickly
in 5-10 years, starting 1985. The need for people managing the work
disappeared. Forest industry
that saw to it that there is demand on round wood started to
invest elsewhere but in Finland in the mid 1990s. Fast growing tree
species produce fibre for pulp
at rates that can not be achieved in Finland. Paper mills are
needed close to the customers, where pulp can be transported long distances.
And consumption of most paper
goods is declining in the old market areas, Europe and North
Am. The low demand shows in prices of round wood - you need a much
larger clear-cut to pay
for an apartment in 2012 than in 1970. Zero-lines that were pushed
aside by foresters and the markets (demand), and now returning and
the profitability of
forestry is going down. Forest companies are willing to sell
their forest ownerships that were strategically so important still
15-20 years ago. People's values
have changed as well. The baby-boom generations still had a 'connection'
with basic production (agriculture & forestry), but their kids
and grandchildren, the
new forest owners do not see it that way anymore, and do not
need the income from forest. Many of the 300,000 forest owners have
small parcels, that are too
small for sensible forestry. We can afford this, and we can afford
to be very 'ecologic'. What used to be a stand is now an ecosystem with
biodiversity. Foresters
were good at counting trees but we are armless against generalized
argumentation stating that forestry is non-ecologic and kills species
in extinct. We may ask
for data to support the green statements, but since that sort
of thinking is not mainstream among biologists, they can just answer
that question with another
generalizing argument. Everyone wants to save the planet, as
long as it is not affecting their standard of living. We want the wolves
to survive, far away from where
we live, and we wave our hands to people near Talvivaara mine,
but use the 230 VAC produced by Uranium that is being mined from there.
Foresters have lost the war on ecologists. They made the same
mistake of being arrogant 15-25 years ago, as Nokia folks did 15 years
later :) And the same thing
has happened already elsewhere - the number of forestry faculties
is truly not increasing but decreasing. We see foresters merging with
biologists etc. I would
not mind going back to where we initially left 90 yrs ago, the
technical university and surveying. But that discipline seems to be
under pressure as well. GPS has made
anyone a surveyor... All in all, the structures (and people like
me)in universities seem to be relicts from old times and not perhaps
in best way surving the society these days.
But who has ever seen a public institution stating 'we are not
needed anymore!', and who has ever heard an objective statement (honest)
from a researcher saying
'actually what we are doing is not that important in the big
picture' - nobody. Instead, I encourace everyone to critically rethink
all statements by researchers,
because they (we) are human beings and may exaggerate things
just to assure future funding. And I think this a particular trait of
ecologists.
----------------
Donno if it is true, but it
seems that having three R-windows open does not slow down any of them.
The four processors (EliteBook 8460p) share the workload, and
each processor is running a separate
R window (25% of total CPU load) under Windows 7. The simulations
take quite some time to run.
8000 trees imaged 100 times and classified with the Mahalaobis
sum -metrics. For-loops, multiplications,
1536 x 1536 var-cov matrix (3 x 4 parameters for 128 images),
and some file I/O.
TBD:
1. Leaf-off terrestrial photography -campaign, leaf-on campaign
agreed: select locations and prepare Riegl/ALS60 data for the plots.
2. DL May 31 for submission of BRDF-paper.
3. Prepare multi-footprint LiDAR campaign plans, as well as UCD
"fixed exposure" flight plans.
i. Tune the frame sensor imaging simulator (limited view zenith),
probability function for discernibility f (Hrel, theta)
ii. Simulation script for synthetic frame image data classifications
(simulation of calibration errors in upto 128 images).
April 19, Simulating multiangular aerial imaging and spectral
classifications with R
THe LAS-WDP issues were reported back and the findings we made,
also made sense.
Tuning the R-scripts that carry out the simulations. The flight
simulator part was made toproduce a new tree and observation set for
each simulation run (100 in total per trial).
The frame sensor flight simulator needs further tuning,
as we cannot exceed the xy-geometry in the ADS40 data. NormalAngleSensor
geometry will be needed. The simulated
observations can now be inputted to QDA-classifier (averaged
features), or the BRDF-match-algorithm. The tests will include scenarios
of reflectance calibration errors. Normal-
plausible, and very accurate-optimal. R is slow (in my hands),
so it takes some time to experiment.
The last snow are melting, slowly. Indoors, the six cuties are
delighting us. Turning 4 weeks, they are, all six.
Compiled a little photopage
of some Hyytiälä film-photo scans. And another one from Botanical
trips.
April 13, LAS obscurities uncovered
LAS 1.3 format specifies a
point (.4) format, in which waveform data is associated with discrete-return
points. The first sensors produced first/last data, while
the successors were able to extract 1-4 points per returning
backscatter surge. Now the format reserves three bits for the number
of points in a pulse that the point
is a member of. LM680 was able to extract as many as 10, in postprocessing,
using the saved waveform samples. However, these are stored in proprietary
format
SDF files that we are reluctant to read. Three bits sets the
limit to 8.
AH also recovered the cause for points without a waveform sample
(PacketSize == 0, WDP FilePointer == 0). If the point was exctracted
"at the end" (border between two consecutive)
of 80-byte wave sampleblocks, the latter block might be missing
in the LAS (WDP) file. They seem to be stored ok in the original proprietary
format waveform SDF file.
The data with missing sample blocks was overlayed on the map
- and it showed correlation with the vegetation type (80 samples x 0.15
m/sample = 12 m).
The Hyytiälä
FWF LiDAR datasets of 2010 (July ALS60 1, 2, 3 km) and
2011 (August ALS60 0.7-2 km; November LM680
0,75 km). LL corner is 2510000, 6850000 in KKJ.
Colors depict pulses per hectare. 2011 scanning includes Lakkasuo
mire (to match the 2006 campaign), and Susimäki old growth. A larger
area was scanned to support
course marv106.
This shows that research can also benefit teaching although cost
accounting at UH doesn't see it that way. A year back some researchers
felt that the 112%
overhead costs in research projects are exaggerated - a week
back we learned that it is going to be 128%! I needed to pinch
myself, but to my misfortune - I was awake.
The likely explanation to 128% could be to avoid or delay, termination
of tenure jobs at UH. Research projects have non-tenure jobs, which
are secondary, it seems. Secondary to
rise in wages of tenure jobs etc. A few years back many tenure
jobs were established in our faculty - despite shortage of funds - and
the resolutions included a clause
"the department will see to it that (external) funding suffices
to pay the costs". Typical or what, but these clauses are now, 2-4 year
later, "just air". So, those who had the
guts to go for such jobs were the winners in this 'zero-sum-game'.
That's called academic politics. If you are not playing it right -
you may soon notice your domain even disappearing.
If you play it right - the sky is your limit.
Work on the ADS40 paper was resumed. Systematic classification
trials with real / simulated data.
Old pics scanned - nostalgic...
The start is always a bit clumpy... Two pics from Hyytiälä
May 2001 preparing for aerial photography that never took
place owing to poor weather that summer. That is Mr
Stuart Fish driving the 1973 Volvo 142. "Lena" I called her,
since 1993 when I made the purchase for 600 FIMs. It was about to fall
in pieces in 2001, but being just on some grants
would not buy a proper car (still haven't got one). The pic on
the right is from July 2000, and shows first-class RTK measurements
(positioning) on the Hyytiälä GCPs. In 2000,
we had yet not heard about direct georeferencing or the Network
RTK (VRS service) - and went around Hyytiälä with a 30kg drill
and metal bolts. Funny now, in retrospect.
When one visists Hyytiälä, it is good to have some
extra time to calm down and get to know the surroundings. These pics
are from 1995, which is when the WLAN wasn't
yet operational, nor any open internet (came with SMEAR II in
'96), and thus you were not tempted to work 24/7, while there.
One thing I've learned over the years about photography is that
you should always use some sort of stabilizer to assure sharp imaging.
In one's excitement you travel long distances
to take pictures and greediness will have its revenge. You end
up having lots of pics with poor sharpness. And during the film era
you couldn't see it from the camera screen instantly.
Lakkasuo white moss, meadow in Nummi, Polytrichum moss in Leivonmäki.
April 10, Winter lasted for 3 months - if measured as the length
of the cross-country skiing season.
The Riegl LM680 waveform subsampling at below 1
ns is not feasible in LAS 1.3. Disappointing for our DEM efforts. One
cannot determine the exact time for the waveform samples.
The November 2011 waveforms were sumperimposed in below-canopy
pics from June 2011. It turned out the we were the first customers ever
to ask for the waveforms of the sensor,
which has its home in Belgium...Discrete-return data suffices
for customers these days - shows that waveforms haven't yet exposed their
secrets or benefits to the general public :)
The pulse data structure that AH proposed has a fixed (pulse-level)
part and a number of Echo-parts (36 bytes each), which hold links to
the waveform sample blocks.
I had thought that "structs", or user defined data types in VB
could not have dynamic arrays as 'attributes', but they do, so we abandoned
the fixed-length
(207 bytes per pulse) philosophy, and the Riegl data has no "zero-echoes"
(as I used to have redundant space for four possible echoes in the
comprehensive pulse format since 2004).
The reading is no longer possible with a single Get (get the
whole fixed-length array) statement (in VB-kuvamitt), but now we a
little bookkeeping is necessary to know where to put the
read-head in order to manage the variable-length pulse records.
AH managed to get the per-hectare-split going at fast pace with Java.
The focus s.b. put back to the ADS40 project again.
April 8 (59º46'N,
21º22'E)
Utö: Västudden - Enskär - Fladan.
Bird-watcher cult was strong on the island over the Easter. Binocular
and telescopic lenses pointed at the sea - they waited for "the thing".
April 5, Great skiing in Sipoo, still.
The Wed LiDAR-BRDF seminar was successful, with 18 participants
out of 22, and we all got some good feedback, I hope. LiDAR simulators
were constructed
by others before us - and the brdf in trees - it might not be
possible to find prototypes - think of bkground signal in sparse vs.
dense crowns.
The Riegl LM680 data conversion from LAS1.3 to 1-ha-pulses-in-one-file-one-record-per-pulse
-format continued. Indeed, the waveform data is in
fixed-length (80) sample blocks. Now, one block can give raise
to several DR-echoes. LAS1.3 (point 4) has a special attribute "RetPointWaveformLoc",
which is a single precision float (4-byte float), giving the
time difference between the start of the recording (first sample, at
1 ns sampling), and
the range for that echo. Now, our data shows different values
for each of the echoes within one sample block, but the times are all
"##000.##" picoseconds.
This indicates that the 3D distance (range) would be multiple
of 15 cm between echoes. Now, something is seemingly not according to
the specifications (RiProcess?).
The geometry of the DR-data was verified in one flight line,
where the LAS file had a size less than 2 GBytes, which is the limit
for VB. AH codes the
final program in Java, where the file pointer can be set above
2GB. Yet, in Java, we need to be careful with ENDIAN issues. The picture
aboce shows
points < 1 m from the DEM, coloring from -0.4 m to +1 m. Leaf-off
LiDAR data - the first from Hyytiälä, Nov 2011.
April 3, Still skiing in Sipoo
April 12 is the record in Porvoo, from 2007, if
I recall right. New fresh snow and skiing-enthousiasts are back on the
tracks.
Mon-Tue we prepared for the mini-seminar and investigated the
Riegl LM680 data, which was delivered to us in LAS format.
To our surprise, it seems that the sensor is not recording waveform
blocks of free length, but our data shows that the blocks
are always 80 samples (@ 1 ns) long. The sequantial order of
the echoes (pulses) was also somewhat distorted so we need
further processing to collect the waveform blocks and discrete-return
echoes that belong to each pulse. We will anyhow distort
the order by rearranging the data into 100 x 100-m blocks, which
have proven handy in the past. The picture shows an 80-sample
long block. The next block is not necessarily following the first
without a gap. Or so it seems at this stage. We have to investigate.
March 30, Taking a pause to see what we have done and where
to go next.
Continued tests with a general
BRDF correction - in most cases it failed in improving classification
accuracy when compared to averaged image features.
In spesific train-validation scenarios it helped. So, in real
data the general BRDF correction does not help if BRF images are available
(for tree species classification).
In addition, the gain from using the BRF-match features and Mahalanobis
algorithm in real data (with substantial calibration errors) is low
(compared to traditional
classifiers and averaged image features). Still need to systematically
verify it. R scripting is starting go smoother, owing to help that
is available.
For one day (March 28-29), we were on the frontpage of www.helsinki.fi/yliopisto
(news
on us).
The November FWF scanning was received, now in standard LAS format.
Q/C remains next week.
The LiDAR project was set to a new course - a few vegetation
types will be simulated in more detail. Discussions on "what do we
see in the returning waveform that is
reproduced by the sensor?" I.e. what we see reflecting back from
a well-defined perpendicular surface is not necessarily 1:1 copy (shape-wise)
of the transmitted pulse.
As to how much is the signal distorted, and why - remains still
a bit uncertain. We would also like to know what happens to the transmitted
pulse, when the system
is modulated faster (PRF), or the output power (energy) is adjusted.
This sort of information is needed in the long run to reach quantitative
RS with LiDAR.
We had decided to give talks in order to get some expert feedback
- and it will Wed April, 4. Talks on "Waveform LiDAR" & "BRDF of trees"
are stored in this folder.
March 28 To BRDF or not to BRDF?
The whole idea that you consider the BRDF of tree
species, and measure the match of your multiangular observations to
these, and select the
species according to the match - was questioned by some results
in which the features were simply of type mean(Im1,Im2,..,ImN), and fed
to a classical classifier (QDA, LDA, knn) that was taught with
a subset of data in images 1..N.
In real data the gain from the "match-classifier" was marginal,
gain over using just the mean features. However, the classifier works
in that sense
that increasing observations, improves the accuracy, although
it saturates quickly at 80%.
In the simulations the situation looks brighter (85% oa-%), except
when the between- and within-image refl. calibration errors are allowed
to grow, to levels that
we have observed in Hyytiälä with refl. tarps (and
are observed from space, for example). That is when the simulations
match results with real data.
March 26 Must Do - The
Strongest Motivator?
The government is cutting back University funding
for 2013. Is it going to ruin everything? Will we collapse? The answer
is of course no, and the
argumentation is simple - universities, high-schools, colleges
etc. rationalize their modus operandi only when they have to. This simple
rule in life
that was taught to me by TT, works universally at the level of
individuals and communities (of public administration, any public squandering
of taxpayer money).
I just hope that the oppurtunity (to see what is essential) will
not be wasted by excessive "politics".
March 24, Excitement in the air...
Tassu delivered
six cuties !
March 23 Anisotropy classifications
The simulator is functional. It seems that the Mahalanbis
classifier works with almost any VAR-COV structure, but the results
are also sensitive to the
imaging geometry (overlaps, solar illumination vs. flying dir).
It would be perhaps better to simulate a new set of trees and a new flight
for each simulation round.
Also, tests with real data must be tried - elimination of reflectance
calibration errors using known trees as targets across images.
So, there is plenty to do. The mixed effects model offers some
possibilities to refine the classification results, and these must
also be analyzed.
We must also check for overparameterization of the anisotropy
prototypes. Limiting of the FOV in the simulator might be the right cure.
TEKES project 2010-11 came finally to an end. Accounts were checked
by a third party.
March 19 Old films go to scanning - R simulations converging
to a solution
Does time fly? In 1993-2003 I had access
to a Canon F1 film camera. These pics are from Hyytiälä in 1995
scanned by Markku Korhonen. We started a process to scan 'em all.
The simulation of multi-angular (aerial photography) image classification
of tree species is making progress like Tango, a few steps forward
and some back.
Now, the simulation, using reasonable accuracy for image reflectance
calibration, provides overall accuracies from 81 to 88%, when using
from 1 to 8 images
per tree. LM's help in making the var-cov matrices for the simulation
of mutually correlated (between bands, between images) errors was indispensable.
We need to debug the classification algorithm still. Now it applies
Mahalanobis distance, and the source of the needed covariance matrix
must be analyzed
further.
March 16 ADS40 & R week
Mixed effects modeling of the
ADS40 reflectance dataset from 2008 revealed interesting stuff. The
model is
Refl = Anisotropy_Polynomial + Tree_effect + Image_effect + Epsilon
The polynomial is the fixed effects part, and it explains anistropy
with respect to the xy polar geometry of BRDF (omitting solar zenith).
Tree_effect explains the variation (random) due to the tree itself
(being bright or dark in all views), and the image effect explains
the random
variation in the 20 reflectance calibrated images, taken between
07UTC...09UTC Aug 22, 2008.
These models were fitted in 24 examples: for sunlit and shaded
tree crowns, in four bands and three species.
The image effects showed time-dependence (flying height + solar
elevation trends), and variance explained (in %) were largest in BLU
band (as can be expected),
while reducing for GRN, RED and NIR, in that order. Tree
effects were largest in NIR, which also makes sense. In NIR, anisotropy
does not explain reflectance
almost at all!
The question now is, what is the anisotropy that we observe in
shadow (diffuse light)?
First attemps to simulate the reflectance calibration errors
were tried.
AH cntd with the LiDAR simulator. Initial plans for the study
report I were agreed. Three main parts:
i) FWF vegetation simulator, ii) feature analyses for short vegetation
(simulation & real data), iii) DEM enhancement with waveform traits.
Data was sent to UEF/LM for the crown modeling / crown penetration
trials (2011 FWF + 2010 DEM).
Two PEER reviews await, for remote sensing journals. Feedback
on the Python course came: "Simplify the exercises for 2013", now many
of them were
"undoable", and this did not awake "killer insticts" but in few,
who actually thanked me for setting the bar high and giving a challenge
for learning. In Finnish
we say "Bog here, ditch there, where to go?"
Are all students equal?
In shools in Finland, the words are "Eriyttäminen"
and "Integraatio" - the first means that the same stuff is taught
individually for each pupil, considering his/her
talents. Integraatio means that all students occupy the same
room. But because of lack of time, the consideration works only in
one way - those in trouble preoccupy
the teacher and those with particular talents - no time left
to nourish these. In the early 1980's - I recall my high-school class
being split into "advanced", "middle", and "short", or
something in that direction - being split, when a math, english,
physics, swedish or biology class started. You could concentrate to
learn, owing to the "disintegration".
But that' system is no longer in use - probably for a good reason,
which I've missed. And I wonder what it'll be like for my kids...
In University, all students (who do well in extrance exam) are
given the right to pursue a degree. And it is basically free of charge.
In the old days, rights used to go
hand in hand with responsibilities. Now, to me, this does not
seem to be the default way of thinking anymore. "Responsibilities, what!?".
"I have rights in this society, go and preach elsewhere!".
On a course the teacher should maintain objectivity - I know
- but I can't help thinking positively about active, humorous and motivated
students. I wonder how wrong
this is?
March 15, RrRrrRRrr... 
Trying hard to get something
sensible done with R. It seems that the standard way of error reporting
isn't as good as it could get. For example,
when slicing a list-structure (vector?), a[start:end], "start"
or "end" cannot be expressions. Or they can (techinically), but the
expression is not evaluated, or
an error msg is not raised, and the error msg from a function
that is supplied the "slice" says something cryptic, not related to the
sliced object.
I suppose that the developers have strict rules on the syntax
and error-handling in functions that are supplied by the developer community.
It is a whole high-level language for math/stat/graph. If you
don't know the philosophy, it is probably easy to write very slow and
inefficient code.
At least so it seems, when I'm indexing list items in loops.
I got ready-made code by an expert and it shows in that in good hands
R is an efficient tool.
I just feel like having learned enough programming languages.
But, if yours is no longer in use by colleagues, how can you communicate?
With R I have
the same feeling as when abroad in a country that you don't know
the language - you end up eating the same stuff every day, that you
learned to
express in that strange language.
March 12 Academia - "Publish or Perish!"
My PhD friends with past experience
in the Academia gave some comments on my March 10 thoughts: "The difference
between Academia and other jobs
is that there is politics and harrashment in both - but industry
jobs have something fun to add on top of that - which is entirely missing
in the Academia". "It
can't be that bad?" - I responded. I got serious smiles back.
Now, this is tricky. The fellows were saying that they are much
happier doing the R&D in industry than they ever were at their respective
universities. Is R&D
in the industry to be taken seriously? Is that science? Their
reports will not be seen by many - but products sprouting from the R&D
will. And you can always
reverse-engineered those somewhere in SE Asia, and start to benefit
from that too. So the harrasment is outsourced (to China) in some
sense!
Results that are published are free for anyone to take advantage
of. You might reason here that if you find out something that could
be turned into cashflow,
it could perhaps be better to withdraw from publishing. It could
even be wise not to tell any of your colleagues about it, and just silently
quit the research institute.
Well, this is known to have happened often. And this is why we
nowadays have fake scientific journals and conferences, which try to
allure (e-mail spam) you into sending your
reserach material to them. Think of that! Yet, in many fields
the pace of new scientific articles is record-high, and it has become
very difficult to follow the state-
of-the-art in ones domain. There's simply too much garbage being
published, as electronic publishing has lowered the thresholds.
Can we therefore trust that the best findings in medicine for
example end up being available to experts around the world? We have to
hope that it is so.
The moral of the scientists is constantly tested. One needs to
publish at high pace - the individual articles become thin in content.
It is your interest to publish
the smallest publishable unit of information, to add an article
to your CV. That increases the number of articles and pages that one
has to read to find the valuable
information. There is no sense here. PhD-theses in the last 10-15
years have changed from monographs into 3-5 article compilations. Can
you expect high
quoality, state-of-the-art research from a PhD-candidate with
his/her papers 1-3? Well, in most is cases no. Yet, the system is such
that these first papers are
normally accepted and published and the last one are manuscripts
in the compilation. Shouldn't it be the other way around, and then the
last two - by the skilled
PhD-candidate, finally, would be published?
You notice that publication activity is "too wild" from reading
the texts, where people quote you. Too often the references are out of
context. The people
quoting you haven't read (or grasped) the paper, just the title
or keyword list. Or, they repeat a false quotation by a previous paper.
The quality of research articles has not
gone up I claim, together with the number of papers. And if you
make a larger article, that takes a deeper look and considers multiple
effects at the same time,
you don't get any readers. It is like the yellow press (tabloids),
you can only have 1 or 2 things maximum to bother the reader. This
is scary, because we need not
only those who are highly specialized, but also those who can
bridge the gaps. But in the end this is not so alarming, since people
in Academia mostly stay
there and are therefore harmless.
"Fake journals" is one form of obscurities around academic publishing
as was the cat-mouse -game around the leat publishable unit of information
that adds your
merits as much as any larger study would. So, what does this
do to the moral of people pursuing academic jobs? They adapt to the
game.
As a student I recall hating when we were given scientific papers
to read. They were not comprehensible to a young student. And I was reluctant
to read them still when
doing the MSc thesis. And I have this feeling that practical
experts in my field are uncomfortable reading them. So much to read
for so little. This explains why
seminars are popular - you have the academic people telling you
in 45 mins what they (and people in their field) did in the last two years.
That is efficient. And explains
why you never get feedback for the research papers, but people
might approach you after reading your web site, blog text, seminar
talk, etc.
But, it is cool to author, and co-author (on any grounds?). You
are more merited.
March 10 Academia - The place to work when wanna
avoid harrashment and bullying?
The headlines of Finnish media have in the recent
days focused on harrasment, at schools and other workplaces. Prof. emeritus
Kari Uusikylä
writes in
Opettaja and tells his views of the academia as a workplace.
Constant evaluations, fierce competition around the few vacancies, undermining
of teaching
skills, staring at the length of the publication list as a merit
etc are his topics. My own experience was, in the last almost 20 years,
that the academia is really not the
fortress of fairness and equity, unselfish work-fellowships or
any of that sort. The rule is, confirmed by experience, that it is easiest
to collaborate with someone abroad,
next comes your domestic colleageus in other disciplines (multidisciplinary
stuff is valued), in other institutes. Last on the line are the people
who
you compete with domestically - at your institute. And the sad
thing is that the (us) academic-tigers not only harness them(our)selves
to the extremely selfish activity, MSc or
PhD-students can be a part of that game too. Too often have I
seen "too active or gifted people" leaving the academia, or changing
institutes, because their
superiority makes the masses or bosses look grey. Often have
I seen critical evaluations or feedback being "archived" - what is not
visible, does not exist.
I have the numbers and the slides - be silent.
So what is it so nice about the academia that makes people wanna
stay there? In forestry, the tradition in Finland until mid 1990s was
that very few would
do a PhD. It was utterly difficult to enter forestry at UH (and
UEF), only medicine was more popular among university studies. My professor
Matti Leikola,
could still, in 1990-91, list all the, maybe 40, dissertations
in forestry since the beginning in the early 1900s. There are now some
140 after 2005!.
I don't remember much about the basics forest entomology, but
one thing I recall was when prof. Matti Nuorteva told us that he does
not expect any students to major
entomology in the coming years, because the (domestic) need for
experts was filled by a few graduates a few years back. You don't hear
such talk anymore.
Now, teachers of all disciplines fight over students and, basically
do not seem to care about the job opportunities in forestry, because
forestry is not the denominator
anymore. The job opportunities were always given to us as the
explanation as to why there was fluctuation - in the past - in the frequency
of PhD-students in forestry.
My own cohort started at UH in 1988. And we would graduate 1992-1994,
in the midst of terrible financial crisis and unemployment in Finland.
Thus, the number of
pHDs from my course (of 60) is 12+. Is academia only the second
choice, a backup? Well the job opportinities in forestry (in Finland)
haven't much improved since
1994, and we have seen an explosion in the number of pHDs. And
this does not only apply to foresty but to many other branches as well.
Forestry studies still make a great 'holistic'
degree combining applied biology, engineering, math, economics
etc. But, in this world of specilization, do we teachers still perceive
its benefits anymore? Is looking
back in time also stupid?
It is a fierce competition. The teacher-researcher-admin jobs
at universities are stressful if you wish to maintain high moral, good
quality of teaching, keep your research
up-to-date, your PhD-students in funds, and the admin sorted
out. It is a fierce competition for these, not-so-nice-anymore jobs.
And the competition for research
money? There are 1400 new PhDs per year, and the post-doc money
is scarce. In the last 15 years that I've sent in applications, I've
seen the ratio go from 1:3 to 1:15
or worse. The ratio of people getting and applying funding. The
fierce competition has created a pool of young people in non-tenure jobs,
willing to grap any offer.
This is great for example for those responsible for teaching
activities - this pool of people is at your service, YES SIR! I did
19 years of temping myself... What an idiot,
someone might say. And sometimes it feels that too.
But, every year there are new students, motivated to learn. That's
perhaps the fuel that keeps the academia going. Their belief in brighter
future.
Enough of pointing out the defects in the system. What should
be done? In the whole of Europe the unemployment rates of young people
have gone up. So finding
jobs is not just a problem in Finland. How to get jobs for the
biologists, foresters, software engineers etc? It definately
would be nicer to teach at the university
knowing that there are jobs out there. The news about young people
being uneasy with the situation do not struck me. But I don't know
what to do either. We should
ask the economists perhaps, the Nobelists.
March 9 Rewriting the
anisotropy paper
Before tackling the R-simulator, the manuscript
needed to be re-written, also to get familiar again with the topic.
LM supplied
me with interesting reading, so certainly there's plenty to do
next week, before the tuning of the simulator's reflectance calibration
part starts.
AH ctnd with the simulator. Check the possibilities for the Metla's
LIGNUM model, at phase-II, when looking at tree crowns (or even for
pine saplings?). AH should check the current literature in 3D
machine vision & computer graphics for purchases. The JavaGL visualizations
were a good add-on - it is important to visually verify that
the simulated vegetation structures look as they should. The filling
of the
footprint with random rays vs. systematic array of rays (and
weighting those), needs to be checked.
The Hyytiälä UAV data from August 2010 was made available,
and should be stored as a part of the Hyytiälä RS experiment.
At UEF,
they will use it and the seedling stand measurements of
2010.
March 6 Learning R-language
The statistical R-language has some python-flavor,
it seems. Trying now to learn the syntax to be able to continue work
done with
the BRDF modeling. Classes, attributes, functions, scripts, sequences,
control flow, lots of parameters to control graphics etc.
But the best part is that R is free and maintained by experts.
Documentation is ample in the web, too.
March 5 LiDAR simulator
to real use?
The response of LiDAR in short (or tall) vegetation
might be possible to recover thru realistic enough simulation. That
would call for
a simulator that sends out photons filling the footprint realistically.
We have accompilshed this in the simulator for the python programming
course.
This way the un-even energy distribution across the footprint
is handled.
Now the vegation presents a case of complex structure. How to
describe 1.4-m high Epilobium or 1-m high Calamagrostis growth?
We need to simplify things. But, ultimately, if it will be possible
to present the "extreme cases" of vegetation characteristics with the
simulator,
that might enable us to say something (general) about the feasibility
of LiDAR-based target classification. For example, it would be possible
to derive the effects of footprint size, quantazation pace (1
nsecs vs. 10 psecs), SNR of the device, etc. But the simulator is only
as good as we can make it.
A footprint 40 cm in diameter can easily illuminate hundreds
of leaves of short vegetation (grasses, shrubs, seedlings). If there
are
1000 rays making a pulse, we easily have 1E6 intersection computations
awaiting. This calls for good implementation.
Describing leaves and vegetation structures will be the bottleneck
(difficult to implement) , especially as the pulse is made of a
photon-bundle (slowing computations).
March 4, Vargspårvandrare bad i skolsken.
The Sipoo's skiing event was organized again this
year - now third year in a row there was enough snow. -4C..-0C and clear
skies.
It's a nice event in a wonderful landscape. In Hindsby valleys
it feels like in Ruhpolding (Bayern) almost. In all, Sipoo countryside
has
not been spoiled with wild construction, as in many places.
Speaking of Ruhpolding. Here's Viljo's view of our Biathlon sportsmen.
March 3, What is the reward in teaching? Tuition fees - please!
To begin with, there were 33 enrolled to
the programming course. A total of 26 showed up on the first lectures.
Seven weeks later,
21 took the exam. A day later, it turned out that 5 had grasped
the essential basic contents, 3 did reasonably and 13 students
either got the lowest grade or failed in the exam. Five
learned the stuff that everyone was meant to get, pathetic.
A look at the mirror: Was I too fast? My English? Were my slides
bad? Did I propose wrong reading? Were the exercises and homework just
waste? Did I not highlight the need for independent work and
studying, with the book?
The book was the first obstacle that I hit. I could not get but
a few students to buy the fantastic, low-prized 40$ book by Zelle. So,
I could not take any homework from there, or I could not direct
people to page # for an explanation. Students just refused to buy it,
and the UH admin said I can't force anyone to buy a book...Great!
The homework I set to be more demanding than the Qs would be
in the exam. And I gave my solutions as hints - but that embeds
a danger - the responsibility of learning is with the student.
I made it clear.
My conclusion is that teachers should not get too much involved
in their teaching, because ultimately, the students won't, not all
of them anyway. You look at the audience, ask them stuff, and
find out that many of them show up to browse the web or facebook
while you lecture. Puzzles me why they don't do it at home, or
why are they taking the university studies. Everything in education
is for free in
Finland, and that shows. Some tuition, even smaller, might make
the students (and us teachers) not to take everything for granted.
Just my few cents.
March 1, Teaching period finally ending
The pseudofinal reporting of the 2009-2011 UH-project
awaited. This 3-yr project funding by the University is something I
recommend to anyone.
You receive funding, annually, and that sum of money is not charged
any additional overheads. You just have to keep watch that the salary
costs
are charged as they should (given the admin jungle of different
budget types for projects). And, moreover, you can split the usage of
the funding
over time and across cost domains. It only calls for your own
decision. That is really flexible. I realize now that I was very fortunate
in being granted
this particular funding by the University. Concurrently, Suomen
Luonnonvarain Tutkimussäätiö (2008-2010) funded our efforts
in airborne RS of forests.
Without these two, I don't know where we'd be now. My humble
thanks, now at the end.
The 52 hours of lecturing programming are now past and the exam
remains. While in Joensuu, I spoke with the students there (UEF). Their
course is
3 cp, and, according to the students, does not cover OOP or data
structures. Just the very basics, and those who need programming will
continue
on a Java-course. Our course was 5 cp (135 hours), and data structures
as well as basics of OOP were included (writing of classes with attributes
and methods). The UH course should be able to give the needed
kick-start for anyone interested in applying programming. Whether I had
the right recipe
or not, remains to be seen. I increased the homework to be sure
that students use that 80 h of independent work for learning. The role
of the project
must be thought over in the future - anyhow - without it, the
course will not have any "forest-flavor", and thus, why would we give
it at Dept of Forest
Sciences?
-------------- BRDF-differences
to support species identification in passive optical RS? --------------------
While teaching, other stuff has been piling up. The BRDF-simultion
research needs some kicking now. We are trying to investigate, by means
of
rather complex Monte Carlo simulation, what are the possibilities
to improve image-based tree species classifiction, by using multiangular
image
data and knowledge of the per-species BRDF of trees.
The sensor has Nbands. There are Mspecies. This gives rise to N×M BRDF-models.
For simplicity, assume that they are dependent only on two variables
that describe the view-illumination: reflectance = f (x,y,band,species). The model tells the anistropy of reflectance
for a species and band, in geometry (x,y).
If you look at the BRDF
page in wikipedia, it is a 4-dimensional function, and considers
just direct illumination. And the notion of surface normal is important.
When viewing vegetation close enough, but not too close, we cannot
define a surface. A pixel with its iFOV
captures photons from a area or volume. And
the illumination depends on many things. Aerial photography is
always carried out on sunny cloud-free days. As the Sun climbs up, the
irradiance incident
increases. The atmosphere scatters light to the target as well,
and to be quite exact, its angular variation (across the specra used
by the sensor), should be
known. An area (volume) that is visible to a pixel is not entirely
lit by the direct light source (Sun, 0.5 degrees), in penumbra, or
in shadow. The shadow is
defined by the target casting it, its transmittance, roughly
speaking. The canopy has gaps thru which direct or diffuse light can
reach the forest floor, even.
Now, we may say that there are tree crowns in direct sunlight,
or in shade (illuminated by diffuse light). This is vague talk. If crowns
had flat tops with planar
leaves having their normals pointing yp, we could say that these
are in direct light, especially if trees would not exhibit height variation.
A branch down the crown,
may have be in direct Sun light, but most likely it does not
see the entire 2×PI [sr] hemisphere. It remains occluded by nearby
trees and the tree itself.
Reflectance is a surface property, it roughly tells the ratio
between incoming and reflected energy, 0-1. If you look at the canopy
using a IFOV that captures a whole
stand, say 100 x 100 m in size, it is possible to define the
energy incident (and its angular distribute across the hemisphere) on
that hectare quite accurately,
given proper solar data and status of the atmosphere. That's
when reflectance again is a meaningful concept, a hectare of forest
seen from a distance (space)
appears as a 'surface'. But, when attempting detection and classification
of single trees, we really have trouble using the concept of reflectance
(or BRF, BRDF, HRDF etc.).
That owes to the complexity of the targets, self-shading, shadow-casting
("fractal shoots"), mutual shading and multiple scattering. You don't
know the illumination
incident, because it varies a lot at large scale (shading by
trees or the tree itself) and at small scale owing to shadow-casting,
as the viewing is away from the hot-spot
geometry. So, ultimately, a pixel sees scatterers of some reflectance
(e.g. needles and twigs), in different illumination, and background,
usually darker.
Recently, reflectance calibration of at-sensor-radiance data
has become possible for airborne photogrammetric imagers. This is old
news concerning multispecral
spaceborne sensors and hyperspectral airborne sensors. A photogrammetric
sensor is mainly characterized by the geometric properties - it is
guaranteed to
provide stable geometric accuracy, sub-pixel, preferably. Measuring
at-sensor radiance across some band (spectral responce
function), accurately, is challenging.
If the radiance field in front of the lens is flat, i.e. each
pixel should measure the same radiance, it is required that the lens effects are compensated.If
there is a shutter,
it is required that opens similarly every time, at each exposure,
for all aperture stops. Consider a multi-head camera ("PAN-sharpening
device") - there can be
eight shutters to control. A CCD is thermally sensitive device.
They have dark-current.So, having them (the focal plane) in stabilized
temperature and pressure is
not a bad idea, for accurate photon counting. So assume that
we know the spectral sensitivity of a sensor, and it is similar across
the focal plane, it is possible
calibrate the outputs from the CCD array to at-sensor radiance
data. If this is achieved at 1% accuracy (W·sr−1·m−2·Hz−1), it opens possibilities for
reflectance calibration.
Now, in the perfect case the at-target radiance is known. How
is that possible? Well, in laboratory you can have a calibrated light
source, control over multiple scattering,
very short path between the light source and the target, as well
as the target and the camera. The effects by the medium are neglible,
and there is just direct illumination
in a narrow divergence angle. The Sun seems like a stable light
source, so that is not a problem. On top of the atmosphere (ionosphere,
even), the solar radioation is
quite stable, across the optical spectrum. Now, the photons need
to cross the atmosphere to hit a tree. Unfortunately, the atmosphere
is not stable. Gases and particles
influence the photons trying to make it to the ground, and it
is a wavelength dependent process. The fine art of atmospheric modeling
tries to solve these matters.
It seems rather obvious that the accuracies here are limited.
The incident light (at-target radiance), direct and diffuse, will be
known at limited accuracy. And the
path from the target to the sensor needs too be modeled as well,
to see what happens to the photons that reflected, and to collect photons
reflecting from the atmosphere.
Doing that, it is possible, to bring the sensor down, right on
top of the target, and then compute the reflectance as the ratio of
incoming vs. outgoing. It seems obvious
that the bottleneck here is the atmospheric modeling.
Examples on how the
solar irradiance is changed from top of the atmosphere, to the ground
(http://www.greenrhinoenergy.com).Notice how there are
sections, that are entirely
absorbed (1400 nm), and areas, where the attenuation is minor
(1000-1150 nm, 1600-1900 nm).
What other chances are there? The first thing that comes to the
mind of a photogrammetrist are signals. Not signals in some exact XYZ
location, but signals with known
reflectance and reflectance anisotropy, across some spectrum.
It would be nice to have for example 10 x 10 m man-made, flat targets
across the country. We have
experience of FGI's 4 x 4 m tarps. You bring them to the area,
but only to one location, because we have just one set, from 5% to
50% nominal reflectance. If the area
is large, it might take 30 minutes to fly it, but the tarps show
maybe in just 5 images on one flight line. Five observations along
a line in the 2D view-illumination space
(view zenith x solar principal zenith). And five observations
with maybe 20 seconds, not representing the whole 1800 secs of photography.
In 30 minutes, the sun moves
7.5 degrees across the horizon, and some degrees vertically.
The atmosphere may also change a bit during 30 minutes (building of
cumulonimbus). So the representativeness
of a set of reflectance calibration tarps (signals) is questionable.
How to get more good signals?
In Finland there are a lot of trees, and lakes. Could they be
used for reflectance calibration of image data? Lakes are a bit tricky,
because of the mirror reflections. And they
are dark, sometimes there is flora (sub-surface even).Tree crowns
remain quite stable from September to May. Fresh shoots and leaves appear
in late May. This means that
it may be difficult to use trees as absolute calibration targets.
However, if we would know the "BRDF" of trees, it might be possible to
roughly correct the at-sensor-radiance
images to "reflectance" data that is homogenous across the photographed
area. Moreover, multiangular observations reveal the anisotropy of
the target. And if it is, in this case
tree species-specific, it might open up possibilities for improved
target classification.
BRDF (reflectance anisotropy) is in every remote sensing textbook.
We thought that BRDF would be used, in the context of multiangular remote
sensing - to support target
classification. Despite ample existence in textbooks, to our
surprise, we haven't found studies - yet?
Now, consider that we'd know the "BRDFs" of classes A, B and
C. And we would have reflectance data, from several bands. We fly over
an area consisting of A, B and C
targets, and each target shows in 6 views. Now, it becomes possible
to compute a feature vector that describes the "match of BRDFs". Max(match,
{A,B,C}) results in
classification, that is the best match of the observed anisotropy,
and one of the BRDFs for A,B, or C, across the bands. What are the
requirements here, to make this
to work? If the reflectance (image) data have offsets, the offsets
will show for all classes, and the brightest or darkets target class
may end up being heavily favored
by the classifier. So, reflectance calibration errors (between
images, for the whole set of images) need to be within some limits.
Also, we know that trees exhibit large
within-species variation. TO BE CNTD... This variation represents
noise in the classification of A, B and C.
If a large set of trees of species A, B, and C are viewed from
many directions, in many images, we observe:
reflectance
= reflectancespecies +
BRDFEffect
+ TreeEffect + ImageEffect
The basic (mean) reflectance of a species is the
average for example at nadir or in the hot spot. The reflectance spectra
of trees varies between trees, but the deviations
across the spectrum are always highly correlated. Bright in BLUE
predicts bright in NIR. This is the tree-effect, a correlation structure
making trees to appear darker or brighter,
in any viewing direction. The BRDF-effect tells how the reflectance
varies with the view-illumination geometry. This effect most likely is
correlated with the Tree-effect. The last
term is the error arising from reflectance calibration. It is
not a constant, but probably dependent on the view-illuimnation geometry
as well (the atmospheric effects are dependent
on them). Also, the effects are probably correlated between exposures,
as the assumptions on the atomspheric status are violated similarly
for exposures taken a few
seconds apart.
If the varianve-covariance structure of the outline model is
known, without the last term, we can play with the last term, and see
how it affects the results of using the
multiangular data for the classification of A, B and C. This
calls for real data, and this is when we again resort to the 2008 ADS40
campaign in Hyytiälä.
Feb 29, "Oh snow!"
The hare had quickly stopped
last nite in the garden, at a spot where we measured the snow depth. 32 inches!
(80 cm).
This is the record since we moved here in October 2000. Oh snow!
Feb 28 Surfing on the waves of LiDAR and digital photogrammetry
- is the beach empty? (A little longer line of thoughts)
Is the hype on
LIDAR now history?
The hype on environmental LiDAR has had its peak. We've seen these
hypes before as well. Recall how the satellite images
were expected to turn forest inventory upside down in the 1980s?
Aerial orthoimages were the 'promised solution' in the 1990s.
In applied math, some might remember how multiple regression
and other techniques of multivariate statistics made many to
model and explain, everything, far beyond the limits of these
methods. This was in the 1970s and 1980s, when the researchers
got access to mainframe machines, and the statistical software
on them. There's always some hype - it is needed to get us going?
How do I define hype? Well, as I've experienced it, it could
be described as an unconscious fraud. Most research is about banging
your head on something and not getting really anywhere (or repeating
work already done a dozen times, and covering for the fact).
So, if someone has a 'promising alternative' - a way out of the
misery - you easily grip it. It bears resemblance to getting a new
domestic device - it's cool, and there is absolutely no reason
to read the manual. Namely, reading the manual might reveal that
the machine still won't do the trick you would like it to do.
You are in the hype set fire by the salesman.
Was LiDAR just
hype for foresters?
But LiDAR wasn't just hype. For decennia forest inventory experts
had been projected to the xy-plane. This must have felt like a handicap.
Trees stick out from the ground and are not captives of any 2-dimensional
thinking. I had never seen a forester climb and observe the tree
apexes, but then, all of a sudden, 10 years ago, we had a device
for probing the apexes, the ground, and everything in between.
Some people realized the potential first. These are called pioneers
in research. They are people with vision. In the case of LiDAR,
they did not have to build the instruments themselves, but surely
it wasn't easy to do the first experiments. The ratio must be 1:100 -
the ratio between people supporting your ideas and vision and
those undermining them. I was one of the mildly discouraging people
still in 2003. Back then, I had put all my effort to be 3D, leaving
the xy-plane behind, not using LiDAR but image matching, and I had
to defend that option of being 3D. It takes some time to absorb
the facts.
How did it come
about?
Laser was invented already in the 1960s. Crucial to pulsed LIDAR,
is the powerful transmission of pulses that preferably remain stable
over
time (same energy, length and shape). In meteorology, terrestrial
LiDARs, looking up, were used years before topographic airborne
LiDAR appeared. And the first airborne trials were carried out
with profiling instruments, in which the laser continuously illuminated
targets right below
the aircraft that possibly swayed as it flew. The technology
had to develop towards making the pulses short in duration and powerful.
The photons reflecting back must also be detected somehow. The
photodiode had been invented in 19¤¤s(?). To enable accurate
ranging, the shape
of the returning photon surge, e.g. arrival time of the front-line,
must be captured at fast pace. As light travels 300 meters in one millisecond,
even
1MHz sampling is not quite enough for forest applications
(consider tree heights). The diode
needed to be fast and the voltage output must be
quantized at extremely high pace. Getting the orientation of
the sensor, continuously, is another story, as is the pace at with
pulses are transmitted (see below).
"More pulses
per second - I say"
Flying is expensive, a fact from photogrammetry that hasn't changed
with the emergence of LiDAR. With a proper airplane, flying costs
are at least 50€ per minute. Groundspeed of 80 m/s indicates
that 5 km are covered in one minute. A LiDAR sweeps the area below the
aircraft. Preferably there are many pulses per m2, sampling the
area efficiently - in a regular pattern. An airplane has to travel at
some
minimum airspeed to stay airborne. Thus, lowering the speed is
not an option to boost the density. The width of the area that is swept
can also be tuned smaller to increase the pulse density, but
that is expensive, as the area covered by a flown kilometer will be
lower.
The effects of occlusion will also have a larger impact the more
oblique the pulses are. The doctor orders faster pulse transmission
rates as
the cure. We have witnessed a development of making the LiDAR
sensors faster.
"LiDAR was enabled
by advances in positioning technology"
Another aspect of LiDAR is the orientation. There is no redundancy
to rely on, as with overlapping aerial images. The pulse is a vector
that starts from where the LiDAR is, to some 3D direction, and
extends as far as the range says. It is much like a total station.
But, it
is miraculous how the orientation is known for each pulse, when
there can be 300,000 of them per second. The transmitted pulse is
deviated from the sensor to some direction, usually by means
of an oscillating mirror. That involves the first rotation in space.
There has to
be a way of accurately synchronizing that angular position with
the pulse data. A terrestrial laser scanner sits on top of a tripod and
the
deviation of the pulse is by means of two spherical rotations.
But an airborne LiDAR cannot be leveled, the platform is too unstable
for the bubble.
The sensor is subject to the rotations of the aircraft as it
moves ahead. It is just amazing that by combining (smart) GPS with
(expensive)
inertial measurements, the position of the LiDAR is solved at
accuracy better than 10 cm. This is true for any time point.
But there is more. The attitude of the sensor is also known at
accuracy better than 0.005 degrees. The instruments for this
accurate on-the-fly
orientation are really expensive, and heavy, so you cannot expect
to see light-weight UAVs to carry them. And if they stop working,
the data is garbage - there is no way of figuring out the 3D
point cloud from mere distance data. The synchronization needs to be
perfect as well - between LiDAR, GPS-receiver and the inertial
unit. The LiDAR community owes a lot for the positioning people.
Well, LiDAR is a positioning tool.
More pulses per
second - eternally?
How high can you go in LiDAR pulse rate? If the sensor is consuming
1 kW at current pulse rate, doubling of the rate means that the
device will consume 2kW. I don't know how efficient the lasers
are, but if the efficiency is 80%, the power dissipated as heat will
increase from
200W to 400W. I would assume that there is a maximum that can
be cooled, and therefore you often notice that the per-pulse energy
(and peak power) is reduced as the pulse rate is increased. Again
no free lunches here, it seems as we bow towards pulse rate,
our butt shows at the photons transmitted. Another issue is the
restricted velocity of light, 300,000 km/s. From an average distance
of 3 km, the maximum rate is 150 kHz, and there still isn't but
one pulse in the air. Above that frequency, there are more that one pulse
in the air simultaneously. So, increasing the rate can
be problematic for the engineers that construct the devices. At 300
kHz, there
are 3.3 milliseconds recovery time between pulses - the receiver
must process the data in the backscattered waveform that is some
hundreds of nanoseconds long in the worst case (tall buildings,
vegetation). 3300 nanoseconds correspond to 330 cycles using a
100 MHz DSP. Everything at the receiving end must thus also be
extremely fast. A long waveform occupies also quite a lot of storage.
Quantizing with 16 bits and at a speed of 1 ns, gives 100 samples
for a 15-m deep object (omitting the width of the transmitted pulse).
That is 200 Bytes of data, and if these keep coming at the pace
of 300 kHz, it is 58 MBytes of data per second to be set aside.
It calls for a fast bus and write head, and clever engineers.
Receiving the
few photons that reflect back towards you
The photon detector in a LIDAR probably turns every photon received
into a voltage peak. That voltage, or signal, is very low and needs
to be amplified. The detector probably has some internal noise,
as low voltage peaks rise abruptly now and again. There are also photons
from other light sources, such as the Sun during daytime campaigns.
These signals are background noise, amplified also.
The real signal reflecting from the target-of-interest has to
outdo the noise. The illumination of the pulse, in terms of W/m2, decreases
with
the second power of the distance. A leaf 1 km away receives 25%
of the photons compared to a leaf 500 m away. That is 6 decibels.
The same leaf 3 km from the sensor gets only 2.8% of the energy.
That is 15 dB in illumination. But a single leaf that far may not trigger
a measurable signal, though.
A receiver that is intelligent can do better in the presence of
inherent noise. Think of it. If you know that the previous pulse produced
a
strong reflection from an elevation of 100 m (a.s.l), it is pretty
obvious that the next one ought to produce one near that elevation
as well.
So, the receiver can start to save the waveform, for example,
when the elevation of the possible target is less than 150 m. The entire
backscattered signal over a depth of 50-60 m is some 500 samples
at 1 GHz sampling rate. If we can afford to store that sample set,
e.g. at 8 bits per sample, it is possible to delve into the signal
afterwards, in post processing at the office. The other option is to
be more
cautions, not to embed a priori knowledge about the expected
range, and set a threshold for the signal. If signal is above, start
a
detection of a reflection. The threshold is a precaution, and
results in lost dBs. So, there clearly are some dBs to gain or lose both
at the transmitter and receiver side, by choosing a different
strategy. But the physics can not be deluded. Enough photons need to
illuminate the target, in order to get a measurable signal. A
low-noise receiver helps, as does high transmitting power. But, there
are
limits for eye-safety that restrict TX power, and the overall
power available (230 VAC) in an aircraft is not endless either.
What is there
to be observed about forest canopies with LiDAR that we do not perceive
yet?
"LiDAR produces a point cloud" was the perception of the community
until waveform digitizers started to appear in the receivers.
Trajectory of the LiDAR of course was there, for each dataset,
but people were not interested in it - to see from which direction,
or along which path, the pulse that produced the one or more
echoes had traversed. You could do a lot even with the point cloud.
The terrain under a canopy was out of reach with photogrammetric
data. LiDAR data gives the terrain profile even under leaf-on conditions.
And the accuracy is amazing, at the decimeter level. As foresters
had been accustomed to 2-3 m accuracy, this was a giant leap.
And you are left with a lot of points that do not represent the
terrain, but vegetation. Early pioneering researchers were clever to
show
the statistical dependencies between the traits of the vegetation
point cloud and the forest parameters. Empirical, statistical modeling
does not tell you the exact nature of the dependency, but if
it works - who cares - the forest parameters are there. But there are
no free
lunches - the simplicity of empirical approach comes with a cost.
The models linking the LiDAR data and forest features are not universal.
Changes in the data acquisition settings ruin the dependencies,
at least to some degree. There is a need to have samples to train
(calibrate) the models - and it costs money to measure trees
every time there is a campaign. Here is where research could help
still in the future. Another interesting aspect relates to the
dampening of the pulse as it traverses thru the canopy. The pulse usually
never hits anything that entirely fills the cross-section of
it. Thus, it is possible that different parts of the pulse reach different
depth
in the canopy volume. This is why there can be multiple echoes
from a single transmitted pulse. Now, it puzzles me if the width of
the pulse cross-section
and the size-distribution of intracanopy gaps or scatters have
any correlation. This would show, in a data set that combines
acquisitions using different footprint sizes. Namely, I believe
that going multi-footprint is more interesting than going multi-spectral.
Vegetation, if it is bright in one wavelength, is bright also
on another. Don't put your money on multispectral. Look at the
gaps, not
just at the scatterers. (Joensuu train Feb 28, 2012).
In Joensuu to lecture photogrammetry. Some UAV talks - see if
the UAV 2010 seedling standdata could be re-used.
Feb 22, Scanning film photos - tests began
Before the CCD -era, film was used to capture
the images. With the help of Markku in Ristiina, we try to get the
old diapositives scanned. 2400 dpi gives 4000x3000 pixels.
If its ISO 100 diapositive color film, the resolution is max
80 line pairs / m at some decent constrast ratio. But, since most pics
were taken without a stative, you do not get that. And, the
optics is there too to worsen it. 80 pixels / mm is about the
maximum thinkable scanning resolution. The one on left is from April
2002, and has 10-month young twins in it.
A bit later - reading. Pearls of the fotoarchive, from Kumpula
in 2000.
Markku has been scanning old (much older) photographs also in
the past. This one has Atte Korhonen (grandfather of mine, father of
Markku), during WWII.
He was born exactly 100 years ago in the spring in Perniö.
An inventor, artist, building master and self-made engineer. I owe him
alot for the support in electronics
& ham radio as a teenager.
It's snow all over the place. You need to spend, daily, 2-3 hours,
to keep the constructions, roads, pulkkamäki etc. free from snow.
Nearly 80 cm in places in Kulloo (Feb 22).
Feb 19 Snow
The heavy snow-work made Perttu to think of ways
of demonstrating. "This fellow stays on alert for blizzards coming our
way" - he thinks.
Some 200 m2 in need of clearing in Kulloo. That was 6tn this
time + 6tn from the roof. The appletrees also needed rescue - they were
moaded -
otherwise they will be hare's feast. The 1-m-high fences don't
give protection in 75-cm-high snowdrifts.
Feb 16 - Explaining the Simulator
The Python project's under-the-hood & code-explained
part was done half-way. (pdf).
Feb 14 - Finalizing he simulator with a viewer.
Python probably comes with many graphical modules,
but I used a simple one that comes with the Zelle Book: Python programming.
Plotting isn't very fast as can be expected for Python, but it
works. The figs have the first-return data with small dots (z is the
color theme), and two-return
pulses (echoes) were drawn with a wider pencil. The scene has
just some ground planes (some sloped), two buildings and two trees. The
taller
tree in the background has a short crown. Some hits in the trunks
too. The view rotation matrix should also be changed to be based on view
azim, elevation and
camera rotation (about the view axis). The focal length changes
in the two views as well as the view position. (python code)
Feb 10 - Pythonized week.
The homework & project are keeping
busy. The documentation of the project "LiDAR simulator" took considerable
time (pdf).The
python code is here (py).
Feb 4 - PileUP! aftermath in Helsinki.
Thu Feb 2, 2011: OH2PM & K1DG (CQ WW 160 m
trophy). K9LA, AE9YL, OH1WZ (&G3SXW) (PileUP! trophy).
Sun Feb 5: -30C in the morning. Attick ventilation
seems to work, that's what cold weather looks like. Gradient of 46C.
Feb 4 It is cool in Finland
Apart from AH's activities (Hyperspectral, Hyytiälä
tree database) not much else got progressed this week (researchwise).
The LiDAR simulator
for the Python/programming course was put further and it now
works for planar surfaces polygons (buildings, terrain) and with simple
tree objects.
The LiDAR pulse is simulated by a ray-pencil. Each ray intersects
a target (reflection) and fills a photon counter with a (time,
voltage)
observation - voltage being related to the scanning distance
(squared) and the target reflectance and incidence angle. With tree
crowns, the
reflection is a probabilistic event - the crown is traversed
along a path (ray) (discrete steps), and the radar cross-section of
possible targets is
randomly drawn. If and how to embed some structure (clumping
of scatterers in branches etc.) needs to be figured out (or we omit
it).
The photon counter -list gets to be filled with (number_of_rays)(time,
voltage) observations - these need to be processed into echoes or waveform
(convolution with
the transmitted pulse). Another set of assignments was prepared
for the course - I hope with unambiguos definitions this time.
-15..-25C here in Porvoo. The pileup crash of March 17, 2005
in Helsinki roads was repeated on Fri afternoon as busy Uusimaa folks
were driving home from work in the blizzard. It takes 7 years
to forget?
Thu evening I was at the CCF cruise gathering in Grand Marina.
There were 6 DX and 5 OH. I recall an earlier CCF event, some 15 yrs
ago. The numbers
then were 10+ / 30. I met with Carl K9LA, and he was given the
"first class contributor's" trophy for his writing achievements 2006-2011.
The Sun was quiet (link)
- did we already have the solar maximum? Why bother? Well it is dramatic
for many - solar cycles last 11 years and every ham would like to experience
a good one. Maxima of 1959 and 1980 were particularly good (I
wasn't around). Now it seems that 2013 will not be. Well there'll be
another one in 2024... If this isn't
depressive then what is? 6 m enthusiasts (50 MHz) had the last
proper F2 (layer in ionosphere) propagation in 2002, when the solar
flux (2.7GHz,
ionosing radioation, makes the ionosphere thicker) exceeded 250.
Now, with flux around 120-140, it seems that 50-MHz F2-reflective areas
will remain
above North Africa, and thus out of reach for hams in Finland.
Feb 1 LiDAR simulator
AH got the HS trees done quickly. As expected,
birch is scarse, but r = 15 m pure monospecies plots were found in over
400.
Working now on the LiDAR simulator. There is a module that produces
the EO for each pulse transmitted. The plane moves with
roll/pitch fluctuations. 3D Plane polygon - ray intersections
need testing for the special cases vertical walls (x- and y-aligned walls
being
special cases). The resulting xyz point gives the range (LiDAR
observation), stored in a photon counter. Python seems quite slow,
so we
will have to keep the PRFs down and area small.
The scene had two taller buildings
(h > 2.5 m) and a triangular sloping ground (h > 2.5). This could
imply that the point in polygons works
for
non-vertical surfaces at least.
Jan 27 Skiing is good in Uusimaa.
Aftermath of aftermath. One
more PileUP! trophy was prepared.
B) January photos: we were fortunate to
visit the Tahkovuori in Nilsiä, and the site was loaded with Russian
tourists. No
wonder they would like to swap obligatory Swedish into Russian in schools
of Eastern Finland. C).Viljo
with equipment.
D) Finally something worth watching on
the TV. E) "Duudson's visit
families" is sure hit in Kulloo.
The
Python course is taking 50-60%
of the hours + the lectures. Coming up with new homework, preparing the answers
and checking the students' e-mails
is a full-time business. The programming exercises still need
to be defined. I'm thinking of not using the same that we had last year.
One option would be
to define a simple LiDAR simulator. The scene could consist of
planar objects, xyz-points all sitting on the same 3D plane. Simple
scene with a few buildings
and surface types (grass, asphalt, gravel, roof types). Oblique
scanning would result in wall-hits. How to define the resulting echo?
Maybe the pulse should
be simulated with a large number of rays - fewer towards the
edge of the footprint. The intersection has to be defined in the coordinate
system of the xyz-polygon,
e.g. a house wall is vertical - the surface normal points to
the horizon. It would be possible to include a few trees too in the
scene. The reflectors inside crowns
need maybe to be stochastic, to allow a pulse to penetrate or
even pass thru.
AH worked on the Hyytiälä
hyperspectral 2011 campaign all week. He set KUVAMITT to work
in a 100 x 100 m grid, and the idea is to find pure samples of
pine, spruce or birch vegetation, that are seen in the HS images.
RGB orthoimage versions of the HS data showed on average 2 m XY error
on the DEM. This
limits the analyses, but is still reasonable.
Photogrammetry lectures (pdf),
labs (doc),
Excel examples & lecture in powerpoint (zip).
Included a new GPS.xlsx, which demonstrates trilateration
using 8 'satellites'. The UEF package still needs refining the
Qs and tearcher's answers.
Jan 23, Elections
AH got the Rautawaara exercise
done and reported. It took a week to notice that the field plots were
too small and noise prevailed in the data. Now started the
Hyytiälä 2011 exercise, finally. We are looking at
orthophotos made of the hyperspectral imaging, and will try to locate
(assisted by a systematic grid), samples
for the UEF team to analyze. First step is to assess the geometry
of the orthos.
Feb 28 lectures & labs in photogrammetry were tuned (from
2010) as well, Theodolite.xls has now the cases for two and three theodolite
positions.
Regression.xls was added to show how OLS regression and non-linear
regression differ (x = (ATA)-1ATy).
Donno if it made any sense to try teaching the "under the hood"
part at the start of the programming course. In the short time you only
get a glimpse, and
maybe that only confused people. Well, it remains to be seen.
Behind schedule we are, that's for sure. Today I tried lecturing without
a proper preparation,
and it showed, at least it felt like that. It makes a difference
to teach in a foreign language since when you are tired, the words escape
you, and things become
awkward. You don't have to prepare a lot, if you really master
what you are teaching - in my case I dont master Python or the programming
vocabulary so well
that I'd be comfortable.
The elections were on Sunday and my candidate did reasonably.
Finns (me included) would like to have a stronger president but our MPs
disagree. Our coffee room
in Viikki was devoted to the elections - it is seldom that you
hear a Finn younger than 40 to talk politics - it has been exceptional
during these elections. Maybe the
recent European economic turmoil has ingited it?
Jan 20, Teaching programming
80% of the time went still this week to prepare
the start for the course, 8 hours of lecturing per week. This
time, we did not touch IDLE and Python but after
6 hours of lecturing about 0's and 1's. Although Python hides
the bits and dynamically changes the type of an identifier, there are
the basic types, and the precision
of floats is limited in Python too (standard library). XVI32
hex editor worked fine, but I'm not sure if all students grasped what
we did with it, when we looked and
mangled the bits of the very simple files.
IPO. input, process, output. We tried to learn
this using paper slips as data records to be ordered according to the
number (age). The computer was simple. It can
read slips and assign them to two memory locations (variables,
identifiers), called A and B, but only if the memory location is free.
Contents of A and B can be compared,
and we get a Boolean saying if A is greater than B. Then there
is a command to output A or B. The outputted list of records is then
subject to another input if needed for the sorting.
Jan 13, Slow start for the new year continues
The week was spent preparing for the programming
course & helping TA with ESPA/multitemporal AT. Espa did not
support DSO, so we needed to establish
geometry transformations back and forth. The old xls-files implementing
solutions from R to (omega,phi,kappa) systems came handy. TA was taught
to use the ESPA-obs
to metric-right-handed-obs and WLS-AT software. This is taking
far more time than initially thought.
Python is funny - it is a language I though having learned last
year during the 7-week course. But, just like with any skill / language
- if left unnourished, it dies and you
forget it. It's a high-level language - so the options (things
to remember) are almost endless. And for an old fart like me - who comes
from the imperative era, OOP and
Python represent certain mental friction.
Python syntax recognizes some special characters, and if you
are the teacher - you ought to have some idea about what to call them.
In English.
To try internet
programming, as a possible option in the exercises, I did a small application
that checks if it is cloudy or if the dark sky
is visible, and if the magnetometers imply Aurora. Python code example: "Might Aurora Borealis
be visible here, now?" (html).
Must be able to run in parallel the two projects from on next
week. Research has now been idling for too long. The darn OH1-hassle.
Must return
to the stuff the financers pay me.
Two weeks from release the Last Blast had over 2500 downloads
from unique IPs. Feels still a bit sad? (odd) to leave it and a-25-yr-old
hobby behind.
Amateur radio is a hobby that goes deep in one's bones, and among
all possible sub-hobbies therein, contesting seems to create serious
dependencies (addictive that is).
Jan 5, Slowly back to business
Preparing for the Basic's of Programming -course.
This year I wish that we have a peak under the hood, before we commence.
The dynamic
typing in Python hides everything as does the automatic carbage
collection. Yet, static typing exists in many languages.
Free XVI32 Hex-editor was found to work in the XP-machines. We
will play with that this year. I heard that in the future, with Win-7,
the plan in UH is that no exe's
can be run from other folders than those where you cannot copy
them. I just wonder what can be done with the PCs? Safety first.
I hope the students like to think in base-2 and base-16 numbers.
The SAC trophies were put to boxes in Espoo/Suomenoja, the mailing
awaits. ARRL contest Update acknowledged PileUP!.
Only ND is missing from WAS now. DXCC is almost there too. 2100
different IPs.
RSE proofs came and were returned. Hyytiälä hyperspectral
images (2011) - nothing still.
The Riegl waveforms from November - it seems that there is the
LAS 1.3 option anyway - must study further - and sit down properly.
______________________________________________________________________________________________________________
Dec 29, PileUP! endeavours
are over
Xmas-holidays were spent 50% on the last issue of PU! by the
2006-2011 editorial team. Last minute addition was Viljo, as we realized
sitting by the kitchen
table, that Viljo hasn't appeared yet on the pages of PU! (see
photo below). Altogether 32 people joined the PU! talkoo-fun this time.
The 20 issues of 2006-2011
comprised a total of 860 pages. That's a lot of talkoo hours.
Perttu also contributed to this PU! by drawing OH1WZ ham shack events
(middle).
Dr. Crofthill reported about a new scientific recovery: a new
molecule called contesterol (right). PileUP! was an extremely nice, yet
laborious experience, and now
it is time for something else. Radio contesting has started to
fade already, along with the amateur radio hobby, which hasn't really
received recruitment in the
last 15 years (internet era). A new state of equilibrium awaits
amateur radio - now is the time to buy gear - when they still manufacture
it....
Pre XMAS stuff. Perttu's angry bird & Arttu's comics for
PileUP! to go with the JW5E story from Svalbard.
R2 (RSE) was accepted, and now new projects can begin.
Dec 17 Commencing detonation
preparations of The Last Blast
To keep the intensive period as short as possible,
the actual compiling of the Last Blast, PileUP! 15(5) wasn't started
but on Sunday, 2.5 weeks before the
DL. By Saturday, Dec 17, there are over 10 contributors, 36 pages
and some 12 more still pending by contributors in OH (Finland). The editor-trio
are res-
ponsible for 22/36 pages now - shows again that its not sustainable
to run PileUP! this way.
We editors are not that humble that the front-cover would not
have us portrayed. Thanks and CUL says the text.
Esa put his best to the humor section already in September, and
to complete that, I wrote the first obituary to PU! :).
The R2 was worked on a little and returned to RSE. Admin stuff
now towards the end of the year. Project accounts
seem to be in some order- although the cost accounting concerning
salaries is strange still. The 83% efficient time
and 100% gross salaries seem to shuffle even the experts - well
accounting IT system is programmed some logic, that
is 'the truth', eh? KKM troubles again... Anyway, we seem to
be able to invest into photography eq. to support the
field photogrammetry, which we've until now done with personal/private
eq.
It seems that there is some light at the end of the MARV-teaching
tunnel, as this week, after some dramatic episodes,
the matter was taken into consideration. Still, not early enough,
I need to teach the Python course in Jan-Mar, which is
ok, now that the lectures were moved from early mornings to evenings,
17-19 and 16-20.
Next week starts the R-code / simulator tuning.
Dec 10 R training at home and in Joensuu
In Finnish the letter R is pronounced (technically) in similar
way to D, which requires less tongue acrobatics. So you practice R by
substituting Rs with Ds.
"Letted D is pdonounced as D". The letter R has of course many
meanings. Those familiar with the Morse Code (telegraphy) know that
R stands for "Receiced".
In statistical computing R stands for a high-level language (free),
which has a large community contributing development & code. Thus
far I've managed to avoid R
by doing simple stats with SAS & Matlab, which both are commercial
software packages. In Joensuu, I learned this week that R is comprehensible,
and that it
might be possible to learn its syntax as well. Dec-January will
show how it goes.
We finally began refining the R2 for RSE after 6 weeks of idling.
Time helps. At the same thime LM had worked on the PFG manu (much alone?).
Still no news at UH Dept For. Sc. concerning the ban to hire
replacements for two lecturers in our discipline, lecturers who managed
to get external funding
and thus freed their jobs for younger people. But I've learned
that this kind of saving-strategy is not that uncommon - savings are
first taken from
where they are technically feasible (e.g. from where extra resources
have been obtained) - the consequences or ethical Qs are secondary.
If you
have fire burning at your feet - you tend to put them out first
- and may not see the monsters or possibilities in the distance.
We had the traditional Xmas seminar in Joensuu Thu-Fri. Many
good talks - someone in the audience even hinted that the quality of
the
seminar surpassed by far the Tasmania congress (Silvilaser).
It just shows that 4-400 km travelling can be more productive than 9000-12000
km.
And, I still am of the opinion that forest science is different
from medical science - those with cure to human diseases should travel
more
than us. AH gave the talk on LiDAR transmission losses and I
gave an update on Hyytiälä experiment (pdf).
The LiDAR + ADS40 research material was passed on to us: Data
(LiDAR and image features) + R-codes (classification scenarios, mixed-effects
modeling,
simulator). The work still needs to be finished by us in Helsinki,
while the Joensuu end will have new things on the agenda.
Ruben Valbuena introduced Lorenz curves as measures of stand
structure. The audience was paying attention. R-training at Borealis.
Dec 6 Aftermath
This year Dec 6 felt particularly important in the midst of the
pan-european hastle, which simply seems incomprehensible to a common
man.
Dec 6 Independence
day
The Kulloo school celebrated Dec 6 on Monday -
and have times changed?
The concluding presentation by the teachers was about the history
of Finns - starting from the Ice Age, and the last two slides
had Risto Ryti and Lauri Törni
pictured. This would have been impossible back in the
late 70's early 80's when (we actually didn't) we had Dec 6 celebrations.
A & P with Pattu sang 'Maamme' in front of the audience. Well they
did
not inherit my singing skills or courage...
Sun was about to rise (at 09:11) when the flague was hoisted
at 08:40. A & P singing. Dec 6 party ended in Letkajenkka. Viljo
in blue shirt.
To celebrate the occasion (Dec 6) we manufactured
another wooden object (m/39) in the basement. A few details & paint
are still missing but boys don't seem to mind.
Nov 30, 2011 - More daylight, please!
This is the time of the year at 60N that could
easily be given away. The Sun is 'up' 6 hours per day, and when its overcast,
you barely
recognize the 'daylight' hours. For me this means that performance
is down at least 50%. Pink glasses are lost, and thus the surrounding
world glows in tones of grey that match the glow of your thougths.
The literature mentions ways to deal with the sombre melancholy:
chocolate, vitamin D, fish oil, excessive sleep, alcohol and/or
exercise. I also seem to use some of those.
Peer reviews (3) were finalized, and it seems that you really
cannot do one in less than 8 hours. The DEC 2 talk was also prepared
(pdf).
The semi-automated program (KUVAMITT-version) was done for AH,
to assist him in updating the status and treetop XYZ-position
of the 25000+ trees in Hyytiälä. The oldest field reference
for the trees in from 2006, and especially the snowbreaks of 2009/2010
need
attention.
Ja miksi
LASERKEILAUS on niin hyvä 3D työkalu?
'Vanha' sähkölinjaesimerkki sai rinnalleen uuden valokuvan.
Radiometrian ja geometrian yhteispeliä laserkeilauksessa.
CQ WW
CW was on the weekend. It appears that radiocontesting truly has changed.
Morse code is no longer decoded by human brain, but 'inventions'
such as the Skimmer and SDR, RBN are there to 'help the lazy'.
It seems to be the right time to leave the party when it is still 'fun'.
Nov 22 It all ends - said the ram to be decapitated.
This autumn we spent lots of time
in the basement building stuff. Highlight was the machine gun seen here. One
story came nearly to the end on Saturday, 19th. PileUP! Hall of Fame.
Coming to work to Viikki
(at the Dept of FS, Faculty of Agr. For.) is somewhat uappealing these
days. If it wasn't for the science & students, I don't know where
we'd be.
But they are arranging XMAS parties - that'll sort it out for
sure. Ok, enough for complaining. Let there be more light - soon.
Sangram Ganguly (NASA) lectured in Viikki, and provided nice
insights into the inherent problems of spaceborne optical RS and monitoring.
He even was at the same time
douptful and excited about LAI products. But, the Q remains,
what are we observing when observing change, when variables are correlated
and atmoshere poorly
under control. A map may look nice - even a global map.
Nov 18 Another misc
stuff - week
The new L-vertical on 160M radiated
energy as far as KH2 (Guam) this week, but the old TS-850SAT seems
to be
leaking RF a bit? on both sides
of the freq, as seen here in SDR-data rcvd @ OH2-land. The screen capture
has "OH1WZ", 1830 kHz. Owing
to the digital
technology
you can share even this kind of ham radio fun. Preparations for Saturday
PileUP! -SAC gathering on the finish line - a special issue was even
prepared,
PileUP! 15(4½).
And the SAC-Ilmaveivi success story was celebrated even at YLE on Wednesday
(16th) with some 25 enthousiasts. October-November
are
the radio months, with SAC & CQ WW contests.
'
@ YLE OH6KZP shed light on how
he pulled thru the new concept for SAC, to be followed by the new SAC
CC.
And the audience was hearing, maybe even listening.
This week we tested a joint concept for student recruitment in
collaboration with agricultural sciences as high school students visited
the campus.
It may not be so evident to young people that such thing as multidisciplinary
Forest science really exists (vs. felling down trees). A good idea
by the student advisors.
Nov 11 Overflow
- Systems running on low performance level
"AAAA" is a simple and efficient stress management
scheme, but so easily forgotten. If you forget or just no longer cannot
do the avoiding and altering,
you're left with adaptation and acceptance. It is good to keep
a bit of buffer in one's stressor bucket, just in case unavoidable
or unchangable stuff appears
in the immediate horizon... And skills disappear - at times practise
the two letter simple answer, "NO".
In the midst of it all, this week we fought the August 2011 FWF-data,
to get the data ready for analyses. Heads were banged at the 2 GB wall
too. Admin
KKM hastle caused hair-loss, as well as admin meetings. With
the KKM-budgeting, the old rule of saying salary costs are gross salary
times 1.3, is now
"salary costs are efficient salary (0.83 x gross salary) times
1.53". This "efficient salary" or whatever it is, is there only the
torment us, I say.
Do you hire someone by promising his/her that "83% of your gross salary will
be #### $?". I heard that the KKM is now receiving critics also from
"higher
hierarchial levels", with the immeadiate response being that
colleagues at the grass root level, also, now , start to "grasp the
issues" - isn't that funny?
The invisible admin, please.
And, it came with a 2-year-phase-shift to UH, namely last week
I heard (from influencial admin stuff, unofficial of course) that research
projects exist for the
purpose of collecting overheads. I had heard the same sentence
at UEF two years ago, but I never thought it would come to UH.
In all, elementary knowledge seems to be lost - and by making
the admin complex - this is the ultimate outcome.
FWF flying was tried at Hyytiälä, but the foggy weather
prevented it, WED-THU. We managed to put forward (a bit) the project
for AH, ordered some computers and
spare parts. A talk on Dec 2 was worked on a bit, as well as
UTM-support for kuvamitt (DEM/ORTHO) for the Hyperspectral imaging
control. A few lectures were
worked on as well as demoes on simple polynom algebra...
Nov 2 GOOD and BAD news
PileUP! SAC trophy construction is under way (with A, P &
V). The neighbor owns a "yläjyrsin", using which the edges were
tuned to nice shape. Warm autumn weather (+10C) continued
and the boys enjoyed outdoor activities. Kuusitie tournament
(ref. Nov 2009) was organized as it is every year. This year it was
table football.
Work on the waveform data got started and we hit some obstacles
in the first curve. The discrete-return data was assosiated with wrong
waveforms. On the map it showed
as offsets and folding. Luckily, it seems that the assosciation
was wrong due to a processing software issue, and the sensor had worked
without problem. So it is not so
bad news in the end. November is the month of public presentations
& lecturing, started preparing slides.
The ADS40 simulator is taking steps forward, and serious debugging
has commenced. Leaf-off FWF LiDAR campaign awaits still. EJFR had accepted
our paper sheding
light on change detection in bi-temporal LiDAR data. Our faculty
has set a ban on hiring anyone. So, it seems that I will not be replaced
by anyone (as a lecturer). I wonder
how the teaching will survive this. I suspect that more news
in this direction are unfortunately coming - but I sure hope to be wrong
on this one.
Oct 28 Bundle-adjustment updated
The old bundle block adjustement program (2002-)
was updated such that now all object space points: tie, gcp, or z, are
expressed by unique weights, and not
by point classes. The A-matrix
formation was added the derivates for the (Xobs - Xadj
= 0) observation equations, which in turn were added to the
vector of residuals. Now all points add three columns to the
(observation) A-matrix.
Observations all were included to have the a priori weights (sdev).
In [um] for the image observations, in [m] for the object space
points, and in [m] for the XY and Z-constrained point pairs. This allows
for more flexible
way of handling observations of different quality.
Movements of the 280 camera positions at some 40 camera positions
in Lapinkangas (Photography June) from the expected xy-positions.
There was one sub-block, in which the focal lengths seemed to
be off by 0.5%, the cameras were pulled down (~6 cm), indicating the
that the focal
length was too big. It was late nite photography with a small
f-number, and it might be that the f-number affects the focal length.
AH completed
the block adding tie points across sub-blocks of images, these
stabilized the tensions, and the final RMSE was 3.4 um, which is 1.4
pixels.
Largest 30 um residuals were observed in images that were taken
in windy conditions. In all, the June photography has succeeded well.
Link to thumbnail images.
A review was returned to us, and we are narrowing down towards
the final (improved) version. It seems that journals really differ
in how they scrutinize
the work done. Somehow, the level of the most scientific editors
seems like the right choice of Q/C. Although it feels painfull to re-do
things and
think over. Quick experience of the AISA images at Kumpula -
it seems that photogrammetry with the 2011 AISA data will be out of
reach. The
GPS was meter-sub-meter accurate, and the INS provided attitude
data of insufficient precision.
Oct 21 UAV-Waves at Aalto
Department of Surveying (Aalto) organized guest lectures in UAVs,
UAV-photogrammetry and LiDAR. Experts had come from Zurich and Vienna.
We thoroughly
enjoyed the talks. It was nice to be "the student" again. One
PEER review was finalized as was the Q/C system for the terresrial Lapinkangas
photos. We need
to triangulate them still next week, because the iWitness 3D
(offset + rotation) transformation of the bundle block still shows
some oddities. But these can also
be due to tree movement during the photography or during the
total station measurements (of GCPs). By redoing the triangulation
with bundle block adjustment,
we have some control over the errors. Highlight of Sunday was
another peer review that was due.
Lots of tiny sunspots (190) in 9 active solar regions. Looks
as if the most active ones will be Earth facing next week, just for
the CQ WW contest. This is good
or bad, depending on how calm these regions will be, or if they'll
spit stuff our way. 28 MHz was again good this past week.
Oct 18, Autumn holidays
Autumn activities: Viljo at the movies with James Bond. A &
P with grandpa at Salo military museum, where the highlight was Mannerheim
knight Lehtovaara (WW II pilot).
If you wish to keep kids happy - outside - it is possible if
you can afford a "trampoliini". Bigger toys are also fun, and a trampoliini
attracts older ones as well.
Viburnum in autumn costume. Angelica sylvestris too. Grey alder
seems to lead a happy (Greek?) life, not worrying about the winter.
Oct 13 Logs, R-matrices
and offsets, writing, deleting accidently, - the usual stuff
SAC SSB radio contest took place Oct 8-9, and I was priviledged
to participate in the aftermath - as a log assistant, or help-desk worker.
Namely, radio amateurs
no longer keep a log book on the paper, but try to manage with
computers. This is not always so easy and the helpdesk is confronted
with all kinds of requests.
That is educational though - opens eyes.
The geometry of the Hyytälä H-spectral (line-sensor,
HSLS) imaging campaign geometry was tried to shed some light on. The
overall idea is to check if the geometry can be
improved by using the images and ground points (or other (linear)
features) - in a postprocessing scheme. The sensor has an xyz-system,
and the projection center at time point t,
is at (X0,Y0, Z0)(t). If a target is
at position (Xt, Yt, Zt),
and it reflects ligth across the range of the HSLS, then where do these
rays end up on the image (xy) plane, same pixel
independent of wavelength? A HSLS sensor has a 3D position and
attitude at any time point t. A GPS/IMU cannot be mounted at the principal
point (inside the lens). The GPS antenna, which is the point that is
directly positioned, needs be outside the aircraft. This established an
offset (dX, dY, dZ) in the camera's xyz-system. Similarly, the inertial
measurement unit (IMU), which observes accelerations and the 3D attitude
parameters, has its own XYZ-system. It has an offset (dX, dY, dZ) with
the focal point (origin) of the camera, and since IMU cannot be aligned
perfectly with the camera (three rotations) - the IMU-observations do
not apply to the movement of the camera. This week we learned that in line-sensor,
the EO and IO parameters
can be solved (including the IMU offsets, boresight calibration)
by applying aerial triangulation with tie and ground control points.
How tedious process this would be ramained
still unclear, and the achievable improvements.
The LiDAR + AERIAL IMAGE species classification paper was being
written: Intro, MM and Results have now their basic form. The details
of the simulator and mixed-effects
ANCOVA need to be included for completeness. Having the multiangular
spectral classification simulator included is exciting. The reflectance
calibration error simulation needs
to be implemented though, still.
Ooops, I id it again. I had promised to do two peer reviews,
which were due about now. Some three weeks ago I accidently deleted
a whole pagefull of emails, thinking
I was killing spam. That page had the instructions from the journals...
I don't remember what else was in these msgs. Well I guess mail got
lost also in the old days.
October is the month for good radio propagation, owing to the
favorable position of Tellus on its orbit. The left picture has my radio
antenna, which is tunable between
14-30 MHz. Radio amateurs are allocated spectrum near 25 MHz
and 28-30 MHz. These frequencies have been poorly usable in the last
7-8 years. Quite recently, the
activity of the Sun has peaked up a little and the number of
sunspots goes up. Many hobbies are dependent on the weather, but ours
is dependent on the space weather!
There is no meters like ten meters - great openings this week
to North Aerica. Solar photos: (http://www.solen.info/)
Oct 7 Simulator of multiangular
single-tree spectral classification cntd
Some issues still remain with the whole concept.
The simulator is expected to have a module that performs reflectance
calibration to an
image, and the relative accuracy of the calibration is likely
to depend on the wavelength and reflectance itself. In addition, we assume
that the BRDF
of the reflectance calibration errors is bowl-shaped, increasig
with the view-zenith angle. There should be high image-to-image correlation,
because
if the atmospheric conditions are estimated wrongly in one image,
they are likely "similarly wrong" in the other images that are exposed
shortly
after (in an image block). Also, depending on the method, errors
might also be correlated between bands, if for example the parameters
are estimated from
the image data itself (e.g. dark-pixel method).
The simulator assumes that it captures pixel data for tree crowns
that have first been detected in the 3D scene in an xy-image-location
(polar representation of the BRDF).
A real tree detector / feature extractor would aggregate the
pixels into a mean reflectance. Now we assume that since we know the
real species, we can use
the BRDFs that have been estimated for all species and bands,
to obtain a mean reflectance for the detected tree in that view-illumination
(xy) geometry.
Now, this would result in zero variance, unless we assume some
intraspecies variation in tree reflectance spectra. By simulating an
reflectance spectra offset
(tree effect), we can create a realistic population of trees.
Then, we must assume that in making the image observation of that tree
from dirrefent viewing
angles, noise is introduced to these observations (as with the
real tree exctrator). In creating any erors, offsets and random, correlation
structure between bands
must be realistic. Also, the total variation must be realistic.
These we deduce from the real ADS40 2008 data sets.
---------
If you don't remember how many barns the Koskenkorva label has
- you can always return to this image. I took it on the way to OH8X
SAC CW, in search
of something that is true Finnish.
Some other flavoured versions of Koskenkorva have a label, in which
the barns are engulfed by some symbols.
Oct 5, Revision returned, first look at Hyytiälä 2011
Hyperspectral images
Hyperspectral images were acquired in Hyytiälä this
summer. One idea was to try to improve the exterior orientation to enable
photogrammetric use of imagery.
The raw images were processed to RGB in Kumpula, and these show
reveal quite bad distortions due to the sensor movements (platform movements).
In such image
areas it will be difficult to point known points, (but it is
not a desperate task). Revision to RSE was returned, and now two reviews
wait, mm.
Sept 30, ADS40 simulator - Understory
mapping
The complexity of the simulator is causing grey
hair. We need 8 correlated BRDF-features, tree-effect (a dark tree
is dark in all images), observation errors (random,
slightly dependent on the view-illumination geometry), and simulation
of reflectance calibration errors (offset + trend, error-BRDF). But help
is on its way.
The understory LiDAR mapping manuscript is being re-written and
shortened to match the work done in August.
The new results in transmission loss compensation caused a lot
of rewriting. The figure shows loss-corrected data on the y-axis. Due
to the inherent selection of subsequent
targets in LiDAR - the losses cannot be compensated, but trends
remain in the data.
Two new promises to do a peer review this week. Two moments of
weakness.
The frame sensor simulator
can be adjusted for forward and side overlaps. Such a sensor fills
the BRDF geometry with observations.
Sept 22 - Some progress with the ADS40 simulator.
The idea is to have a small area with trees, and this area is
flown over using four S-N flight lines, with both the B16 and N00 CCD
lines mapping the trees. The
Sun azimuth changes between flight lines (real change is continuous
over time). Perpendicular E-W-oriented lines can be flown to increase
the sampling. Strip side overlap is
a parameter used to control the number of multiangular observations
per tree. The above case had 12 minutes between strips while the lower
one had 24 mins.
The left image is a map of trees, strips, views and Sun vectors.
The right image shows the BRDF-geometry. X-axis is the principal plane
zenith, and the y-axis
is the cross plane zenith angle. This functionality allows us
to create image observations in the wanted and reasonable BRDF-geometry,
for testing the basis of
multiangular spectral tree species classification in reflectance
calibrated image data, given tree-species specific BRDF-patterns as
a priori information.
Another anaglyph stereo image of the Arcala Mammoth. Red-Blue
glasses required for 3D experience.
Sept 18, SAC 2011. Radio
Arcala.
Some pics from radio Arcala, where I participated the Scandinavian
Activity Contest. Left image has the Mammoth, while the mast in the
middle (M5) is easier to grasp. Autumn
colors on a local cottongrass bog.
late Tue evening, and its shock wave arrived to Tellus 3 hours
into the contest, 09GMT Sat. Magnetometers swayed to honor SAC - of course.
Radio propagation suffered - and more
towards north, where I was (65N). But that's just a part of the
game.
Sept 14, Meetings This
week was dedicated to "admin stuff", while trying to advance the three
ongoing projects (with poor success). The 1-day-delivery win7/64 laptop
arrived after
almost two weeks, but it's MAC-address is now registered and
the admin rights remained with the user. Meanwhile, all the network
card addresses in the other computers were
removed from the database for security reasons (these have Windows
2000 and XP). The evaluation of the research communities had its climax
on Tue, when the RCs met the panel.
Well, it turnd out that all RCs were more or less invented for
the evaluation, and the feedback of the panel seems to concern more
about the evaluation process than those being
evaluated. The head of the panel put it nicely "Soon ust PhD-students
have time to do real stuff, as the rest of us spend our time cross-evaluating
our academic proposals etc.".
Roughly a week after the release, PileUP! had 3000
visits from 73 countries and 2200 different people. It will be fun
to travel to Radio Arcala to participate
the SAC CW.
By Sept 14, the Sun has
been behaving quite OK. Solar flux is reaching 130, which could even
produce some 28 MHz propagation. Just hope the Polar
Path works for OH8X.
This pairs shows "vertical" images taken such that the lens was
pointing up. We had positioned points in the forest, where a LIDAR pulse
had intersected the forest
floor, 1.3 m above ground. And later placed the camera in these
points. In the image pair above, three pulses with path near the (right)
camera position were projected.
This data hopefully enables us to assess the relationships between
needle density and waveform traits. In the optimal case, we would just
see the pulse as a point, but
the positioning accuracy was ~40 cm, and the pulses pass the
lens from the side.
Sept 8 Recovering...slowly.
Oddly, the spare part HDD arrived on Tue, and Lenovo was revived, slowly.
It takes quite some time to go thru all updates since 2009, in
MS-Windows XP. Meanwhile, on Monday, PileUP! 15(4) was released
to get that process into an end. The orientation of the first terrestrial
Powershot was done using
iWitness, which we got in the summer. It has a handy camera calibration
procedure, and image orientation tools, that are based on stepwise
procedures: relative matching
followed by 3D similarity transformation (i.e. not bundle adjustment).
The observations weights cannot be set, which is a bit of a drawback,
in the low-cost version that
we've got. We had the camera positions measured with total station,
and the orientations match well with those of LiDAR 2010, as can be seen
in the first sample image.
Amazingly, it worked right from
the start. Waveforms in a pine tree.
Sept 2-4 - Laptop HDD
failure. Dramatic life change. The now 2.5-yr-old Lenove turned
me down. Guess the hard-disk just had rolled too many times. The previous
one, my first,
(HP) lasted for 3.5 years, but the Lenovo can perhaps be brought
to live with a spare part. Having the systems up again (productive)
will be challenging. There are software that
require XP, and then there are those that work with win7/64.
Will try to maintain both environments still, and Win2000. And, to complete
the list, there is still one machine running
win3.11/Dos 6.2 that has 1.2MB and 1.44MB floppy disks, just
in case. I'm not sure if the UH IT has this kind of service, I doubt.
In Finnish we say, "oma apu - paras apu",
meaning "trust only yourself". But how can you, if the IT people
just won't let you? Now I start to understand what is meant by those
who are worried about parents controlling
their kids too much - they become absolutely handicapped. This
is what the IT people seem to do - in any organization - I have learned
when talking to different people.
Arttu wanted to try out the 60 mm MACRO lens, which has not been
used at all. The Vaccinium photos are taken with that while the rest
with a 18-200 mm lens.
Over the weekend PileUP! grew to be a 52-page thing. At the same
time more the 100-hour limit was exceeded. Crazy!
This SAC edition is darn good, though I say it myself. It will
be fun to work SAC 2011.
"Pulling out teeth" - that's what it feels like, at times, when
trying to get people to contribute to the radio newsletter.
Aug 30 - New students
at the Department -day.
Aug 26 Leaving LiDAR
behind - back working with image data.
We had earlier noticed that the I1 data near the ground
differs from that further up (right, trees 0.5-3 m and > 3 m)).
The relative intensity histograms show this too. In second return
I2 data (left) the differences are marginal. Yet, also in that data,
the low values are rare in the other sensor, whereas the other
sensor produces also very weak values as expected.
specific BRDFs are known. The simulator draws observations from
these, in different imaging scenarios. Errors are simulated, such that
they have offsets, trends, correlation etc. These observations are classified
using distance metrics that show the deviation from the species specific
BRDFs. The idea is to check, how noisy can the data be, for this approach
to work.
The errors made in reflectance calibration are assumed to be
dominant. The figure show the two BRDF-angles when the nadir and back-16
lines are flown 20 degrees off the sun azimuth.
Aug 25 Time for simulations
The structure of the scatterers influences the I2 × I1
patterns in LiDAR data. A has diffuse small scatterers, B has large
leaves - larger than the LiDAR footprint, while C has well-
defined surfaces of different reflectivity (filling always the
footprint). The greyed area denotes the noise level, where echoes are
not triggered. I1 = intensity of first return, I2 =
intensity of second return.
The %-histograms of intensities in understory and trees show
differences between sensor, in their ability to produce low I2 data
near the ground. We need to constrain the
histograms with the h-information further, to see if the effect
(difference of relative histograms) is really related to height (near-ground).
AH got the normalization models estimatd,
using relative constraints on the slope, and mean intensity of
corrected data. Area-based analyses remain to be redone using the combined
I1+I2 data, where I2 data is
made equivalent to I1, first-return data.
Visited Otaniemi to register as a student for the 24th time.
Well, does time fly or what?
Aug 24, Reverse-engineering
photon detector offsets - KKM oscillations are settling down
In a desperate attemtp to figure out, where the receiver (intensity
measurement unit) has its origin of intensity measurements, we tried
finding pulses intersecting pine
trees (intensity 1, x-axis), and producing an echo from a well-defined
homogenous surface (intensity 2, y-axis). It turned out to be ill-posed,
we can say that an offset surely exist, but the
exact numeric value is impossible to determine (crossing of the
almost parallel noisy lines)..
The way KKM project managers get access to possible "excess-overheads"
was agreed this week. We still need to question (challenge) the 112%
overhead costs at UH/Forest
as they are clearly higher compared to our (public) competitors,
who do not have the obligations of teaching.
Aug 21, Radio "Art"
There are different ways of visualizing the radio logs. This
is a new one - the log spiral. The angular rate at which QSOs are printed
needs careful consifderation,
as does the speed at which the radius increases. By adjusting
font size the center of the spiral looks nice. Downsampling was used
to avoid aliasing artifacts.
The white flague theme seems better (black is quite sad, actually).
However, the speed at which the radius increases needs still tuning.
:)
Aug 19, War on losses
wasn't lost.
This is the kind of structure that seems suitable for the
correction of 2nd echo intensities, based on x and y, both scaled from
0 to 1.
The term r is needed (AH), because the intensity values, which
have a numeric value of 1...M, do not start from the bottom of the scale,
but from the noise level. The exponential
terms are scaled from 0 to 1, being 1 when x and y are 0 (a and
c being negative). We can now compensate the upper canopy losses in the
range that stays above the noise
level of intensity measurements (losses that are small enough
not to prevent an echo being triggered). Quite linear responses (in x
and y) were observed. x is the intensity
of the first echo and y is the estimated pulse path "depth" (explaining
deep volumetric reflections, estimated by Beer-Lambert law -based ray-tracing
for example).
The idea that intensity scale (observations) lack the
noise level is very important as it also means that we have a way of
normalizing first, single and second-return data, which
we had thouhgt to be impossible. This one we owe to Aarne.
Two PEER reviews + project management this week. The test-TEKES
project (starting in March 2010), which was the first "kokonaiskustannusmalli"
(~very-high-overhead) project
turned out to be suh that we actualy could use 30-35% of the
incoming financing to actual work. Attmpts to say (bookkeeping-wise),
that UH actually invested tenure-hours into
the work resulted in these tenure-hours being charged from the
project although it wasn't supposed to be so. I just hope that KKM-implementations
would be somehow harmonized.
It is, after all, mostly taxpayer money that comes thru KKM financers
of R&D.
Stupid or what, but to fill a few pages in PileUP! (and to occupy
myself for 48++ hours on this project?), I launched
a special SAC trophy program for PileUP!. The picture is from
2001, when such trophies were built the last time. Tubes (also QB3/300s)
were donated by individuals, so
I won't be in trouble, (I hope) later, when the manufacturing,
shipping and sending invoices -phase starts.
Aug 15, Fighting the
losses
The transmission-loss hastle is becoming a nightmare. I just
hope it would sort out somehow. It is now clear that our assumptions
were wrong from the start, and that the
real data is not actually behaving according to our expectations
(multiplicative dampening). We now need to decide what is good enough,
and that is not easy, if you are
uncertain about the validity of your assumptions. There is no
guaranteed stop sign, arghh...
Excitement
was in the air Tue mng, when the boys had the first schoolday. For Viljo,
it was the very first ever. It was good to share the excitement with classmates.
Saturday was en
exciting day too in Tampere!
Aug 11 Quick Visit to Central Finland
Juhtikylä
in Ikaalinen is 140-200 m a.s.l. and this is where Seppo (OH1VR) has
his contest QTH. Not far away is Seitseminen National Park, and the Kovero
museum farm.
After
Vantaa, Oberschleisheim, and Deutches museum (aviation), it was time
for Tikkakoski (EFJY) and Halli (EFHA). Brewster fighter retrieved
from a Lake in Carelia was the one
the twins needed to see. Tikkakoski museum filled all expectations,
or, maybe a simulator was missing.
On the
way to Halli, we stopped
at Oravivuori to check that
the "globe's degree of ellipsoidness" hasn't changed. The oldest hole in
the rock was from 1834 (Struve) and great views over
lake Päijänne
from a geodetic tower, which even had the instrument table aligned
right on top of the bolt in the rock. Halli aviation museum was such
a disappointment that we did not
take a single picture. Even the GNAT fighter in the centre of
Halli had been replaced by a Saab Draken, of which there are 13 in a dozen.
The museum, if we had actually found it,
was just a small storage area with uncut grass and a rusty Mig-15
in one corner. We decided to drive home via Hyytiälä, and in
Pilkottu, next to the forester's
log-house, we stopped
to do Botrychium inspection as we spotted a small dry piece of
open land. "Hopsansaa", we found two specimens of B. lunaria.
We were
so excited about moonworths that we decided to check an old meory of
mine. Namely, sometime 1991-1993 (it was May) we were with the course
in silviculture basics (MH1), to see the Murray pines in Siikakangas - near the old WW II airfield.
I remember that some student had brought a strange-looking leaf of a
plant, and it was doomed to be B. multifidum. So we went to see the place
now, after 20 years. It is a strange opening in the 20-30-year-old pine
forest. Maybe it has something to do with the WWII activity. Botrychium
multifidums were many,
despite the high vegetation, we spotted them in tens. We got
home late....
Aug 8 - LiDAR selects
its targets on the way thru the canopy
on the selection of subsequent targets. The picture shows a simulation,
where the radar cross-section (~silhouette area) of second echoes is
affected by the losses as measured
by the first-return intensity. We had notied this before, but
could not see the full spectrum of outcomes from this.
Aug 7 - Harvesting time
Blueberry
time in Kulloo forests. The boys did not smile a lot when in the woods
:) "Long-time-no-seen" Galinsoga
ciliata turned up in our own backyard!
Calluna is in flower, which coincides with the first attacks
of elk flies in the forest.
Aug 5 - Recomputing
canopy transmission-loss models in LiDAR data
In an attempt to model within-canopy transmission losses in LiDAR,
we had done modeling and parameter estimation in the absence of true
values assuming that by minimizing
the CV of intensity observations (per class, weighted mean),
we would find the right parameters for our multiplicative model that
amplifies the second- and third-return data that
for sure are subject to losses. We had considered transmission
losses as noise (dampening), such that a second return intensity value
(I2), should be amplified by k×I2, where
k depends on the pulse path and the previous intensity.
The red slopes in the corrected data differ between ground and
Birch targets, which of course should not be the case. A good normazation
model reduces the variation across the range of the explanatory
variables similarly producing trend-free data.
Maybe trees constitute entirely different targets compared to
ground? We will next try estimating the models separately for vegetation
and the ground.
Aug 2 - Heatwave is
over, so are the holidays.
Highlight of the week, if not
the whole summer, was when Tassu participated a "beauty contest" for
her race in Turku. The comments from the foreign referees
were
"I'd like to take her home". Tassu was the winner of all possible classes, a true pedigree,
and the car was filled with pet food and other trophies. Loniceras in
Salo reminded of the end of the
summer. Viljo and the boys got the chance to swim the lake Pyhäjärvi
in Säkylä after meeting with cousin Marcus.
It is only two weeks to the start of the school. Preparations
for another issue of PileUP! commenced.
July 29 - Warm summer
continues in Finland.
Porvoo
does not have a decent baseball team, so we travelled to Hamina to see
the game. In the 1970s Hamina was a top team in Finland, and the playing
looked good also
in 2011. Hamina played against Imatra. Nr 5 Botrychium (moonworth)
species was the highlight of the week. B. virginianum is very rare
and one nees to know from where to look for.
July 25 - Visiting Savonlinna
(links to Bavaria)
(A) Impatiens glandulifera grew wild in the Freising forests
(48N) and seemingly was an annoyance. Well, a week later I learned
that it is also growing across the city of
Savonlinna (62N) in Finland, but not in forests. I saw it in
lush and wet shores and in gardens. The size was max 1.5 m, whereas
in Bavaria it was over 3 m. Herniaria glabra
(Parkumäki), Geranium palustre (Savonlinna) were two other
more excotic species that I came across in both places. (B) The forests
at the test site of TUM in
Freising surely deviated from those in Finland. They were rich
in tree species for one thing. The tone of light that reached the ground
was nice in the Fagus and Acer stands.
(C) In Savonlinna our good old Esko knew the air taxi driver
and we got a ride over the old town for a really low price. For the boys
this was a really special case.
Olavinlinna with the opera festival (ongoing) tent. "Eskoslinna"
in Kellarpelto. Perttu (aviation enthousiast) sat next to the pilot,
who normally flyes in East Africa!
Eastern Savo offers good opportunities for moonworth hunting,
which I got to do a bit while there. Dry meadows are vanishing, and Botrychium
too. We stopped with Tassu at the
Rantasalmi road to look at the TVH-meadow in the crossroads.
Well, the 10 x 3 m area had some 10 specimens!! Most missed the "flowering
part" (itiöpesäkkeet) due to active
treatment. In Parkumäki, which is known for 1789 warfare,
we again saw Impatiens glandulifera taking over sites. The aliens species
are not so big issue (on land) in Finland that
the public would be aware of the problem, but I guess that they
come with the globalization of matters (along with many other unfortunate
'issues')
Savonlinna is known for good Ice Hockey goal keepers. With some
2000 other aboriginals we went to see the Stanley Cup trophy, which was
brought to town by Tuukka Rask,
Boston Bruins winner of 2011. The local (tearchers) band "Lipsaset"
tried their best in getting the Savonlinna people to party, but I guess
they faced a desperate task. Too serious
folks was our observation...
July 18 Austria - Bavaria trip
completed.
Lofer
(Austria) was hot, +30C at the fine ski resort. Boys preferred the
pools over the mountains. The Fish family (Paul, Christl, Stu) !. Viljo
at Steinplatten showed no fear for heights.
Legoland is not found only in Denmark, but also in Gunzberg (GER),
100 km from Munich. Nice location with lots of "participatory, do-it-yourself
stuff" and less "centrifuges".
Arttu at Oberschleissheim aviation
museum, which was nice and well completed by the great Deutcshes museum,
which calls for a week of visiting time. Perttu 's highlights
was
seeing the ME262 fighter
and the V1 and V2 rockets (saves us a trip to Peenemunde). The electric department
had some nice high-voltage phenomena combined with loud bangs.
cause hastle amongst us. We had namely not quite thought to catch
anything and if it wasn't for Teemu-the-Neighbor (photo), we would not
have the fish in the fridge, tnx.
80 cm & (roughly) 7 kg.
July 8: As the
campaign succeeded smoothly we could host the karaoke (as promised)
on Wed evening. WX was favorable the whole duration. 16 plots in 6 sites,
with
almost 3000 trees and 3.0 ha were mapped by the 49 students.
For the first time, we prepared also some area-based LiDAR estimates
for comparisons and that stuff
was being demonstrated in the field too, (by AH). (link to web-page).
RSE returned our manuscript so we know what do to when the holiday
season is over.
Details of the FWF-LiDAR of 2011 were also agreed. Data will
cover Lakkasuo, Susimäki and the regular area E of Hyytiälä.
June 30: Travelling continued - 20 years
of MARV1 teaching 1991-2011, This year's MARV1 course will be the last
of its kind (1966-2011). It was 5 weeks in the 60s-70s, four weeks
until now, and 3 weeks (but integrated) in the future. Fifth
try at measuring trees with the marv1 students for research purposes.
Altogether 16 plots with 49 students this year. We had a reasonable start
- hot but at least it wasn't raining.
It is hard to ignore the Siitama road sign in the summertime
- so we went to see moonworths as we had failed to find them in Vuorijärvi
anymore. The boys had their first Hyytiälä trip.
"Marvi on mukavaa" is the phrase. Hot WX as usual during the
start of the course. Marv1 combines LiDAR from 2010 with Ilvessalo's
tables from NFI 1 times - that is 1920s.
Hamina-trip: Sturm "Aili" @ RUK-museum now has roof on top of
it, and Ylä-Pihlaja bunker museum had been upgraded since our last
visit with some buildings and sites.
June 23: Finally visiting UMB in Ås Norway. Participating
LiDAR in ecological applications -seminar. I gave a 30-minute talk
about airbone (optical) RS of mire habitats (pdf).
Interesting talks e.g. about using interferometric SAR in palsamire
monitoring, LiDAR DEMs in detection of forested swamps, LiDAR DEMs revieling
cultural heritage, description
of habitats of predators and prey, as well as birds. UMB is the
place, where the potential of LiDAR was realized early on.
Recruitment of good students is an issue also
at UMB (smilar to UH and UEF). This ad probably is aimed at female applicants?
En konstigt "vedbit" hälsar på dörren vid UMB.
A sign in the lecture room reminds the students and staff about
basics of "snus-spytning".
Sample of UMB LiDAR experts: Ole Martin Bollandsås &
Erik Naesset. Hans Tommervik comes from 69N (Tromssa). Prof. Sorin Popescu
taught us how to respond to a "Howdy!"
-----------
June 19: On botanical
trip. Viljo and Perttu and myself we went to Karkkila, to an eutropfic
fen, where I spotted Cypripedium for the first time exactly 20 years ago.
The flowering was
past the peak, but we saw some fine examples. Tiny Moneses uniflora
were seen too, making us very happy amongst the mosquito clouds.
Summer is the time when Finns celebrate things after the long
winter. June 13, Arttu & Perttu turned 10 years!! June 19: our "local
road association" had summer party (yes, not all roads
are public). And midsummer is the peak for weddings - we were
also honored to be guests in one.
June 16: ISPRS Hannover. Ridiculously, to save money we
flew to Hannover by first waking up at 04 a.m. Tue, and arriving home
at 2.30 am Fri. One should
not do that anymore at this age. Anyway, a good meeting in Hannover
again Tue-Thu with participants from for example DLR, U GA, U Hamburg,
U Melbourne etc.
To enable fast IO of our cameras, we made the purchace of iWitness.
Good talks on CRFs (spatial context in classification), and multitemporal
CRFs.
The LiDAR-undestory talk was on Thursday (pdf).
On Monday we got the photo-albums done for Katriina, who's farewell
(retirement) party was on Wed.
June 8-9 Wed: Trying terrestrial photogrammetry
in the forest: With LK, we began preparations for the photography.
Tacheometer was tested and found operational. The signals
in the lower branches were finally built from polystyrene balls
(150 mm) that we bought from the sports shop (fishing gear). These were
painted in different colors and attached to
tight ropes that were (n=12) thrown up there with a (fishing)
spinning rod. A total of 35 camera positions were established as well
as 9 ground positions, which had been hit by
a LiDAR pulse producing more than one echo. We learned that it
is difficult to map targets higher than 45 degrees in vertical angle,
which slows down the tacheometer
work. Namely, we need to see the polystyrene balls from at least
two positions to get four angle observations and solve the XYZ. 28.5C
and cloud-free skies for the 2nd day
in a row. Great wx for aerial photography, but too hot in the
woods. Thu: Theodolite solutions
(2 azimuth + 2 elevation angles) for the "balls" had an average SD of
0.04 m. The sway
of trees is unavoidable.
The tacheometer network had an RMSE of 12 cm, when fitted to the UTM system.
From 12 to 18 azimuths to trees (marking GCPs) were observed for 4
tacheometer points,
and the RMSE of 12 cm is the residual error of the rigid XY-transformation
(shift + rotation). Thursday nite, all is ready for Friday's photography.
Hot weather
continues in Hyytiälä.
Fri: The great cloud-free,
hot WX cntd as we started the photography. We decided to have four sessions,
morning, midday, late afternoon and evening. Some
150 pictures were taken in the first two. The azimuths to GCPs
were remeasured but no major improvement - we need to observe
some interpoint distances!
June 6: Visiting
the land of ancestors in Säkylä. On the way back, we payed
a visit to the shooting tracks in Säkylä, where both father
and grandfather have won trophies.
On the 300-m track, we checked a .223 rifle. Boy, was that exciting!
At the actual site (85-yr-birthday party), Viljo was courageous and
sang Kevätsää.
June 4: School
is over - Perttu! Reading those grades Arttu, Eetu, Juho & Perttu.
Viljo bought military (anti-aircraft) glasses in Hyrylä. Here,
baron Viljo v. Richthofen pictured
in kauppakeskus Kaisa, where he did an intermediate landing before
attacking Korkeasaari Zoo..
June 1 Reports, pre-holiday
hastle, celebrations.
On Wed, we organized an ad-hoc seminar for AH
to tell the folks at the Dept, what the "LiDAR Boys" have been doing
since May 2009. To go with the cake,
we had a special guest in the form of Felix Rohrbach, no longer
from ETH, but from fm-International, Malmi, Helsinki. The Geodimeter
460 was fixed at Geotrim,
as expected, "the levelling bubble" was injured, and the batteries
had run out, since the device was last used in 2008. I had another
weak moment - thus another
peer review is due to IJRS.
AH presenting to a happy audience.
We did final adjustments to our share of the TEKES project "Social
Forest Inventory", involving VTT, Stora, UH, MosaicMill and Simosol.
The image matching products
(UAV-based imagery) were assessed for quality. Second last board
meeting on Wed - we still need to help one partner with the satellite
image interpretation.
The activity resulted in a request to prepare a proposal on how
to implement the KKM-practices at the Department. Must remember to do
this.
Both of JT's language reviews arrived - now we are still waiting
for the decisions from FEM and RSE (since 5 weeks). RSE published the
BRDF-work.
School celebrated the spring in Kulloo. Arttu played Peer Gynt
and received a hug! The new Kulloon koulu (Finnish) and the old Kullo
Skolan (Swedish) form the Kulloo Educational Centre.
Friday was Minna's big day, and we enjoyed
it thoroughly. Päivi & Bosse. Prof. Kari Korhonen (opponent),
Niilo & Minna. Professors emerita: Laasasenaho & Poso.
May 25 Preparing for
field photogrammetry
AH returned to work on the UAV-stuff (Social Forest Inventory)
project. I had a weak moment and promised to peer review for ISPRS
JPRS. The good old Geodimeter 460
total station was picked up from Old Viikki, and taken to Geotrim
for check-upp. These instruments are rented for 100€/week, so it might
not be so wise to fix the old one
if it is in bad shape. Discussions about possible airborne hyperspctral
scanning in Hyytiälä this summer - some concerns over the
geometry remained.
Rectified the PowerShot G6 image (see below) - the yellow bk-gnd
shows as stripes that show the distortions.
The disrotion free image follows the principle of the pinhole
camera (simple interior orientation). The intrpolation
of the missing pixels needs to be incorporated (antialiasing).
PileUP! 15(3) was released
on Monday afternoon. 56 pages this time, of which 66% were finally
done by the editors, which is sub-optimal. Some 1500 readers in the
first 48 hours from 70 countries. Direct links to the pdf do
not show in these numbers.
Marv3 course (forest management planning) and Marv1 (course preparations)
in Hyytiälä. Hectic times: some trojan virus attacked and
we spent 2 days trying to
salvage stuff. Well done by anyone who does these. With AH we
measured alltogether seven sites with 20 plots for MARV1. LK came from
Joensuu and we tried
photogrammetry in the field. Not very easy, but seems feasible.
The idea is to know the XYZ-positions of the cameras as accurately
as possible. Then, by placing targets of known length and orientation
(either levelled or collinear with
the plumb line), we could solve camera orientations also in the
field. Must re-consider the use of omega-phi-kappa angles when doing
terrestrial photogrammetry. If the
cameras are tilted almost towards zenith, they are ok, but otherwise
it is utterly difficult to get the initial approximations for triangulation.
The fig shows an image par
from Hyytiälä machine hall, LK stands, where 2-echo
pulses were known not the exist. The HSV colors depict the waveform (sampled
amplitude) at 15-cm intervals.
The circles denote echo-positions (dscrete-return data in ALS60).
Camera used was the Canon powershot G6. The vertical aerial image is
from 2010.
(1) Carex nigra ssp. juncella
(marv3). (2) Poa remota, Chrysosplenium alternifolium,
Cardamine amara (marv3). (3)
A spruce seedling lifted up by rouste (frozen water pillars).
Pictures from the field (marv3 / 2011) - the last marv3-course
at UH.
May 8, Mother's day,
boys' stuff
There are not many things that make countryside a better place
to grow up compared to urban neighborhoods.
May 6, marv1, marv3
& tekes hastle
Installed all image (1946-2010) and LiDAR (2004-2010) data sets
from Hyytiälä into the win7 machine (~800 GBytes). Kuvamitt_hyde2011
was tuned, such that the
marv1-tree measurements could be carried out easily (first time
not by me) and using the latest LiDAR data. Course marv3 starts next
Monday and the final tuning of the aerial images
and GPSs was done. A full car load of stuff again. This will
be the last known marv3-course, mine was in 1990, and I taught it 1997-2000.
Tekes CHM comparisons
went smoothly ahead.
front cover.
May 3, Spring slowed
down in Kulloby.
February daphne
is "April-May daphne" in Kulloo. Ice
melting finally in a culvert by Laversintie-road. Salix caprea was still in flower after
two weeks of blooming.
Lupine is an
early bird and on standby for more conquests. An alien much disliked
by me, but loved by many during midsummer. Star-of-Betlehem (Gagea lutea?) is one of the
early flowering plants. Kulloby
chapel - the copper roof is now matt after the renovation a
couple of years ago.
April 29, Snow breaks
submitted. Proposal-writing for the Academy.
On Tue we sat down to do the final adjustments and FME was submitted
the article. mmvar37 exam #2 was organized on Fri. We planned the final
leg of the TEKES
project Social Forest Inventory and Aarne began on Tue tuning
the reference-, foto- (UAV) and LiDAR-CHMs for the two test areas of
2010. Wed-Thu we prepared a SHOK
proposal for Academy of Finland. While writing that on Thu afternoon,
came the "Real Jytky" from the
Academy - they had nominated me for Senior Reserach Fellow position,
9/2011-8/2016. "PileUP! GOES image processing" was added to the
draft of 3/2011 which is being processed. Image processing was also used
for the FME bicolor
images to turn the greys into reds :)
X-ray image of Sun, October
1959 (PileUP!). Bi-temporal LiDAR point clouds in a snow-damaged forest.
April 21, Snow-breaks
collected into two piles
This week was spent finalizing the snow damage mapping articles
for Galgary & Forestry journal. Submitted also the Mapping of understory
trees -paper. A journal
article was peer reviewed. Discussion on how old courses can
a student "claim points for" when making a new degree. Similarly, activities
around the IT-services.
The scan zenith angles (x X-coordinate) in one plot. Optimum
of Mikko's snow-break detector, Contrasted CHMs and the snow-breaks.
April 15, ISPRS 2011
Hannover + Galgary. Tampere & Joensuu
Snow-damage work got expanded towards area-based
analysis (just a pit), and an ISPRS report for Galgary was written (Laserscanning
2011).
We finished and submitted the understory mapping 6-page paper
for ISPRS Hannover. Another one going there too - it is about the radiometric
EUROSDR-tests. The understory mapping (long) journal paper is
also ready for submission - just some final polishing.
Course preparations (forest management planning) in Tampere,
where we saw "LiDAR in action" - and heard of the new inventory system
in Finnish
private forests. Travelled to Joensuu to see Lauri Korhonen defend
his thesis on canopy cover measurements, ground- and airborne. Faculty
board meeting on Tue.
Photos from the thesis-defense by Lauri Korhonen, who is by far
the best expert in canopy measurements in Finland. This was also the
conclusion of the opponent, Dr. Varjo.
April 8, Understory+LiDAR -research culminated this week. ADS40-project
that started in Hannover 2007, too.
We pushed hard to the get work under way before
new teaching assignments commence. We learned about the echo-triggering
probabilities among
tree species in the understory layer, models for correcting the
transmission losses were evaluated (and it seems that despite partial
success, the data
is not worth much), and we played a little with area-based estimation
in the understory, using normalized h-distribution metrics that should
be invariant
to the variation of the overstory. Quite exhausted after all
pushing - now we need to sit back and see carefully what we actually
did. Quick comparison to
the plan made in Nov showed that we had done mostly what was
planned (stuff). Aarne did a great job with his R and Java code.
This fig shows the near-ground
intensity-height distributions in the two plots by LiDAR data sets.
Black dots are single
returns, while red depict first of many echoes. ALTM differs from ALS.
Graphs A and C show that even first and single return data are
subject to transmission losses that
reduce the intensities in the understory. These losses are weak,
and there is no echo-triggering in the overstory.
The ADS40 paper was finally accepted after being entirely re-shaped
from the first version (May 2010, to another journal), to this report,
which was greatly
influenced by the reviewers. The project had began in Hannover
2007, where I met with camera different manufacturers and
prof. Petrie, who knows all about latest technical gimmicks in
photogrammetry. In August 2008, ELB flew the sensor in Hyytiälä,
and winter 2008-2009 was
spent implementing the sensor model. Fall 2009 and spring 2010
Felix and I, we did the atmospheric & BRDF corrections in the data,
programmed the tree-level
pixel retriever and a feature-extractor that computed from the
pixels of each (163000) tree the statistical features. During 2010 this
data turned into the article
that was now finalized and accepted to be published.
April 2, Gas on the
understory pedal.
Ville had good news as the ADS40 tree-sp classification study
(which was based on our previous work still not yet published)
was accepted to IEEE TGRS.
We were figuring out what might be the accuracy of area-based
change estimation (e.g. intermediate felling in a forest) from bi-temporal
h-distributions:
<><> Left: A 22 x 22-m area with modest snow-damage,
and h-distributions from 4 LiDAR strips having differet average scan
zenith angles (3,10,10,15 degr). Due to oblique scanning, the
2010 distributions deviate considerably (gnd-penetration). Well, we've
seen this when estimating LAI. (Right)
We
started also doing area-based analysis on the understory plots. To
show what would be the most extreme effect, figure on the right shows
the change in LiDAR h-distributions (another 22 x 22 m area) due the
complete removal of faily dense understory (in 2009).
This is how the point patterns of the upper and lower story trees
look like in our plots.
"Pseudoecho maps" drawn at the approximate height of the crown
base of dominant trees.
These pulses produced echoes from below the mapping height. Colors
depict the number echoes per pulse.
Circles estimate crown perimeter. HOw these patterns affect the
estimation accuracy of the understory is what we also investigate.
March 31 "What a winter!"
Snow in almost all pictures this winter. Skiing season has extended
over 4 months, which of course is great.
Maybe my opinion changes when the next electricity bill arrives.
Some miscellaneous pics from the camera.
OH1RX and snow-work @OH2PM. Arttu
and the snow-glide. Viljo skiing to kindergarden March 31.
Viljo (6), Eero (82), Perttu (9)
and Arttu (9) in Märynummi, Salo (Halikko) March 27. Viljo getting
a mönkkäri-ride by Timo OH1NX in Paimio (March 26). View in Kulloo,
March 31.
The snowy winter will increase cash-flows in garden shops this
spring. 48-m mast of OH1NX in a DX QTH. Tassu has fully enjoyed the
free-snow-running.
March 25
Monday spent preparing a presentation for FGI
(pdf). Tue at FGI also to discuss about the
fwf-work. We should soon
start with the low-vegetation and then
with sp-detection (assuming that peaks are relevant and previous
losses must be accounted for). The h-distributions from 2004, 2006,
2007 and 2010 could
also be used for snow-break
detection, area-based, it seems. Thursday we Alumni celebrations
in Viikki. We had a little "lasershow" for the alumni (our DPW
with aerial images & LiDAR was made available for people
to try it out, pics and videoshows on the computer screen, old and new
posters and publications).
The main task of the week was to finish the revision #2. MARV3
course will be given for the last time this year. I will "return to
the scene" after 11 years.
Slides from FGI talk
that looked at the past activities in Hyytiälä.
The audience at FGI was
active (J. Hyyppä, E. Ahokas, N.N.). Juha's Darth Wader mug reflects
his spirit.
The alumni foresters gathered to try LiDAR. My old professor
Simo Poso was courageous and tried out digital photogrammetry. Simo's
tree h-measurement was "eighteen-ninety".
We had to remind the audience (Prof. emeritus Leikola, left)
that it was meters - not a year! (Photos Mikko Havimo)
March 18, ADS40, Snow breaks & LiDAR undergrowth
Multiangular ADS40 work: It seems that the optimal use
of multiangular spectral data in the presence of view-to-view offsets
and within-view errors in the atmospheric
correction is not very trivial. Normalizing to nadir reflectances
using a single BRDF is not optimal if the species-spesific BRDFs are
known to differ from one another.
Yet, if the differences of mean reflectances (per band) are 5-10%,
and the atmospheric correction has an accuracy of 5-15%, combining
the different views (and corrections)
requires some support from the object space if anything else.
Let's see if we find anyone else saying "been there, done that!".
Snow break manuscript
was (Tue-Thu) edited and it seems that the concept of vertical canopy
cover, or canopy cover in general is not unambiguous at all. I recall Prof.
Nyyssönen
sayng "Kas teoria hyvä on, mut metsäs aivan mahdoton"
("theory is fine in the office, but often won't survive forest").
Started making the 2nd revision
for the ADS40 manuscript. This experience confirms all earlier
findings about the importance of good experimental design when doing
empirical reserach. Our 2008 design surely had some flaws and
this makes the interpretation of the results often ambiguous or subject
to speculation. We see statistical
effects, and claim that some factor is contributing #%, but the
contribution may actually be an artifact from something we could not
do in a perfect manner. All this was
not so obvious to us as it was to one of the reviewers, who is
putting in a lot of effort to help us.
Faculty board meeting on Tue, and in Thu afternoon we had a nice
skiing relay race with students from Viikki campus. Not many teams
but a good start we hope.
PERS 3/2011 had the front-cover + a 3-page article about the
airborne optical RS activities in Hyytiälä!.
Sliced LiDAR data shows broken, fallen and bent trees. Tree crowns
projected in side-lit RED band ADS40 image. That image and the image
showing the angles was redone,
maybe for the fifth or so time this week for the new revision.
The last pair shows how the LiDAR ('06, '07, '08) scanned the two plots
for the understory mapping.
OH2PM birthday celebrations in Kiikala. OH2PM; OH1NS & OH1TX;
OH2BO & OH1VT.
March 11, ADS40 BRDFs.
The binary image operations seem to work, but
we had to re-define what the Y-variables were in the snow-damage mapping by LiDAR -work. There
is vertical canopy
cover (0-100%), but when a tree looses 30% of the living crown,
VCC might not be affacted, but crown volume yes in such a case..
The internship arrangements failed, and it seems that we have
to do without a French intern this summer (after all the preparations...)
Mon-Tue I thought that I'm a great mathematician, who has found
out that proc reg and proc glm in SAS
software have bugs as they were not giving similar results to
R (in Joensuu) and Matlab. We were estimating a general BRDF-pattern
for trees in reflectance calibrated aerial images. No, it turned out,
as usual, that my SAS script had a bug. I thought
having introduced the term x^2*y^2, but instead it was x^4*y^2. In a polynomial surface
model this had some significance. And it meant that a lot computations
for an
earlier paper needed revision (in revision). The ADS40 BRDF-corrected
(to nadir) reflectances helped, when combining observations from several
flight lines. Yet, not always. Sometimes,
the atmospherically corrected reflectance data produced higher
classification accuracies compared to our BRDF-normalized data. In
such cases, it sometimes helped if the blue
band was removed - as if it had strong offsets between strips.
This checking took almost all week, and it seems a very complex matter.
+ plan different Training - Validation Scenarios (geometry, leave-stand-out)
+ check the remaning trends in strips were the BRDF correction
is not helping.
AH started to program the "optimizer" which we use to find the
optimal parameters for our non-linear model for correcting losses in 2nd and 3rd return
LiDAR intensity
data. Beer's law -based path integrals were not, at least at
the start, any better than simple geometric slicing of the crown. 5
parameters makes the optimization comptationally
demanding as we are trying to find the set of parameters that
minimize the wieghted mean of intensity CV (for the species in the understory).,
and maximize the kappa
of echo classification, based in intensity alone.
+ report SLVTS
A: After the
high-order tem was corrected the brdf patterns were redrawn. Here,
Maple 7, from 1999 clearly outperformed Maple 13 in export quality (eps).
Well, others have
made the same conclusions in the net. B: We got this picture from SMEAR II
in January 2010, snow-loads are accumulating but crowns ok.
C: Hyytiälä
LiDAR 2010 data in the front-cover of PERS 2010!
March 5, More daylight,
farewell winter blues!
Perttu, Arttu & Viljo enjoying the 50+ cm snow. Mickelsböle,
Borgå: willows.
Construction starting near our home. Logs made from trees felled
from the site are being sawed to battens etc. Arttu on the kick-sledge,
which was the #1 vehicle this winter in Kulloby.
See the amount of snow March 2011.
March 3
Python
course ended with 17 students at the exam. 11 passed and 6 failed. Here
is the cource code for ex1 (Stem bucking)
and ex2
(Forest inventory simulator,
Zip with trees, stand border, simulation
control file). The exam looked like this (pdf). Moodle.helsinki.fi has
the other material (Faculty of Agriculture and Forestry, Basic's
of programming. Login as Guest).
Snow damage work:
workflow for the binary image analyses was done by MV this week.
Transmission losses:
We found a candidate model for correcting the intensity data of 2nd
and 3rd returns. There is substantial reduction
in CV of per class intensities after compensation. Compensation
considers non-seen losses, geometry and correction by previous
echo.
Internship negociations. Preparation of proposals. ADS40 hastle.
Feb 26
Python
course on the finish line. This week we did the Monte Carlo stand cruising
simulator that send out a crew N times to a forest
with 641 trees enclosed by a 0.58-ha polyline. The crew measures
M circular forest plots with radius r. These are randomly located inside
the forest, entirely fitting. The crew selects the trees (does
Gaussian(mu, sigma) distance errors), measures the species (classification
error-matrix),
the dbh (additive Gaussian errors),and the height (additive Gaussian).
The stem volume of each tree is computed with volume functions that
have Gaussian model errors such that the sigma of mu (sdev of
stand-level bias) is 5%, while the noise (within stand stdev of model
errors) is 9%.
The simulator reports the mean and stdev of stand variables:
(stem number, volume, basal area, basal-area weighted mean dbh, and
height).
Normalization of the
transmission losses does not seem to be as straightforward as
we thought. Now we need to carefully re-consider the model
for the losses, i.e. how to take into account the intensity of
echo 1 and 2 in the interpretation of echo 3, for example.
Snow damage work:
Some figs were done and we must now reconsider the binary image analysis
that is done for the delta-CHM images.
Also, how to assess the overall performance by considering at
the same time {correctl found trees, omission errors, commission errors,
and the
delta-VCC estimate}. Make all steps into a script and perform
sensitivity tests for the 2-3 parameters.
L: The 10 plots
from which we collected field data in 2010 (red = snow break). M: Plot P_M_08 (from MARV1 course in 2008).
LiDAR of 2010 and
snow-broken trees in yellow. R: LiDAR of 2007 (red) and LiDAR
of 2010 (green) shows where there are changes. This plot was damaged
severely.
Binary difference image. Removed small connected
components (CCs) + 3x3-Median filter + Another removal
of small CCs
Feb 17
The Python course
lecturing changed into demoes, where we go thru two exercises: €-optimal
stem bucking and Monte Carlo stand cruising
simution (forest inventory simulator). Remember to have the Python
2.5 uppgraded into version 2.7 on the computers in the GIS-lab.
Python seems to be very complete high-level language- graphics
we did not do so that side remains to be seen. Two weeks still to go.
18 out
of 24 who showed interest in the course are still rubbing along
after 5 weeks. Remember to 'force' students to buy the textbook next
year.
The ADS40 + LiDAR stuff
was planned during the week. We will try the empiric BRDF-correction
approach, which would normalize all observations to
nadir reflectance data. Image features (mean, sdvev, min, max)
would be invariant to the correction and it would be possible to combine
different
views, image lines and even flying altitudes that way. AS is
putting these efforts forward.
The understory work is
still dealing with the normalization of the transmission losses occurring
in the upper canopy. For single and first returns in
the low canopy - the losses are small - by default - otherwise
it is likely that they'd not be first echoes but 2nd - 4th. Aarne modeled
the multiplicative
dampening of the signal in the crowns along the path (intersection
with crown model), and this way it is possible to correct the downward
bias, but
the correction increases noise, because there are also such real
pulses that were not subject to losses - and there is no easy method
to detect them
among those that lost photons "up there". We need to continue
with the multi-return pulses next. The manuscript was sent out to RSL
after being tuned.
The snow-damage manuscript
was finally revised on Friday. We need to find examples of monitoring
applications, in which multitemporal LiDAR data
were used, and the strength of accurate geometry (sampling errors
=> 0) was used. We got the basic information about snow-damage, yet,
might there be
NFI data on the phenomenon?
The ADS40 paper came
back. Luckily there is no time limit to do the revisions. Another very
accurate and critical review (50%).
Tue, 2-h faculty board meeting. Wed, 3-h general exam superivising.
Evaluation of research -stuff.
# Approximate
Pi the Monte Carlo way
import random, math
pi = math.pi;
N = 1
for j in range(20):
i = 0; In = 0;
N = N + N ** 1.01 # Through this many
stones
while(i < N):
# Count those inside the unit
circle
if math.sqrt((random.random()*2.-1.)**2
+ (random.random()*2.-1.)**2) <= 1.0 : In += 1
i +=1
print "Pi %.4f" % (In/N * 4.0), "%8d" % N, \
"stones, delta: %.2f"
% ((abs(pi- In/N * 4.0)
/ pi * 100)) + " %"
Pi 2.0000
2 stones, delta: 36.34 %
Pi 4.9827
4 stones, delta: 58.60 %
Pi 3.4636
8 stones, delta: 10.25 %
Pi 3.9171
16 stones, delta: 24.68 %
Pi 3.8624
33 stones, delta: 22.94 %
Pi 3.2611
67 stones, delta: 3.80 %
Pi 2.9022
137 stones, delta: 7.62 %
Pi 3.2695
282 stones, delta: 4.07 %
Pi 3.0741
581 stones, delta: 2.15 %
Pi 3.1728
1201 stones, delta: 0.99 %
Pi 3.1293
2491 stones, delta: 0.39 %
Pi 3.1474
5185 stones, delta: 0.19 %
Pi 3.1322
10833 stones, delta: 0.30 %
Pi 3.1398
22721 stones, delta: 0.06 %
Pi 3.1284
47840 stones, delta: 0.42 %
Pi 3.1375
101123 stones, delta: 0.13 %
Pi 3.1352
214598 stones, delta: 0.20 %
Pi 3.1402
457228 stones, delta: 0.05 %
Pi 3.1418
978104 stones, delta: 0.01 %
Pi 3.1417
2100869 stones, delta: 0.00 %
Feb 6
Too busy with teaching and administration to keep this weblog
alive. Python programming -course, while AH, AS & MV are keeping
the research alive. Interesting findings concerning the probabilities
of snowdamage. Understory research is also making good progress
with expected and unexpected findings. I just don't get it, how
can I create immutable lists in Python?
class UniversityLecturer():
def __init__(self):
Busy = True
TimeToConcentrate = False
def getMoreEnergy():
while not(self.Busy):
return getMoreEnergy()
Jan 18
We tried to locate trees that would have strongly swayed July
19, during the fw-LIDAR in Hyytiälä, which combines several
overlpping strips
to reach the high point densities. Despite high winds, > 6
m/s as 1-min average, we could not find such trees. To be sure that the
patterns observed
in the point clouds are really due to wind, some immobile targets
were observed, which included roads (paintings) and buildings. The picture
shows
LiDAR data superimposed in an aerial view from three overlapping
strips. The green strip possibly deviates 10-20 cm in XY from the others.
Jan 14 Hastle began
The LiDAR features were re-computed for the 15800 "ADS40 trees"
for the 2006 and 2007 data, and first results in RF, knn, LDA were obtained.
The AGC parameters for the 2010 LiDAR were solved.
Highlight of the week was when the 1 Gbps switch arrived on Thu.
It speeded up the data transfer by a factor of 5.
A bit of hastle around Python - its syntax is dangerously
similar to old and familiar languages. Getting familiar with it and
the Moodle system.
Point clouds arrived for the TEKES project.
Waveform analysis - it will hopefully continue?
Aarne cntd fine tuning of the java-programs, and presented with
first 3D viewer tools.
A few meetings and an interesting evening with the odd fellows
of Porvoo on Tue.
10 days after the release,
PileUP! had more than 2500 "registered" readers.
Jan 5, 2011 New start?
Since beginnig of this year I have started to work as a member
of the 'tenure staff', with more teaching responsibilities and less
research possibilities. The idea
is to even more conduct co-operation with people that will have
more time for actual work. Aarne luckily started to work as research
assistant Jan 1. Btw, I did serve UH for
19 years temping. One gets used to it, so it will take time to
get used to the idea of not being under constant where-to-get-the-next-funding-pressure.
I feel as if I'm returning to do stuff I did 1995-2000, although
the infrastructure has changed a lot at UH/Forest and in the business
of forestry, forest inventory and
management planning itself. In many ways and mostly positive.
In 1995-2000, we had assistants (young people with Master's degree, aiming
at PhD), now the same
elementary teaching tasks are done lecturers, who all seem to
be qualified docents. I wonder if this makes sense, entirely.
So its not a fresh start, byt has some flavor of freshness.
TBD
Reporting: LVTS, UH ?
To UEF: ADS40 data and feature extractor, Compute LiDAR
features for the 15800 ADS40 trees.
Point clouds, Jan 15 - from Mosaic Mill - comparison to als60
data
Prepare the PYTHON course - The installation
Read the waveform analysis.
Xmas holidays, PileUP! 15(1-2)
was prepared and released Jan 2, 2011.
A record-thick 72-page issue this time.
Dec 17 Joensuu Xmas meeting
Good news from RSE arrived concerning LK's LAI & CC-work
that we finalized last week. Also VH's reviews from ieee tgrs arrived
and were deemed positive.
The mammoth's language review came on the 15th, and mammoth.R1
was returned on the 17th.
Tue was the F-meeting & departure for Joensuu. We filled
the skeleton for FM during the trip. Seminar at UEF & meetings
to start new co-op. Some data
preparation awaits tbd. No skiing this time in Joensuu - too
cold - we went to Vesikko instead.
Friday ended the 19-yr temping.
Dec 12 Killing old projects
I have learned that the long-term temping for UH might be history
soon. This means that existing projects need quick finishing and the
start
of new projects will await for some time.
Helped colleagues in Edinborough with LiDAR data in the vicinity
of the SMEAR II station:
Hyytiälä DEM from
S-SW
The point cloud (10 GBytes), a Word-macro and a DEM were delivered.
The DEM was filtered from
the 2-km data of 2010 scanning @ Finnmap.
Helped PP with a LiDAR campaign in Africa to see what is feasible.
Three companies to operate in Kenya were found and issues related to
operating
there were uncovered a little.
All financial matters were put in order, incl. also the TEKES
project, which needed a new invoice to be sent from us.
Painfull to find ones way in the KKM jungle.
The MÄ & KU plots in Hyytiälä were marked
for thinning.
Dec 2 Mammuth was revived
Quite a bit of effort was put to the manuscript, as we had to
recalculate some parts due to wrongly defined brdf-angles. Numerically
the changes were
minimal (sin(a) vs. a, a < 0.5), but
now the data makes sense. All but one figure was enhanced etc. Rebuttal
letter became 30 pages long.
AH has made great progress - we will now write the skeleton for
the article, and hopefully get some more experts to join us.
A Nepal seminar at FMI, where I contributed with a small
LiDAR talk.
Trying to get the administration / reports / done for 2010 for
the projects. It seems that the kokonaiskustannusmalli is really painful.
It is creating lots
of job opportunities - far away from research and teaching. I
hope that I'm wrong and that things will improve. Everyone is saying
the same - the taxation
for administration from the research projects has increased side
by side with the increasing amount of hours that we need to put in administration
ourselves.
And in their wisdom, the project administration was out-localized
to what are called the "service centres". I don't know - maybe I'm just
not qualified
for the administrative work. For sure - I'm not motivated.
Another CQ WW CW Contest is behind. 2000-2010 I have been
a guest at OH2BH. Until now the visits were single band efforts, but
I had the courage to
try SOAB again. Last of that kind was in 1999 in Sysmä.
How time goes...
Returned the review.
Nov 18 More change detection
Canopy height model (0..20 m) of 2006 and 2010, and the
difference raster | Hdiff > 3 m and H(2006) > 12 m.
Nov 17 Mammuth is back - Snowdamage mapping got started.
The Mammuth is back home from RSE with four reviews containing
lots of improvements (clarifications definately needed) to the article.
Picture from August 2010. Plot P2 (Pine, 60-yrs) seen in an aerial
view with 2006 and 2010 LiDAR. Snowbreak occurred Feb 2010.
Reference and LiDAR data for the 10 plots are in this folder.
Nov 15 Dissemination of results
-season is over
Busy disseminating at the Finnish Remote Sensing Days, "Does
the LiDAR see the Trees from the Forest" seminar, and at the Gävle
Leica Workshop at LMV.
Using multitemporal circular plots centered at well-defined surfaces
with and without GNSS-control, we computed the Z-discrepancies in the
2004-2010
1 km LiDAR campaigns. These were 0.231 m (2004), -0.051 ('06), 0.011 ('07), -0.042 ('08), and 0.201 m ('10). Using these corrections,
the Silmäpäänlammi
fen data (fig. above) shows nicely how the mire was filled with
the ample rainwater of August (23), 2008. The Z-errors are due to range-calibration
issues and
direct sensor orientation (GPS) errors.
Nov 10 Elevation Dynamics
Optech 1233 (2004), Optech 3100 (2006), ALS50-ii (2007), ALS50-ii
(2008), and ALS60 (2010), 1-1.3-km LiDAR data from r=1.5 m circular
plots in Hyytiälä. Asphalt and low grass should produce
stable elevation data. These data were produced relying on the GPS XYZ
and range
calibration of the LiDAR (no ground truth was available but maybe
for the 2008 campaign). There seems to be 30 cm inaccuracy in the Z
of these well defined surfaces.
In Hyytiälä forests prevail, and roads are the only
option for well-defined surfaces. In the example the road was 're-done'
in 2006, and especially
a section that had 'sank' in the peatland was 'lifted'. Roads
do not comprise most stable targets, but they'll have to do.
Nov 7 Powerpoint slide production
week.
Mon-Wed prepared for the Remote Sensing Days (@VTT, Nov 4), LiDAR
Seminar (@ Evo, Nov 9), and Gävle Workshop (@ LMV, Nov 11). Thu-Fri
participated the Remote Sensing
Days. Promised and worked a bit on a peer review. Final adjustments
to computations by AH.
Oct 29 Applications and more
Another Academy of Finland week, with many lights on in the Metsätiedeiden
talo office windows still after 8 p.m., when leaving for home. Participated
in two this time.
Waveform processing took some first steps.
On Saturday some 70 people attended the funeral in Ikaalinen
to say farewell.
Oct 22 Meetings, application
writing and waveforms.
Built a simple application, using which it is easy to point targets
in aerial images (such road points, treetops) and collect waveforms.
The figure above shows averaged waveforms for different targets;
flying heigths 1.3, 2, and 3 km combined. As can be expected, the more
diffuse
structure of pine results in longer fall time compared to spruce.
The low seedling stand vegetation includes 0-100-cm high Calamagrostis
epigejos,
Rubus idaues, Deschapmsia flexuosa and Epilobium angustifolium
(and probably small willows, rowans etc.). 1 meter in target height
is equivalent of
7 nanoseconds. Comparison of the asphalt and low vegetation waveforms
is difficult.
This figure has one more target added. "Grass" (0-5cm) produces
similar intensities to low (0-1-m) seedling stand vegetation. The lenghtening
of the
echo is seen by comparing these two classes.
On Wednesday we had a 2-hour Social Forest project meeting, and
our WP was in schedule and the promised deliverables had found their
way.
On Thursday, the Dept board had a meeting. Writing a proposal
for the Academy in between other tasks. Three peer reviews were completed
this
week.
Oct 19 More waveforms
Splitting the als60 data into 1-ha files took 24 hours on the
freshest Win-7 machine. Need to do a redundancy check for the format,
that is storage
consuming currently (use of 8-byte floating point numbers etc.).
Started to look at the waveforms just to verify that the data is in
order.
For well-defined surfaces having
different reflectance properties, the received waveform seems to have
a an ending tail (fall time deviating from the rise time?).
Aspen shows triple peaks, with about 1.5 m and 2.5 m in between.
This pulse probably hit the upright branchnes. Maple, as expected has
a high amplitude (intensity) and
since the foliage is on the outer surface of the crown envelope,
the width is narrow. Single pulses tell, however, only a little.
Oct
14 SAC SSB & ALS60/WDM65 Waveforms.
Sat-Sun I was priviledged to participate the 52nd Scandinavian
Activity Contest from OH2BH's QTH in Pusula. Here's the report.
OK'ed to do two peer reviews, both
in LiDAR (DEMs and CHMs). AH got the DEM versions that we pending
and we decided how to model the crowns of the trees in the upper layer.
Familiarized myself
with the FWF-data and programmed an application that splits the
data into 1-hectare files, separately for the discrete-return (fixed
four returns) and FWF data. The FWF-
data are simply 128- or 256-byte records. Each DR-pulserecord
holds a file offset (address) linking the waveform (if recorded) and
the DR-pulse. Also, the time distance
between the DR-returns and the start of the FWF-digitization
(in picoseconds!) is stored for each pulse, so that the 3D path of
the waveform is traceable. The VB-applications
that does the I/O and data splitting is slowish - with 69 Gbytes
of data it'll take some 60 hours of processing time. Faculty meeting
on Tue. More dark clouds above research projects?
ALS60/WDM65 Waveforms of
2-or-more-return pulses. Color depicts amplitude. 2-return pulses are
seen e.g. at the edges of roof.
Waveforms of 3-return pulses.
Sparse birch crowns and edges of crowns (tips of branches) produce
such backscatter.
Waveforms (excluding near-ground data) with more than 1 echo
(return) are at the edges of crown extent as can be expected. The Nadir-view
(from 2004) on the right depicts this.
Oct 8 Preparing applications,
talks & teaching
Participated in an ESA-project application preparations. On Tue,
in Evo with AH to give the lectures
and the kuvamitt-demos. Wed-Thu prepared the 45-minute talk that
was
given at NLS on Friday. Almost 120 participants in Pasila (Link
to News).
StepIR went up on Wednesday, finally. KÖ was there to help.
ACOM and GU74b's arrived on Tue and provided a mood boost. PY4XN was
the 1st QSO on 18 MHz.
Oct 1 Dancing to the Administration
Polka - Mammuth out again
This was the get-the-administration-done week. Chased after invoices,
travel bills, salaries etc. to see where the projects are, economically.
It seems that all companies
from which we've bought anything have received their money thru
the university system. Unfortunately it seems that some extra money is
going to be left unused (2010), because
I was not able to spend it not knowing where the projects were
and because I managed to get the consortium running for the airborne
data acquisitions, which saved our funds.
I just hope that the financial administration by the University
would work better in the future. This year has been catastrophic at all
levels of administration, it seems. Projects pay
for a service but the truth is that you still should keep a 100%
parallel bookkeeping and scrutinize and correct the official bookkeeping
based on that. Also, you need to be an
accounts ledger and see to it to that people that you buy stuff
from have actually got their money. When applying projects, you don't
know what you can promise the funder
budgetwise in your project-plan, because it is still unclear
what "taxes" the university puts on your project, ranging from about
70 to 25% of the total budget (money received).
It used to be 10-20% before the start of 2010. This year everything
is with VAT, next year we are back at VAT-0%, because that's what we
had for dedaces and I guess such
traditions are impossible to break. It is only fair to tax external-funding
research projects for the costs that they cause (a share of the infrastructure).
However, the basis for these
'taxes' are unclear and it seems that differentiating between
the basic stuff (directly taxpayer-funded) and the rest (indirect tax
money, real external money) is challenging
(but not openly admitted). Somehow I have the feeling that whatever
we now have at UH (or other public universities), isn't the most effective
way. We had high hopes for the
new University law (Jan 1, 2010), but it seems that the same
shavelling of dirt continues and departments (heads) are in between
the wood and the bark, as are other institutions
above it. I don't know where the last layer of bark is...
Prepared for a day's computer
lab -teaching in Evo next week. The IT-person in Evo had one PC tuned
for KUVAMITT and an image of the HDD was copied to the machines
in the computer lab (temporary). All this took less than a day
of his working hours. The same lab was organized in UEF (Mekrijärvi)
in 2009. There we had to work a bit more because the
installations were done machine by machine. We have to make it
possible also in Viikki, at our "GIS/RS -lab" (1 Gbps + HDD-image based
temporary installations or eq.).
Waveform data was received again - now chopped into max 2 GByte
LAS-files.
Faculty board had two meetings (Tue, Wed) this week.
ESA application preparations caused some extra bandwidth.
Mammuth was submitted finally on Monday evening to RSE.
Promised to participate in the voluntary evaluation activity
(research and post-doc training at UH).
We discussed the options for doing the snow-damage mapping (with
MV). LiDAR data sets need to be prepared first (2010)., actually a.s.a.p.
TBD
- DR-data 2010 into 1ha files.
- Oct 8 seminar.
- Trees and DEMs for AH.
Sept 23 First waveforms
While finishing the ADS40 mammuth, started to have a look at
the FWF-data, because it will needed for some UAV-validations. The
LAS 1.3 format uses
huge integers called "unsigned long long", which allow the files
to go above the 2^31 or 2^32 byte limits. Well, my old VS6.0 of course
would not support
_fseek64(), which can be used for setting the file pointer beyond
the 2GB wall. They saw it better to split the files for me, knowing
that I'm too lazy to learn the use of new compilers.
This is the first 'visualized' ALS60 256-sample waveform in Hyytiälä.
A start if not a grand one.
Unix was down for 6 days and it was faster to transfer 7 GB of
data to Evo by snail mail (or update this file). Testing awaits. Aarne
showed a cool Java application. Must model the
crowns for him for his ray-tracing tests. On Thursday tried lecturing
in Swedish for the second time (pdf).
The mammuth was made another 10% shorter and more concise thanks
to local and remote help. Mosaic Mill got the GCPs for the UAV-blocks
that were derived
in the frest UltraCAM Xp 20-cm images. Natural targets and signals.
Submitted an abstract to the Nordic Remote Sensing days in November.
Hope to g et the StepIR up finally.
Sept 14 Peat is soft
PileUP! got ready late Sunday with some last-minute editing.
Two days later Google Analytics shows 1058 visits by 865 people from
61 "Countries". It seems that
the 'bush-radio' isn't really working. Sent some msgs to contest
club webmasters to boost reader numbers.
In Hyytiälä on Tue to talk (pdf)
and see the wood procurement experiment in Vuorijärvensuo drained
peatland (old cottongrass-pine-bog, now 135 m3/ha of 50-60-yr-old
pine) by Metla and Metsähallitus. LiDAR-based estimates
of tree volume (maps) and elevation models were tried / tested for their
ability to predict the 'carrying capacity'
of the terrain. Even after the very dry July, the machine hauling
the wood loads (30 tn?) would dive into the peat and the experiment was
not completely done. It remained
somewhat unclear what are the primary causes that explain why
in some places there were no problems, while in other places there were
40-cm-deep tracks. But
undestory trees and their roots, near the ditches would be an
option (placement of the roads). One more reason to estimate the understory.
Sept 10, Time flies when busy,
too busy
This week involved a lot of text editing, that is pruning and
hairstyling of two the Mammuths - CCF's PileUP! 14(3-4) and the joint
ADS40-paper.
PileUP! has LX2A's interview (antenna farm), the front cover
has Olli's QSL from The Gambia. ADS40-'BRDFs' of pine, spruce, and
birch.
68 pages plus a 5-page supplement pdf. All this despite a decrease
in the font-size! Esa OH7WV and Kim OH6KZP provided with help as well
as many others
whenever asked for. Great talkoo-spirit once again. I think that
if we add up all the voluntary work of the writers and editors, we are
somewhere at 200-300
man-hours. Some were known to travel to scan old documents and
pictures in libraries and neighboring towns - even. There had been
plans for this kind of
special issue since last 2-3 years, and now we have it. With
a double-issue at our hands, the editors and the talkoo-folks can now
rest for a while.
The ADS40 reflectance patterns were estimated for the main tree
species in Finland, separately for the (sub)crowns that are sunlit and
for (sub)crowns that are in
diffuse light, i.e., shaded. THey were estimated using the view-geometry
of the ADS40 (+/- 34°), and using the target-reflectance data by
the XPro software.
Calibration in it is based on the dark-pixel method (the estimation
of the path radiance, i.e. the proportion sky-scattered photons in the
pixel values). Some
interesting species-spesific differences in anisotropy were revealed,
and these can be used for the optimization of the classifiers that discriminate
pine, spruce,
and birch from spectral signatures in aerial images. In retrospect,
we should have flown differently in 2008 when we had the ADS40 in Finland.
Instead of
flying 4 altitudes, we should have done just 2 or 3 and by using
flight paths having 30-degree offset in azimuth. Now we have too little
variation in the difference
of the view and sun azimuth angles, and the observations were
slightly contamined by the height variation.
Last Sunday I took my camera to forest to document variation
in tree structure.
The high within-species variation in the reflectance values definately
has to do with structural differences as all needles and leaves are
green! Spruce samples in
Kokonniemi, Porvoo, 1) clumping of needles (shoot shape varies,
comb-like, brush-like), 2) Shape of the envelope the tree top. 3) Sparse
crowns, probably trees that were
exposed to a lot more light at some point (cuttings).
A 60-yr-old pine growing on barren site. 90-yr-old pine with
a wide, asymmetric crown, Older pine with dense foliage in the company
of spruce trees.
The leaf density in birch has a huge variation. No wonder our
signals show a lot of variation, be it monostatic LiDAR or image observations.
September 3, Directional reflectance
anisotropy in trees - To Be Or Not To Be...
Converting the azimuth difference (0..180º) and view angles
(0..34º)
of the ADS40 data into the polar representation of BRDF showed that
actually our data did not provide a particularly
poor sampling. There are poorly sampled regions though. I tested
with simple polynomial surfaces, which were fitted to the reflectance
data and computed difference surfaces that
depict the similarities and dissimilarties in the 'BRDFs'. The
example above shows (BRDFpine - BRDFspruce) and it reveals that in
NIR pines are brighter than spruces in the forward scattering side (negative
principal plane zenith), while the order is reverse on the other side.
The difference ranges +-10%. This simply says that near the nadir,
NIR separates poorly
pine and spruce, but that chances are better off-nadir. We still
have to work with the low-order polynomials to be sure they aren't oscillating
and causing bias. The whole week was spent
redoing the analyses, drawing new figs and trying to understand
what the data says about the BRDF patterns in pine, spruce and birch,
and how much can we generalize the resuls. Now
it looks as if different strategies (variable sets) are needed
in different parts of the view-illumination geometry to optimize tree
species classification. The fig. on the right was made to
illustrate the sampling of the crowns in the images: white =
sunlit point, yellow = self-shaded, blue = neighbor-shaded, purple =
neighbor- and-self-shaded, green = point occluded by a neighbor. 5 hour
meeting on Thu. Agreed to go the Evo in Oct 5 to teach and in November
to talk, which wasn't wise given the NLS-talk in Oct 8 and Metla-excursion
in just one week.
PileUP! is right on top of this all...Too much hastle
August 27
New start with ADS40 paper.
Apart from some administrative work and meetings, the whole week
was spent improving the ADS40 sp-classification paper. Analysis of
the importance of relative
height revealed that it explained 10-25% of the within-species
reflectance variation!! Much more than the other factors that were
thought to be important. Some discussions going
on as to how to call the entities that we observe in the pixels
of the reflectance images. Textbooks edited by Schönermark et al.
and another by Jones & Vaughan
discuss the ambiguities involved in BRDF-research. Finally I
grasped how to transform the off-nadir and azimuth-difference angles
to produce the kind of polar
plots that you usually see with BRDF. However, our data is so
noisy that the 2.5D scatterplots reveal nothing, some sort interpolation
is needed,- have to check this as
well as the single-tree study where the same tree (and its neighbors)
are observed for anisotropy. If the patterns match and noise is small,
then maybe there is
something to do in multianugular data and anisotropy estimation.
Also, it would be interesting to see if the classification results
improve by applying the correction
on the relative height, which was shown to be a major cause for
the high within-species variation. This would be new stuff to science.
Mean reflectances of sunlit tree crowns in the blue band for
frontlit-backlit trees (~ principal plane +- 30 degrees) and for sidelit
trees (across the principal plane +- 40 degrees).
August 20
Post-Holiday Existence
Started work on the Mammuth; it needs a complete revision. Slow
and difficult process. Tried to learn the basics (major factors) of atmospheric
correction to include that too. Luckily
UB helped and I should receive some decent litterature (read
easy-to-read) to go with his tutoring. Shortening of Mammuth wasn't
very successful - the first week saw
the reduction of 1.5 pages out of the 20. AH worked on the documentation
of the field campaign. I promised to do some presentations in
Sept in Hyytiälä for the wood procurement
project by Metla that is experimenting on Vuorijärvensuo.
This overlapped with Silvilaser 2010, so those plans were set aside.
I said "ja" also to some requests to do lecturing
in Swedish.
We visited Finnmap to collect the FWF and image data from
last summer. Everyone was on good humour as the project went exactly
as planned. This isn't the rule in the business of aerial photogrammetry
and laser scanning. Only one drawback was revealed, the full-waveform scanning
at high PRF, using MPiA, resulted in every second pair of pulses being sampled
- not every second pulse. I had a visitor (LR from Brasil), who taught
me inventory of Eucalyptus plantations, clonal and natural, as well as
rain forests. Interesting opportunities for LiDAR in those conditions.
This week I learned also that aspens (Populus L.) assimilate with their
bark. Determination of the LAI in such cases, or phenological status in
the spring, isn't trivial. Highlight of
the week was when UH extended my disk space on this server to
2GB. I could have continued the holidays.
August 16
Hirvensalmi-Syväsmäki-Mikkeli
trip.
We were
chasing fish and Botrychiums (Moonworts). The latter had a reward (5€
each spp !!.) on them. Syväsmäki abandoned football field
(N 6835790 E 497720)
finally produced
a great finding of the two species making the boys happy, finally.
B. multifidum
was abundant in about 50 examples and B. lunaria we saw 4 times.
We searched also many other dry meadows and field edges in Hirvensalmi
with no luck.
The pale B. lunaria's
were easy to spot, while it took some effort to distuingish the B.
multifidum's. The site has 0-1-m high alder-birch-pine-willow vegetation,
and the places to find
the moonworts were in the southern end, where it had been moist
enough. Surely, during a less dry summer, there will be Syväsmäki
Botrychiums all over the the old sports plan.
Hirvensalmi is not that
far from Mikkeli that hosts three war-time or military museums, being
an old garnison town. In Mikkeli, it is compulsory to visit the Siiskonen
cafeteria. In the Infantry
museum, the boys were boys again. One room in the museum was
dedicated to the Mannerheim knights, and I spotted the legendary Einari
Schadewitz, who was familiar to me from Hamina times in the 1970s.
The famous Lauri Törni was there too as well as Arvo Pentti.
Lake Puula with extremely clear waters. Our rented home. The
dart-targets. The lower onewas more motivating for the boys.
August 6
Waiting for the UAV, and delineating new reference plots in the
anticipated areas to be flown. Aarne stayed in Hyytiälä to do
the VRS-GNSS. The
Mammuth peer-reviews came finally back and it has to be re-rewritten
into a shorter article. This work will wait untill after holiday-week
in Mikkeli. Leaving Hyytiälä
for this summer with a total of three weeks here. We agreed on
the kind of "DEM-samples" to be taken, where we estiate the height
of the low vegetation
and its canopy cover, in %. In all, the VRS-GNSS work covers
a) rectangular plots 64 - 200 m2, where all trees are mapped and measured
for height,
b) samples of pure vegetation with one species in 0.5 - 1 m2
circle, and c) these DEM samples, where the species mixture can vary,
but he cover-%
and height are measured. These latter data are used in studying
the fwf-response per species and effect in Z of the CC.
August 4
Tue and Wed participated in the mapping of the snowdamages of Feb 2010.
We are mapping some 4 hectares of plots that are 'high' in elevation,
which is above 180 m in Hyytiälä. The snowbreaks seemed
to occur more frequently near canopy openings and wood procurement roads.
This observation
was valid only for one 0.7-hectare plot (Pätsäinmäki).
We did not map the position of the broken tops to figure out in which
direction the trees were bending.
Metsähallitus and UEF promised support for the 2010 airborne
LiDAR and UCXp campaigns in Hyytiälä, which was really nice
news. The canopy-cover
estimation paper was returned with positive comments. Promised
to give a talk at the NLS LiDAR meeting in October. Not very good news
from Salo. Reviewing
SVM paper by VH.
July 31
Joined Aarne for a week's session of field work @ Hyytiälä.
Tried out VRS-GNSS measurements in two 10x20-m plots with 0.3-5-m high
trees on Sunday.
Built "signals" (control sticks) for the Social Forest Inventory
project's convergent photography of forest plots. These were wooden
sticks with BW markers at the ends.
Three sticks to be installed in vertical (plumb line markers)
and two longer scale bars to installed in arbitrary position (see fig
below). On Monday did some photography
in Muistokuusikko, Lapinkangas (#2) and in plot M1_09_P2.
from 2007 that seems to be working still. In the forest, the
trees are photographed using convergent photography, from 5 (in our
case) locations, and the camera is kept
pointed to a target in front of the middle (Photo 3) camera.
The figure above illustrates the idea. If the camera positions are
along a line (which is called the X-axis), the
Y0 values (camera positions) can be 'fixed' (close) to zero.
If the camera is looking to the horizon, values of omega are ~PI/2 and
kappa are ~0. The phi roration about the Y-axis
takes values around zero. Trees can be used for measuring tie
points that are vertically aligned and the XY-similary constrain between
these points is imposed in the
image block adjustment (triangulation). The scale of the ray-block
is determined by the accuracy of the camera constant and the 2-meter
interphoto distances that are
taken simply with accurate steps, in the simplest case!
were 33.2, 31.7, 31.6, 32.4 and 32.3 cm (average = 32.24, missing
1.5%). These images were taken such that the painted 1.3-m marks are
visible to the cameras. Now, if the markers
are not there, the ground in front of the stem has to be visible
for a Z-gnd measurement. The 1.5% underestimation was not visible in
the "control sticks", which
had "photogrammetric lengths" equal to the nominal lengths (see
below). Interphoto distances were taken by foot.
the branch. Tie points with XY-constrain are needed to solve
Z0-values reliably (and to have the 3D scene in the (up)right attitude).
Stems have to serve this purpose, when
nothing artifial is placed in the scene.
July 27 Got the solution working for the image
qvintet orientation with the assumptions on camera positions. In Salo.
July 23-24 Aarne finalized the UAV flight
plans / selection of seedling stands, returned the tarps and fetched
the VRS-GPS to commence the measurements on the
seedling stand vegetation for the UAV and FWF LiDAR. Signals
were also built for the UAV areas to be flow in early august. The mapping
will follow the 2006/2007
proceduredures. Signal positions for the large AOI from 2008
in XLS.
July 21-22 Helped Aarne with aerial maps. The
WBS-code finally arrived for the Social Forest Inventory project. Visited
Finnmap to see/discuss the data processing. The DR data comes in LAS-files
with trajectory and the FWF-data is in extended LAS 1.3 type 4 format
(tiled to max 2GB). Boresight calibration had been successful as well
as both campaigns in Hyytiälä and Evo.
Mon July 19 was a busy day on the
phone as the FWF-LiDAR took place in the evening in Hyytiälä
and in Evo. Aarne & co. installed the 5 tarps in the field. Scattered
clouds
enabled the flying at 1 km from below the cloud cover and when
the 2-km and 3-km strips were due, clouds were mostly gone. The wx
was turning
unfavorable and the timing was just right. In Savonlinna, where
we were, the cloud cover did not disappear completely. Let's hope the
data logging
was succesfull and flying resulted in seamless data. The installation
of the als60 was fast in Tampere.
July 15 paid a visit to Hyytiälä
and surprisingly the hypersectral imaging by Pellikka et al. was ongoing
in the cloudy weather.
July 9- On holidays with Ellu
& boys...Low-cost-travelling / tourism in Hamina, Helsinki, Härmä,
Ähtäri, Seinäjoki, Savonlinna. On the way back from
OH6 on Thu,
July 4-7.
ISPRS C VII symposium in Vienna. We had 2 talks on Tue, one by LM,
while I gave the other. Several talks on ADS40, FWF-LiDAR, and Tree-Sp
classification.
The slides in PDF.
The 6-page article in PDF.
Consider preparing the ads40-dataset for "BRDF-work". FWF-software
in TU Wien. Ask about LAS 1.3 limitations concer-
ning pulse reconstruction / time-stamping.
July 1-2: #4 of the MARV1
campaign saw last measurements being taken and the data was checked
and delivered. An amazing 2355 trees were mapped (accuracy 0.24 m)
and tallied. The total area covered by the 13 plots was 2.62
hectares. Day #5 on Friday was just for some feedback and acnowledgements.
In all, the operation took 3.6 days
to complete the data. Was too busy with the course/campaign to
participate properly in the 100-yr-jubilee at Hyytiälä. No
rain at all in the 5 days.
The fwf-LiDAR campaign:
The sonde-campaign by SMEAR-reserachers will result in NOTAMs at the
air control, and the meaning of these, with respect to our LiDAR
campaign (and the low flying campaigns by others) was under investigation.
We now have all the partners informed and should be able to squeeze
in the flights between the brightly-colored balloons. Usually, "the-more-the-merrier"
saying holds true, but it seems that the increasing activity at SMEAR
II will require more careful coordination.
After successfull completion
of the campaign it was again time for photography! World Cup 2010 in
South Africa was ongoing..
AH will check the obscurities at the MARV1-plots,
arrange the painting of the 1.3-m marks with the station labour, measure
the MK-plot, and prepare the field-campaign for
the snow-break damages, complete measurements of the dimensions
of the small understory trees (for crown modelling), and act as a contact
for the fwf-LiDAR cam-
paign/HUMPPA project at SMEAR.
Notes: The How-to-draw-the-treemap -instructions needed to be
reformulated (for Excel 2007). "Fotopuuluokka" is missing in Hikihelmi.
Remember to claim: Sun-Fri -trip / travel costs.
June 29-30:
Hectic days #2 and #3 of the MARV1-campaign. All went ok apart
from the fact that 2-3 subplots were too demanding. Some obscurities
in
the dbh-measurements (single vs. double directions). Learned
that in July some weather balloons will be launched from SMEAR... Need
to figure out when,
by whom and how it might affect flying above Hyytiälä.
MM successfully completed the first cam-photo campaign.
'
I payed a visit to see the "TLS boys" on plot 4A. Wed evening
managed to do a bit of botany for entertainment.
Silene rupestris was abundant at Makkarakallio. Sadly, Lupinus
is spreading across Tapiolanmäki and killing everything - here
Platanthera. No actions seem to be taken to control it's spread in Hyytiälä.
Huperzia selago at Makkakallio..
Orchids
of Tapiola - Pilvilinna. Haven't seen the Coeloglossum in 10 years.
Listera ovata stand is only
5 m away from the spreading Lupinus. Old sheep pasture (from 1994),
June 27-28:
In Hyytiälä for the MARV1-campaign
and Social Forest Inventory CAM-photo session. UltraCAM XP aerial photography
took place at 13:15 GMT at short notice and
we did not get (had no time) the in-situ radiometric targets
and measurements going. Visited the MARV1 plots on Sun evening and
the lectures & basic training got started on Mon. MM arrived to
do the CAM-photo observations at the prepared targets. Everything worked
smoothly on day #1 - let's hope all goes well. At least the WX forecast
was really good - maybe even
too good WX? Brought some posters to Hyytiälä for the
100-yr-jubilee, on Thu-Fri. EN announced that thru Janne L we can have
some field labour for the snow-break-damage mapping.
AH started to prepare the maps etc. for that task.
June 16-22:
Munich-Muhldorf-Landshut-Zurich-Heerbrugg
-trip. Visited three universities (Munich, 2 x Zurich) and Leica
to see/meet people and technology. FR's seminar for the MSc thesis
was on Monday at ETHZ. TBD (check the template-approach in classification,
send pdfs to FM). Meanwhile AH finished the Social Forest plots and worked
on the MARV1 plots - the course starts right after midsummer.
I still could not get the project administration going because the service
center had no personnel to talk to.
<>
Heerbrugg railway station. Integrating sphere for radiometric
calibration of ADS40 with green leds on. Waveform digitizer board in
the ALS60 LIDAR system.
ALS60 without cover. AGC circuit board is at the far end.
A gyro-stabilized camera/LiDAR mount under construction. Focal plane
of an ADS80-SH82 with filters visible.
Tourism: Massive Mettenheim Bunker ruins from WW II in Bavaria
(for building jet fighters 1944-). Capital of lower Bavaria, Landshut.
BMW museum in Munich.
<>
From Landshut in Bavaria I got some stuff to bring the boys at
home to go with the World Cup 2010. Good WX during Juhannus in Tevaniemi,
Ikaalinen.
June 13-15:
Hyytiälä. Visiting the field experiment to assess the snow
damage in the forests, and to get the Social Forest Inventory project
started with AH and
MM. The general findings were: p(damage) was highest in pine
with mostly broken stems but also fallen trees. Variation, but slender
trees in dense stands were
affected most. Plots MARV1-08-1 (recently fertiziled) and P4
(unthinned) had upto 30-40% of the trees damaged and were unrecognizible!
These plots were 180-185 m
a.s.l. and probably prone to winds (with the snowload little
is needed). Below elevations of 160 m the damage was much lesser (<
5%). Broken spruce trees were seen
too, and bent birch trees.
We decided to map the high-elevation pine plots with the help
of JPs field workers. Preparation
of the maps & tree lists needs to be done. This data may
be of interest
for anyone mapping the damage of winter 2010 or trying to understand
the reasons behind the varation in the damage.We also discussed with
growth and yield
Prof., who said that the impacts in growth on the remaining trees
will be of interest. We will be using the LiDAR 2008 and 2010 to assess
the damage in the whole
Lake Kuivajärvi area.
We visited the SFI-project plots with MM to show him the kind
of variation present in the selection and to discuss on the principles
of the selecting the views for
the cam-photo experiments + markings. The camera used by UH was
tested for the simple forest-owner photogrammetric documentation. The
FOV is just 52 degrees which
keeps the views narrow (concering usage for Bitterlich samplng).
Convergent photography of known targets was used to test the calibration-
see below.
We also visited the MARV1-2010 plots, and realized that some
plots need special attention in the setting of the minimum DBH to prevent
a situation where the students
choke on too many trees to be measured (plot 4, 5). Finding the
plot corners can also be tedious, so we decided to do it by ourselves
beforehand to avoid chaos on day #1.
We discussed on the option of having the 1946-2010 aerial images
converted into orthomosaics, because the current version of KUVAMITT
(@ Hyytiälä) is not supporting
the different kind of images produced by the modern digital sensors,
and the users are not keen on photogrammetry. Seek funding for the
work? Was home 15th @ 17. Claim!
June 9-11. Tried to arrange funding for the required
aerial imaging of Hyytiälä to map the unexpected snow damages.
Wasn't very nice experience but finally got
support from Kirkkonummi. We try UltraCAM Xp 2.5-km flight with
80/50% overlaps. On wednesday I heard that we will give the fwf-LiDAR
a try in Finland, which
was good news for the consortium. Joint planning of the flight
details started right away in order to reach a consencus on them and
make them feasible.
AGC-issue remained as well as the dates.
On Wed visited MMill and FMint to discuss UAV and fwf details.
During the week, AH got all the preparations of the field plots
done for the Social Forest Inventory -project.
Agreed on TLS work at the MARV1-2010 plots this summer. Also,
agreed on joint field work in Hyytiälä with JP's group - if
needed. The VRS-GPS was
reserved for (rental) the August field work in seedling stands.
FR's thesis analyses showed interesting results (and illustrations/visualizatios)-
now remains finalizing
the writing. Made arrangements for the Zurich trip to meet RS
specialists at the University and ETH.
Got some idea of the budget (economical situation of my projects),
but tolerances are still around 10%. Maybe this is how it is in the
new university? I really
hope not.
June 7-8.
The TEKES-project Social Forest
Inventory was started and our initial plans on the activities
at the field plots concerning the Mobile-cam-photo experiment
were confirmed @ VTT (Mon). The snow-break damage in Hyytiälä
(devastation) was worrying a bit since we have not yet received "analytic
reports" of it or witnessed it by
our own eyes. It can cause and will cause extra grey hair and
the 2009 field work might be in part out-of-date now. Rumors say that
the devastation has hit higher
elevations above 180 m a.s.l., which is where I have a lot of
field trees. One option would be to still try to acquire high-resolution
aerial images over the site and
manually determine the lost crowns in the images rather than
doing another field campaign of 20 000€+. Really annoying, although
excellent for anyone interested in
change detection! Last attempts to put forward the fwf-campaign
on Monday. On Tuesday AH and I commenced the pre-forest-preparations
for the cam-photo experi-
ment, which included selection of the candidate plots (to hold
80-degree, or panoramic views, meeting selection criteria in species,
density, age, topography, and occ-
lussion by understory) editing the most up-to-date field measurements
(plot data) into CSV-files that can be inputted to RESECTION and PDFTREEMAP
programs,
which were both adapted to a new kind of input structure. The
main difference being that tree numbers (identifiers) are now treated
as char[], and the pdftreemap
plotter now determines from the XYZ, and dbh-data the proper
scale and orientation of the map, which were earlier tediously entered
in the source code. The UAV
flights need still further planning so as to be most efficient.
The paper work for the project was done 50% (all local dept level
work was easy) with central administration folks being unavailable. I
still think it su##s that a PI has to know all (many) the bureaucratic details
of the project admistration,
even while he/she pays 112% overhead for that work/service also. The
whole idea of placing the project administration people in one room
(palvelukeskus) in the campus is not a good idea. These people
should be spread across the campus close to the people that they "serve".
It is 5 months
from the start of the year, and now I heard a rumor that it might
be possible to get a report of the financial status of my projects. Some
PIs go about and ask
their workers for their "salary tickets", to be able sum up the
salary costs in their project (whcih might be ending now), which they
cannot get from any service/system. Not even if they pay the #-112% extra
overhed. Who'd use the Big Sweep?
June 4
A week of planning, proposing and
reviewing
Programmed the TIFF2RAW programs needed for the RGBN and PAN
tifs (12 bit) of the DMC. The inner orientation requird a 1.5 pixel
vertical offset (see Fig.below) to work
in KUVAMITT. Planning of the Social Forest Planning project's
field campaign commenced. Reviewed UBs article and FRs & AHs thesis
work was consulted. Agreed
to review an EJFOR article. Wrote an initial plan for the FWF-LiDAR
campaign in Hyytiälä, should it see daylight. Will be travelling
to Bavaria and Switzerland
in late June.
In DMC the origin of the pixel is in the center of the pixel,
which is 0.5 pixels different from KUVAMITT. Also, KUVAMITT shifts everything
by 1 pixel because of a definition in the int.ori. functions. Image on
the left has 20 cm pan-sharpened pixels while the image on the right is
the ms-image in 80-cm resolution, from 2 km agl.
Many new cameras apply the PAN-sharpening principle, where a
panchromatic image has a higher resolution, while the resolution of
the other bands is lower.
On tuesday we had the great pleasure to visit Ratina in Tampere
with Teemu who drove 450x2 km from/to Oulu. Three hours of waiting gave
us good view and hearing wasn't a problem.
May 28 AH is making good progress with the LiDAR
- understory tree -detection study. Started working on the DMC/UCD/RC30/ADS40
comparisons.
The DMC image formats need special attention - the are the 12/8-bit
versions, with and without LUT corrections. It is advisable to consult
the camera users
in Pasila-Jyväskylä. The IO/EO does not apply to all
image versions. Learned also that absolute calibration for ASR data
is now available for the DMC sensor,
not that we have it for the 2009 data. Planned the feature extraction
to take place for all 16000 trees per flying height in the different
cameras - storing the image bits
with similar data structure as with the ADS40, but in random
order as the same tree can map to 4-15 images per flight altitude (variable
length files). Silva Fennica was sent the long
abstract in Finnish and corrections (by HO). Sat 3 hours in the
institute board meeting on Thu a.m. and Metla-parkano was sent the 2004
LiDAR from
Vuorijärvensuo (not that these two events sit on the same
branch of boredom). MARV1 package was delivered for Hyytiälä
transportation.
s
PileUP! 14(3) was announced to be another theme issue - this
time "The good old days". If that theme doesn't allure writers
then what does?
TBD:
- Growth -study analysis.
- DMC/UCD feature extraction - DMC data analysis - IO/EO definitions
for different image versions. Solar angle data for all images.
- Preparing for Social Forest Planning -project activities.
There was a drastic change in WX type after the warm period.s
May 25 Got the 6-page ISPRS-Vienna paper done as well as the
powerpoint slides. Reviewed the ads40 multiview manu for VH. SF proofs
arrived and need
attention. TEKES decision was received and it was positive and
so we have funds to do proper field work in 2010 in support of the FWF
and UAV
campaigns. MARV1 plots will also be measured using TLS by Markus
et al. (during the course), which was a nice addition to the setup.
Exceptionally warm weather in the last 10 days - (Rosaceae) trees
in flower and lawn mowers out of the carages in Kulloo.
May 21 Finalized preparations
for MARV1-course in late June. They took in total 13 work days. All
maps, tally sheets, instructions etc. were printed and
maps laminated and the web-site finalized
with all data and files accessible. Preparations for lake Kuivajärvi
LAI-estimation. Ville sent the ADS40 multiview manu.
May 18 Preparing for field measurement
campaign in Hyytiälä cntd.
Muistokuusikko training site & Pehkusuo pine bog plot.
May 13 Spring arrived in Kulloo, while Jari defended his thesis
in Joensuu. Peer reviewed for IJRS and FORECO.
PileUP! 14(2) was released.
May 5 Finally, after 1 year and 8 months of hard work by lots
of people, the ADS40 research article was finally submitted.
VH continues with tuning of the sp-classifications using state-of-the-art
classifiers and multi-view predictors.
Began preparations for the MARV1-activities in June. ISPRS JPRS
proofs came in too.
Active with PileUP! 14(2) with others.
April 28 error in substantia..
Correcting the manuscript about ADS40 based
on feedback. Finally, on Thu, realized that I had
been completely wrong in assuming that CV of reflectance by species
would reduce when the effects by the atmosphere are
eliminated. For weeks, the others tried to convince me on the
fact, but I guess I was too stubborn in admitting that I had not
come up with a grand idea.
TBD:
- Job
- IJRS peer-review
- Marv1-plots 40/3 ~ 13 groups.
- SF abstract in FIN
- Growth -study analysis cntd.
- Vienna -paper
April 24 A week of planning
activities and meetings
A week of planning, meeting with people and organizing the summer
activities. Mon-Wed tried to bring forward the LiDAR-tree-growth
study and Thu-Fri worked on some corrections to a manuscript
by the co-authors.
April 17 Tree growth indicated by the LiDAR signal
Wrote a new version of the TRISECTLAT-program to enable any number
of control points with distance and azimuth
observations. Measured also the treetops (control points) as
plot LK3 for AH, and he reported improved XY accuracy from
17 cm to 10 cm for the seedlings below the 25-30-m-high canopy.
Prepared observations files for the 5800 trees with id
or ir growth observations 2002(05)-2009 and LiDAR
features similar to
those in our previous Tree-SP study. One additional feature,
PulseRatio, was defined as the proportion of pulses passing
thru the tree crown volume and pulses reflecting back (stopping
at) from the crown volume. We began analysis to see how much LiDAR
features predict variation in the growth measurements (tree vigor)
in pine, spruce, birch, and alder. Stepwise regression
analysis of growth observations with and without trend removal
was attempted. By trend removal we mean the effects by
tree size and crown length which are known from field measurements
and which explain 30-70% of the intertree basal-area
increment variation. It seems that at stand level, LiDAR intensity
and relative height features can explain 20-60% of basal
area variation, but after trend removal the R2 started
to approach 0. Quite interesting to be back in tree growth after almost
20 years when did MSc-thesis in the theme. Quickly remembered
the inherent problems of causality and multicollinearity
in statistical approach to growth modeling. SH was working from
UQAM (Can), while the rest of us in Finland we scattered
across the country. Just shows how internet can help to avoid
traveling...
April 10 A week of peer reviewing
and writing revisions
Reading and correcting articles by LK et al. and JV et al. and
did the revision #2 for ISPRS with HO.
Sent the ads40-article for co-authors for their revision. Went
back doing the "LiDAR-TreeGrowth" article, found the data and programs
from Oct.
Send finally the "Vuorijärvi-LiDAR-bins" to Metla/Parkano.
0.7 x 0.7 km2 in 3 campaigns.
Highlight of the weak was seminar presentation on Friday (pdf)
about the ADS40-species-identification study.
We went to see the flooding in Salo.
PileUP! announced a call for contributions.
April 3 Last skiing this winter
Tue-Thu, 23-25 took some holiday to enjoy the last skiing, which
was over in Porvoo in the WPX SSB weekend.
OG7X (@OH4A by OH6LI). OH1WZ and OH6KZP in the very small ham
shack. Li-Connections. Low-cost rotator...
CQ WPX SSB 2010 M/2 three-op. operation. Photo Jukka OH6LI.
Cntd editing the LAI-article for LK et al., which was done April
2. PEER reviewed for IJRS, but not for
Sensors,
who
send me an article that had the appearance of being "in print".
On would assume that an abstract is sent first to the
candidate reviewer, who, based on it, decides if he/she does
the review.
March 26 and 29 did preparations for some sensor fusion teaching
in Evo, which involved a days session of lectures,
demos and labs, which took place on March 30. I had AH to with
me to participate as a student. All went well apart
from the fact that the Word Macro (Visual Basic) examples did
not include plotting LiDAR data in the images, since
I did not have time to find the right function/method in the
Macro language. It differed from VB6. NIce day with motivated
students comprising of senior forest experts from Metsäkeskus,
companies, and forestry schools,
Last week we got one article back and need to redo it completely.
Reviews were hard but right. Started to work on that
right away as well.
March 31, we the PI-meeting
at the Dept, where, now after 5 weeks, we had some reasonable
discusssion about
1) the 112% oversized overheads that make our project bids too
expensive (60% would make sense), and which
are not calculated by Department or Faculty,
but by combining two Faculties and the Viikki campus, with seemingly
different cost-structure and type of research.
2) Coverage of the 20 or 30% budget share that is the Department's
own share, which need to be real money
or tenure hours, making TEKES and AKATEMIA
projects very difficult for non-tenure post-docs and docents.
3) Non-tenure PI hours in the SOLE follow-up system. Who pays
for the 1 month work on proposals, administration and
the 5-10% of teaching?
We showed that 1) might not be as drastic change as I claimed
("50% more money needed for the same hours"), but clearly
worsens our situatoin if the 112% is not correct but oversized.
March 17 Found the Y-chromosome
of Remote Sensing?
This is the reflectance of trees (N=200,000) in Hyytiälä,
in the BLUE band (1.5%-6.5%) as a function of the off-nadir or view-zenith
angle (radians, 0-34 degr).
Trees take two paths after about 12 degrees of view-zenith: some
are in the direct light, some in diffuse light. It is the Y-chromosome effect that
makes BLUE so poor as a predictor
in classifications?
A two-line sensor leaves but few marks in the view-illumination
space. Azdiff is the azimuth offset between sun and camera, and absoffnadir
is the view-zenith angle. Phase in the phase-angle, i.e. 3D angle
between the sun and image vectors, low values = front-lit trees, high-values
= back-lit.
Viljo was ill on Thu so had time to produce this graph, which
shows how much electricity (kWh/Day) our household is consumig. Every
now and then I've taken
the reading in the kWh-meter. Winters 07/08 and 08/09 were pretty
mild compared to 09/10, when we had over 60 days of temperature below
0C in Dec-Feb. I have
some old data covering Jan 2004 - March 2006, which would make
a nice comparison. The summer numbers, 25 kWh per day are produced by
the boiler, and
other home electronics. I'm suspicious about the boiler being
in the worng place (cold basement) and thus making the energy bills high
here.
March 12 Anisotropy, proximity and job applications
week with FW-flavor
Active
5 days of exploring the radiometrically corrected image observations
for tree and crown points. I learned that the
anisotropy effects can influence the observed reflectance +/-
50% in the worst illumination-view geometry and up to
+/-10% variation in mean reflectance occurs, when the view- or
sun-angle distributions change only marginally. The used
instrument measures at an accuracy that is better than 10% so
that real effects were observed. When
this anisotropy is combined with a high intraclass variation,
the odds for very accurate species classification are small, unless
some invariant features are found that are less sensitive. The
intraclass variation was shown to results from age, sitetype,
and other stand effects as well as proximity effects, where the
local neighborhood has 0-30% influence on the reflectance
of the target tree, depending on the wavelength (highest at NIR),
and the neighbor. This finding is quite fundamental and
in a way disappointing, because in a mixed stand the spectra
have even more variation.
Reflectances in pine, spruce and birch trees in the BLU (left)
and GRN (middle) band for the side-lit geometry. Maximal reflectances
(right)
for trees in a back-lit - front-lit view-illumination gradient.
The dots represent 15000 and 12000 georeferenced trees.
Each graph shows 4 spectra for pine, spruce and birch that are
in different colors. The dots are means of 2000-6000 trees per species
obtained from 4 views that represent different viewing geometry:
trees are seen side-, front-, or back-lit and have been sampled for
sun-lit pixels.
March
5 Another ADS40-week
Binary data were sent to VH @ UEF. FeatureExtractor and SAS-scripts
by Felix were working fine and I was producing results for the "Reflectance
of trees as measured in
ADS40 images" article. Twin articles are possible by splitting
and sending in parallel so we could send the general, "feature review"
case separate from the
"Species classification" -paper. Discussions around the UH -
KKM continued. Also, the possible ALS60-FW or eq. project planning
was cntd. Metsähallitus
reporting. Hyytiälä 100-yr Jubilee meeting on Tue,
busy 3 days in July awaiting. Still nothing from Silva Fennica after
2.5 months. Promised RV in Evo to go there for a LiDAR-
Photogrammetry teaching -day.
Viljo, 5 was helping half-blind grandpa Eero, 80 to ski in Salo
on Saturday. Arttu participated in Susihiihto for the 1st time. Great
sunny WX both days.
Feb 28 Snow on Helsinki roofs,
and elsewhere too..
Although our village is not in Helsinki (which is the main concern
of our responsible minister of housing affairs), many, here in Kulloo,
outside
Helsinki and behind the leftovers of Sipoo, seemed to follow
the instructions by minister Vapaavuori, to climb up to the roofs and
through down
the snow that's about to bring down buildings. Following
Vapaavuori's advice, which are, I believe, based on his great practical
expertise in
construction work, I went up on two roofs and Finns all over
the country did the same. Now surgeons are busy fixing legs, arms,
necks, etc. as
Finns rushed, not only up, but down too :). This cancels out
any positive health effects that the work otherwise might have...
Vapaavuori was right. This is our sad "Lautakasa" Sunday morning.
I sure could use Vapaavuori's assistance in moving the ice and snow around.
Luckily we don't live in Helsinki, and we will eventually get the job
done.
Snowload on our Pihakeinu. The ice, when coming down from the
roof, brought down the "killu" ADSL-line. Killu is strong and the weblog
was updated a ok. The snowy winter shows heatleaks, even better than IR
cams. Where heat leaks out, water leaks in - 2-way.
Feb 26 Kokonaiskustannusmalli - the KKM - first
experiences
On Monday was planning a forest
inventory proposal for TEKES (although can't claim the costs anywhere,
travel or hours, officially we have no project preparing funds/hours).
On Tue tried to figure out how the budgets are done for TEKES-projects,
which was really painfull (Not because of TEKES instructions), lacking
logic, and foremost did not impress me at all as to how effectively these
public funds are used at places like our University (publicly funded organization)
in this new way of computing budgets / costs - the KKM. I hope I'm completely
wrong (as a taxpayer) and get convinced somehow of this, later. I think
we need more people in the upper levels of bureaucracy at UH (c.f. UEF
where funds of administration was recently redirected to the grass-root
level teaching and research, auch). I also believe that the current (cement)
state of affairs, in academia, where everyone, at every level, is basically
just looking over the needs in their immediate neighborhood, won't change.
With the increasing specialization, we have less and less Professors with
wider scope and will + skills to develop common building blocks. It is
funny, at UH we have "Service Centres" (list of people and telephone numbers
located far enough), yet most of us, at the grass-root-level, do everything
from seeking funds and doing the administration to the actual work, which
is very educative, but is it relevant?. Someone somewhere again decided
to change the IT-system in administration - I'm sure that new project planners
are allready hired to do the next one. It is like a recursive function call,
without the end statements. Survival - that's the name of the academic
game and only plausible explanation as to why all this is tolerated. Yet,
as to how effective our adminstration is, I don't know. It may well be that
an ever-lasting tension exists between the production lines and the administration
- everywhere.
Supposed to be on vacation Thu-Fri,
but the TEKES application wasn't finalized by some partners but untill
2 hours into the DL on Friday . Well, I got sick so it did not really matter
to work / be on alert those days. Still when signing and submitting
the application, I could not figure our the required adminstration - budgeting
- and I went for peer-advice to learn how these research-projects that
are based on this newish "kokonaiskustannusmalli (KKM)" are really done.
|
This budget-modeling assumes
that UH (any public organization) as a partner, invests 0-30% (0% if
companies pay it all, may happen) of its own resources to the project,
while the rest would be received from the funder/companies committed to
KKM. If the total budget of the research project is 100 units, 44% of that
is reserved for salaries and salary costs. A total of 44% (112% times all
salary costs) is overhead, which covers the infrastructure that, according
to my understanding, does not include for the teaching activities at the university.
How can administration,
lazy hours, or teaching be separated from research,(I have asked
for these figs at the Faculty level and these cannot not be deduced reliaby,
I was told) and is it most efficient that this overhead percentage
(for us 112%) is specified by campus (when all budgets are by Departmets
and Faculties!!!)? If we had very expensive research instruments, they
would increase the costs of (low-cost) philosophic research, should we
be in the same campus! (now we have similar extremes in our campus) 30%
is our own investment in the project. I learned that at VTT it is "real
money", whereas with us it not necessarily is real money, but something
else.
The main lesson learned this week
was that non-tenure post-docs, docents etc. can have great challenges
in getting KKM-projects, because of the required UH investment,
which in most cases, as I learned it now, is tenure hours. I had anticipated
this and asked about it several times at the Faculty board in the last
two years, but there has not been too many non-tenure representatives,
so who'd care, as, it's "not impacting
me". Two years ago I got an answer by the financing experts at
one of these meetings: "With KKM, you'll do 60% of the work you did with
the same funding (cash in)". I wonder where the rest is been blown now?
Or did we just, overnight, became more expensive? I didn't get a raise.
We have a follow-up system and hours are allocated to projects.
I currently have two. Since 1) keeping weblog, 2) attending administrative
meetings, 3) preparing new projects, joint or own, 4) supervising students,
MSc or PhD, are not specified in these projects, I can't claim the
hours or costs anywhere (if I'm honest), so these need to be made outside
office hours leaving no mark of these activities in the follow-up systems
(Sundays). I do not believe that I'm the only one. I 'm also afraid that
the pressures of KKM are pushing people to claim hours and costs to projects
to create book-keeping
costs (cf. "real money") At least, this is what I read between-the-lines
last week when discussing with colleagues here and there. In Finnish we
call it the "Maan tapa".
hope I'm wrong.
|
Some much for complaints, TBD:
- Send VH the new BINARY-files
and latest VB-featureExtractor
- Send report / funding docs
to SLVTS/UH
- Continue FW-LiDAR contacts
- Check the status of the
LiDAR - tree increment project
- Check the bugdets - contact
AH
Feb 19
AM defended his dissertation on Friday, Feb 12 and we had a nice
party that evening. Last week of Felix's intership was spent getting
data and programs in order and ready for shipment to HB9. Reading JVs
work, and the Academy proposal + Faculty meeting made it a week of interruptions.
On Wed we visited Masala with two meetings and on Thursday we were in
Malmi, after which some first contacts about possible FW-LiDAR next summer.
On Friday we organized a short seminar where Felix told us how the internship
went. Some 15 people from the 3rd and 4th floor attended and seemingly
enjoyed Felix's show. He then left the building at 2 pm. The very active
4-month research period is over... I'm getting e-mail announcements that
my work has been cited almost on daily basis. I wonder if that shows
in any index, and if it does, so what?
Perttu, Eetu and Arttu cheering for Lempi. Arttu on ski's. Kullo
Fastlagssöndag.
Viljo, Laura, Perttu.
Feb 11
We found a bug (thanks to the azimuth analysis) in the earlier
kuvamitt_lite computations of the 68 strip_correction_view_RGBNIR files,
where the 12 crown points in 10 heights were not correctly aligned on
the crown surface after the first height down the top (1st one towards
sun, then 30-degrees of rotation between points) due to an index
error in a for loop that caused the sun to lie in east (phi = 0). So on
Friday, we started again the computations on 3 PCs that we did last during
Xmas time, 6 weeks back. They got ready Wed-Fri, even earlier had not the
main PC booted itself after an update on Thu mng :). The revision was returned
to ISPRS, and writing work on the article got started. Felix made his thesis
plans for ETH, documented his work after Oct 22, and programmed the adjacency
analysis variables describing the neighborhood of reference trees. It
turned out to produce interesting results as upto 25% adjacency effects
were seen in the NIR. We completed the list of analysis to be included
in the article and identified few places where some additional programming
is needed to allow the work. Wednesday was spent proof-reading a proposal
to the Academy.
Feb 4
Felix cntd his analyses / programming of "light extinction" and
"solar azimuth". Thu, seminar at MH about the use of new RS techniques
in producing meaningful data for ecological monitoring applications.
Work cntd on the revision. Meetings interrupting.
Feb 2
We planned analysis work for Felix in which he studies the behaviour
of the features with respect to the view-illumination geometry and
some forest parameters. I edited the ISPRS JPRS article for the 1st
revision (still pending and some reading of general LiDAR-calibration
articles remains). We looked at the reflectance data (ATM, ATM+BRDF -corrected)
for the 5, 20, 30 and 50% tarps, and noticed some signs of anisotropy but
also strange behaviour in the 3-km data, where the P30 and P50 tarps were
underestimated considerably. These flight lines suffered most from clouds,
which was revealed by the maps Felix draw. To what degree is this esential
in the Xpro procedures remains still unclear. The terminology of the illumination-view
geometry needs to be evaluated against a paper LM sent, which has the "standard
terminology". However, do we have BRDF in tree canopies (volumetric scattering)
or is it shadowing effects, or is it both? Can we do with standard (old-school,
the ones we have now) aerial image view/illumination geometry definitions?
Jan 29
Academy proposal reviewing. Preparing some PPT-slides about RS
of peatlands. Felix cntd to debug the data. We found some problems in
h-rel values caused by 2008 LiDAR hitting clouds 450 meters above the
terrain resulting in somewhat overestimated Hdom-values! Some dependencies
with image features and age/h/hrel vars. On Monday we need to find the
three basic cases of illumination-view geometry and study those. VH suggested
how we should publish the results in 3 articles. UB send some hyperspectral
+ field verification BRDF examples. ISPRS JPRS reviews arrived and
need to be answered by Feb 26.
PileUP! 14(1) from early Jan was released.
Jan 27
Peer-reviewed on Sun for RSE and IJRS. Felix got the geometry
(brdf-angles) working and we filled the tree arrays with missing h_rel,
age, and site type data. On wednesday produced some pics with kuvamitt
that illustrate the sun-target-camera geometry in those 17 ads40 strips
that we are working with. A line sensor truly reduces the brdf effects
in one dimension.
Jan 22
Debugging of features, Preparing for data delivery in Joensuu
for Ville. Lectures and labs were made for Joensuu lecture on Thu. On
Friday we were examining the behaviour of the features in the illumination
& viewing geometry. Some uncertainty still in the angles describing
the illumination-view geometry, but the BLUE channel shows odd trends,
while others are stable.
In Joensuu tried teaching photogrammetry by starting with the
"theodolite is almost the same as a camera" example. Guess it worked
ok, but when collinear eqs. cae into the picture I saw less happy faces
in the classroom. Doing photogrammetry in 4 hours isn't maybe feasible...
The course homepage is here, with the xls-photogrammery
exercises and lecture notes in pdf.
Jan 15
The 3 PCs were restarted a few times on account of "out of memory"
errors in the w2k machine. Debugging of the data and programming of
the feature / predictor builder was active by Felix keeping him busy
for almost 2 weeks. The program is outputting nearly 200 vars/predictors
per tree*view. First classifications were done for debug purposes.
LiDAR-sp article was returned and corrected, ads40-article writing was
active and and some FGI-reference targets (24) were mapped to the (68x4)
ads40-images. Uli computed ome examples with Modtran, to show the
importance of path radiance (%) per band and flying height.
With Esa, OH7WV we struggled with PileUP! 14(1) and got it done
as scheduled on Sunday Jan 10. I had allready promised not to do any
issues, but doing jointly is another thing?
2009
Dec 22
We got the map_trees_to_ads40_images() routine to work at 0.3-2
trees per sec speed, thanks to predefined crown parameters (modeling
just once for the 15688 trees), and the C-dll function that checks the
inclusions and shadowing by the tree itself and its neighbors. The routine
reads now a tasklist. After debugging, three computers were left to work,
win2k, xp and win7 machine. Let's see what we get for Xmas-present.
Snow in Viikki. The ADS40-research team.
Dec 18
Felix got the new PC with Windows 7 to run the old VisStudio
6. Also the 8 TByte disk was made into RAiD5 mode and storing data started.
To make things faster, the mapping procedure will be split in separate
phases (crown modeling once for the 15275 trees, occlusion determination
17 times for the views, image mapping 68 times for all 4x17 image versions),
where the occlusion / shading testing will be made into a C-dll function
by Felix. We left the new computer to work over the weekend. Hopefully Mon-Tue
we have a solution and can leave the computer to work for the holidays.
We need to debug things before relying on the data / solution too much.
Dec 16
Programming of the Map_Trees_To_ADS40() routine contd. Focus
was in getting the per tree execution time to be tolerable and reducing
memory consumption of kuvamitt. It now takes 3-5 secs per tree to center
the images, solve the crown envelope, solve the occlusions and shadings
and store the data. DEM data type was changed into int16 from single to
save 50% of memory (80 MBytes). Also, the ODF-array was redefined to grow
in size only when new images are added. This saved lots of memory. 3-5
secs x 16000 x 68 is 1200 hours. That's still a lot,
but should we get access to 10 computers it can be solved. One
run would be 17-20 hours with a image set. One option still is to solve
the crown modeling only once and this would reduce the time 30-40% as
one big and slow loop would be avoided.
TBD: Felix lists needed LiDAR BIN-files to be copied on the client
computers HDs.
Dec 14
Debugging the Occlusion-and-shading determination -routine. Self-occlusion
and -shading are simply determined using the vector_angle() function
for the crown envelope surface normal vector and the sun and camera vectors.
Given the high density in the LiDAR data, 10-12 pulses per m2, the results
are good for sphere radius of 0.35 m. Shading can be determined for 3
degrees: self-shaded, self-shaded and neighbor-shaded, neighbors-shaded.
Dec 9
Now starts the implementation of the feature extractor for the
ADS40 species recognition study. The crown model in KUVAMITT was simplified
into y=a+b*H*x^c, (later y=a^2+b*H*x^c) where y is the crown radius,
H is tree height, and parameters are a, b, and c. a is used to create
a flat top in the crown envelope (rotated function for crown-radius),
it takes into account the XY-inacuracy in LiDAR. b is the scaling parameter and c sets the shape. We
noticed that it is possible to get negative values for a in the NLLS-estimation.
Also, we thought that it would be better, if we have some control over
the shape parameter c. In the estimation,
we introduced additional observation equations for a and c. With weights Wa and Wc, it is possible
to assure that a is not strongly
negative and c is near the expected
species -spesific value (look at item 2 below).
Spruce tree forced to have (c=0.1,a>0), (c=1,a=0), (c=1,a>0), (c=free,a>0).
We agreed that Felix would get familiar with the literature,
vegetation indeces and the R statistical system.
The feature extractor implementation consists of parts:
1) Building the tree
list with some 16000 entries (13318 + plots LK3, MARV1-09-1,
MARV1-09-2,MARV1-09-7,MARV1-09-8,MARV1-09-9)
2) Crown modeling
and extending the tree list with crown parameters {a,b,c}
1. Using DBH, H and SP, compute E(Dcrwon), crown width.
2. Using DensityF (0.8..1.2) multiply E(Dcrwon) to get search
radius - sets limits to a cylinder which is used to collect LiDAR data
for the tree to modeled.
3. Setting initial approximations for a, b and c. a0, b0 and c0.
For pine and spruce c0 = 0.5, weight
Wc = 1/1,
= 1
For birch / B-leaved c0 =0.3, weight Wc = 1/0.2 = 5
For all species a0 = 0.54, thus a0^2~0.3
m, and the weight Wa = 1/1 = 1.
b0 scaled to give
y = DensityF*E(Dcrown) at the crown base height, i.e. when x=1 and
H=height in y=a^2+b*H*x^c.
Each LiDAR point inside this overestimating crown
envelope is used for NLWLS (non-linear weighted least squares) estimation
of a, b, and c.
Each LiDAR point is at height x'(i) and distance
y'(i) away from the trunk.
Using the Jacobian A (partial derivates), a diadonal weigth
matrix W, and a residual vector
l. The solution in each iteration
is s = inv(A'WA)*(A'Wl)
Last two rows of A, l, and W have two constraints (and rows) in the
form of additional observations of parameters a and c.
In l, the
observations equations are a-a0=0 and c-c0=0.
Negative residuals (points outside the current envelope)
are also weighted differently from positive residuals to improve the
fit of the crown envelope.
Open: we should perhaps dampen the elements of s to avoid oscillations. I.e. to use something
else but standard Gauss-Newton.
Allow max 16 iterations?
OUTPUT a, b, c and crown width,
rmse and dbh_allometric. rmse should be < 25% of the crown width,
dbh_allometric=f(sp,height,crown width) s.b. also +/- 25% of the true
field measured dbh.
3) Implementing the
ADS40 sampling for the trees in 68 RGBNIR-images. 6x1x2 in 1
km, 4x1x4 in 2 km, 3x2x4 in 3 km, 2x2x4 in 4 km. #1=strips, #2=views,
#3 corrections.
ADS40 sampling consists of a routine that
loops thru the extended tree list and for each tree it performs,
having the R,G,B and NIR images read in (SET_PREFIX_STRIP_VIEW.txt),
in addition to the 2004 DEM.
3.0 Read a tree
Compute the File address using the tree serial-#
Offset = 1 + (serial-# -1) x
SizeOf(TreeRecord)
FileName.BIN consists
of three parts
PREFIX string*4 ("FULL",
"ATM ", "BRDF", "ASR ")
STRIP string*8 (mmddhhmm) e.g. "08230656"
VIEW string*4 e.g. "N00A" "B16A"
Input Tree-record
Output (69 bytes)
Plot string*15 (15)
Num
string*5 (5)
Serial
Integer (2) Address in files is based on this, see above
Xtop
Double (8) TreeTop
Ytop
Double (8)
Ztop
Single (4)
Zbutt
Single (4) Butt
SpText String*9 (9) Species in text
Sp
Byte (1) Species, numeric
Status
Byte (1)
a
Single (4) Crown parameters
b
Single (4)
c
Single (4)
Free 24 Bytes (24)
3.1 The in-image test
- Check that (col,row) -values of treetop are within image width
and height, and since it is L1 data, maps to non-zero image data (margins
have zero data)
Output (1 byte)
InImage Byte (1) 0=false, 1=true
3.2 In-cloud and in-cloud-shadow test
- Check that XY-point falls inside the cloud- and shadow -polygons
- PATHCLOUD gives the folder that has the N1 cloud polygons for
this STRIP_VIEW -case.
- PATHSHADOW points to the folder with the N2 shadow polygons
for this STRIP_VIEW -case.
OUTPUT (2 bytes)
InCloud
Byte (1) 0 = false 1 = true
InCloudShadow
Byte (1) 0 = false
1 = true
3.3 Sun- and camera vectors
- tree-sun vector, tree-cam vector are computed for the tree
top, normalized to lenght = 1
OUTPUT (52 bytes)
TopSun
i,j,k 3xDouble (24) Vector3D data type.
TopCam
i,j,k 3xDouble (24) Use this for all points on the crown,
since camera is so far away (cf. sun-vector)
dist
1xSingle (4) Distance to cam, (X0,Y0,Z0) retrived with d*PointCam+TreePos
3.4 Sampling of 1+120 points in the crown
Using variables Zdiff (0...40% of tree height) and phi (start,
start+2PI), we obtain points XtYtZt on the surface of the envelope.
radius = b * H * (2.5 * (Zdiff
/ H)) ^ c
+ a
^ 2 // According to Felix's suggestion, a^2 restricts the
constant to be positive
Xt = radius * cos(phi) + TreeTop.X
Yt = radius * sin(phi) + TreeTop.Y
Zt = TreeTop.z - Zdiff
Zdiff / H takes values 0.025,
0.065, 0.105,...,0.385.
phi start from azim_sun and loops 2PI at 30-degree intervals.
For each 121 points output (76 Bytes per point, 9196 Bytes per
tree)
Output
x,y,z
3 x Integer (6) Point offset from TreeTop (dX,dY,dZ) in cm, 32000+
cm = 32 m.
radius
1 x Integer (6) Crown radius in cm, 32+ m max.
i,j,k
3 x Single (12) Surface normal of crown envelope
phi
1 x Single (4) The rotation (redundant?)
Soccl 1 x Byte (1) Self-occlusion
using angle of surface normal and cam-vector
Sshad 1 x Byte (1) Self-shading
using angle of surface normal and sun-vector
Noccl
1 x Byte (1) Neighbor-occlusion
LiDAR Z-buffering
Nshad 1 x Byte (1) Neighbor-shading
LiDAR Z-buffering
colrow 4x4x2 Long (32) (col,row) of images R,G,B,NIR
RGBN 4x2 Integer (8) The image
data in 4 channels, all values s.b. below 32768 the upper limit of VB-integer
Free 4 Bytes
For 16000 trees, the storage need is 16000 x 9196 + 16000 x 148
~ 142.6 MBytes per file x 68 ~ 9.7 GBytes. Sizeof(Treerecord) is 9196+148
= 9344 (2336 4-byte words).
Dec 7
Felix is starting to get the hundreds of ADS40 images ready,
so we must start considering the feature extractor in more detail. It
seems that there are several hierachial levels; the trees in plots,
tens of points on their crown envelopes, that are mapped to R,G,B, or,
NIR images, that belong to a strip with certain GSD and radiometric processing.
Over 15000 trees in 121 plots are first solved the crown envelope
parameters {a,b,c}. Where r(x,[a,b,c]) is the function giving the crown
radius at height x down from the top.
Points on the surface of the rotated function have XYZ-coordinates
and normal vector as [dx,dphi] x [dx,dzdiff], cross-product:
1-meter long normal vectors (i,j,k) as seen in two views.
Now, we need to compute the BRDF-angles (3): zenith-cam, zenith-sun
and azimuth-difference(cam, sun). The third is still not clear how it
is done. The cam- and sun-vectors should be projected to the planes for
which we have the normal vectors (i,j,k). The normal vector defines the
Ax+By+Cz=D plane. Since the envelope is not
existing, and it is about volumetric scattering, we may well omit BRDF-calculations
at all points.
SHADOW-OCCLUSION TESTING
WITH LIDAR
Testing of sunlit/shades parts of the crown, we plan to do using
LiDAR data for z-buffering. Namely LiDAR points are turned into Spheres,
which cast shadow or occlude on the target of interest. Example shows
a completely shaded target (all test points drawn in blue, yellow lines
depict shadowing objects), and a tree with top 2-3 m in direct light.
Dec 4
Monday
was post-CQWW recovery day. Felix cntd working on the L1-data. A Q/C
test routine was built by Felix to check the geometry and RAW-file contents
with kuvamitt. Still some obscurities in the form of empty 0-valued files
here and there. Also, the L1 geometry was programmed a patch to fix the
GetSearchWindow() and read__ADS_file(), which need to be programmed in all
versions using L1-data. Worked the LiDAR-tree-growth data sets; field data
and LiDAR features. Some preliminary tests with tree-size-and-competition
-normalized growth data modeled by the LiDAR-features. On Friday programmed
a feature that measures the transparency of the crown, i.e. counts the intersections
of pulses that pass through the crown envelope, pulses with their 1st return
below the crown. Transparency seems to be highest further away from the
stem, as expected, indicating also the errors in the crown envelope, e.g.
non-circular symmetric crowns will show as transparent, wrongly. ISPRS
J reviews the article sent last week. No MATINE money. UEF dean-selection
took place.
TBD: Implement the ADS-geometry corrections in Felix's KUVAMITT
version. Finish Q/C of images. Start implementing the feature extraction.
Prepare 13118 + L3 + MARV1-2009 plots to be used for ads40 feature extraction.
Nov 27
Got the UH-3-a report done on Monday. Faculty meetings x 2. Fnished
the LiDAR-normalization paper and got it sent to the journal and language
review. Started to prepare the LiDAR-growth data sets for TT & SH.
ADS40-work ongoing (validation and L1-production).
Nov 20
On monday LM & EH visited us in p.m. We agreed that we extend
the Xpro products to 6 for the tarp-AOI. We tested kuvamitt, and it supported
the 1x1-km L1-patches directly (unbelievable!). We got the shadow-casting,
self-occlusion and neighbor occlusion testing running using the Z-buffering
approach. But, we decided not to use the convex-hull delineation of
the crown points in the image plane, because the projections of e.g. in-light
areas are not convex, and the Z-buffering results in noise at times /
individual isolated pixels with "error" in the shaded-determination. Felix
was active with cloud-screening, and he computed the areas belonging to
clouds or shadows and the "loss of reference trees". Just, one 30-cm strip
was bad, with >50% of trees lost, because of a single cloud. In 1 km
data, the clouds cover large areas, because of their proximity. Results
for 1 km still pending. Worked hard with LiDAR-normalization paper to get
it out of our hands. JH & VH suggested corrections. Corresponded around
the theme. AH got the marking ready and we agreed that the fellings would
be done after 2010 LiDAR. HL visited for amateur aerial photography using
experimental aircrafts. The elections took place Mon-Tue and it turned
out that there'll be lots of meetings for me 2010-2013 as I got support for
both the Faculty and Department boards. However, I think that it would have
been better without voting, but for some reason such pre-election discussions
never came about amongst the post-docs and grad students.
TBD-ilkka:
Report UH-3a (DL Nov 24),
Send
the article to ISPRS JPRS & UH language centre (Check the earlier
language review! Does Felix have the paper copies left?)
TBD-Felix: Xpro-processing ATSR, ATM,
BRDF, DRDF-ATM 1 km AOI, PANTIFFTORAW-batches (The C-program), UNBLOCK
(VB_unblocker.vbp) -batches, Build HDRs for KUVAMITT.
Nov
13
The GetSearchWindow() routine was given a temporary cure that
helped such that only at line 0 and line N, the mapping from image to
xyz and back to image would not work, but there it's to be expected. It
turned out that our SDK source samples were rather old, in need of improvementfixing
and somewhat slow. Anyhow on MON we believed that kuvamitt now has working
geometry for the ADS-imagery. Our old VB6.0 lacks the uint16 data structure,
which is needed for "proper" reading of the image data. It turned out
that 0...32768 should suffice in 99.999% of cases. The reflectance data
is from 0....10000, for example. FR started to work on the shadow/occlusion
determination and got the convex hull algorithm going. The workflow to
1) produce block-TIFFs with radiometric corrections using XPro, 2) Untile
these to raw-file-blocks, 3) and to join them to unblocked 0...2GBte
RAWs with odf/ads/sup/cam/hdr was tested and by wednesday afternoon we
had 15% of the data processed. The bottleneck is the manual deriation of
the BAT-files needed in phases 2) and 3). Thursday we got ready for the
Joensuu trip that we did on Thu-Sat. Also, I looked at the LiDAR-normalization
manuscript. Thu we met with VH/TT to discuss the project. Friday at the
Forest Faculty for a seminar with 11 presentations by the researchers/faculty
and us. Some social events in the p.m. Felix's presentation about the
ADS40-project. My presentation
in LiDAR intensity.
TBD-ilkka:
Report UH-3a (DL Nov 24), Sphere-casting, convex-hull,
PANTOTIFF-work, UN-BLOCK-WORK. BUILD HDRs.
TBD-Felix:
Xpro-processing ATSR, ATM, BRDF,
DRDF-ATM: 2-4 km. 2km 0753 AOI,1 km AOI, Cloud-screening,
Nov
6
Got the NSF-review done by DL, Mon 12UTC. Xpro still not configured
to work with the AT done @ fgi. Felix working hard on it. Even the bought
1TByte Sata-disks wouldn't communicate with the 2006 motherboard, and
we needed to travel to Mechelininkatu 42 to get 500 GByte replacements
with the slower 1.5 Gb/s protocoll-support (jumpers). Cloud-screening needed
a polygone digitization routine that works with the L0 imagery. Also
a routine for writing xyz-footprint and tarp-sun-tarp geometry,ascii-files
per ads-image got started. Was interviewed for the Hyytiälä
100-yr document (turning old!) we did a short visit to GIS/EXPO to see
JV. Felix supposingly now has all the official paperwork done with last
visit to the bank. The UH-elections paperwork was done, thanks to help
by many at the 3rd and 5th floor.
Wed-Fri: Working from home; twins in fever. Programmed the mapping
of the in-situ targets
to the images; and the footprints (115) using the L0 and "Oct-2008-Fernando-odfs".
CJRS review took nearly all Thu. Fri: needed to program the polyline editing
for the digitising of the clouds and the shadows in the images. Noted
that the gnd2lp() function fails in narrowing down the search space when
the point maps to the first ~2500 lines at the top of the images (Fig below).
Redid the affine tranformation computations, but still not solved. Also
noticed that the ray-casting for "LiDAR-monoplotting" is not accurate in
3-km and 4-km data. The interected point is 50-70 cm off from the pointed
camera ray (thus the camera ray "is turned"), which does not happen with
3-km or 4-km frame images having KKJ-orientation data. Need to test if the
KKJ_2_UTM_2_WGS_2_LSR() is not working precisely when "off ellipsoid". This
could explain the deviation. The xyz-points map in correct positions in
the 3 and 4-km data, and forward ray-interection works fine, but there, the
same tranformations are all made, down near the ellipsoid.
Left: 0852UTC
Nadir 4-km image; the top 3000 rows are "missed from the xyz-space" /
mapping goes wrong. In similar 0843UTC 4-km data there was no problem,
suggesting intability of KUVAMITT's solutions to grnd2lp(). Right: There is 1-m point spread in the
monoplotting results, which depend on camera's height above the ellipsoid.
This suggest that the KKJ_UTM_LSR_WGS-transformations in KUVAMITT don't
work but near ground (are not 3D, but 2D)? Thus the monoplotting / ray
casting against the LiDAR s.b. done in WGS/UTM; in which the EO of the
ADS40 data is defined. Must check also the Earth curvature effects, which
might not be fully taken into account (the XY error with respect to the
direction towards the anchor point). It seems that, the problems start to
appear when playing around with several coordinate systems simoultanously.
TBD-ilkka:
Report UH-3a (DL Nov 24), Aug-23-2008 & May-31-2009
datasets, HYDE_REF to unix. Fix bug in grnd2lp(). Programum the featre extractor. Cloud screening.
TBD-Felix:
Xpro-processing
proposal: what versions to be included and their rationales. Cloud-screening.
Oct
30
Hectic writing of the Academy proposal Mon-Wed. Felix did the
language review and got the laptop running, Xpro worked out installation
as well, First trials with 20-cm L1 image production on Wed with the
original raw-orientation data. An improved graphics card will be need
to support the ADS-viewer (DDR2 256 MB vs. DDR3, 512 Mb). Ordered and got
a pair of 1 TByte 3.5" hdd's to go with the Xpro. Discussed with IT-people/Sorvali
about the option of "distributed computing" when we do the feature collection
with KUVAMITT. Teppo promised to help in this by setting up an image-server
(with volume X:\Images\) "called" (fopen) by the KUVAMITT-clients. On
Mon-Wed some correspondence about SIMO/MELA analysis/simulations of the
Hyytiälä site/forests to gain knowledge for commenting the
cutting activity. HO promised to keep the Normalization-manuscript and
contribute to it. I got the honeybee/users/korpela/ -files, 2.4 GBytes
from the IT-dept on a DVD. On Thu the proposal was sent, and some activity
around the ads40 16/8-bit versions for AhK. Fri at FGI to get the needed
Xpro-files + metadata and to discuss options for the next steps. We received
the rsu-files for the Hyytiälä forests for some SIMO-analysis.
JV sent the manu away. Saturday @OH1XX, NSF reviewing too.
TBD; Tue a.m. interview
Hyytiälä 100th jubilee, p.m. GIS-expo in Pasila.
TBD-ilkka:
NSF-review,
CJRS-review,
Report UH-3a, Aug-23-2008 and May-31-2009 datasets,
Save-copy honeybee-files, copy HYDE_REF to
unix. Programming of
feature extractor.
TBD-Felix:
Xpro-processing:
what versions? With EH, UB, LM, IK. Sun elevation & azimuth computations
in Matlab, Cloud-screening using LiDAR/Images - manual.
Niklas, champion of the
autumn hockey tournament @ the Kuusitie stadium. Perttu & Arttu
sharing the excitement. Fanny too.
Oct 23
Felix arrived on the Sunday and we used the week for getting
all administrative and preparatory work done incl.
the accomodation, computers, contract etc. Planning of the 4-month
period was also in the programme, a little as well as meeting with people,
which we did on Thu-Fri at the Nordic Rem Sens days in Kumpula, where
we also had a poster presentation, during a 15-minute coffee break :).
Sort of reflects the event! HO had done work on the normalization-manuscript,
and I had done my best in confusing him by even sending the wrong file
to be updated. Thu-Fri we realized that its us who need to do the Xpro-processing,
and we hope to be able to start with that early next week. Reviewed a
gradu and a manuscript during the week. On Friday we got Felix's laptop
working with KUVAMITT and the sample data material.
TBD-ilkka: NSF, CJRS, Academy proposal-5a,
Report UH-3a, Aug-23-2008 and May-31-2009 datasets,
copy HYDE_REF to unix. Programming
of feature extractor, L1-debug.
TBD-Felix:
Silva Fennica language review, Xpro-installation, Results of the triangulation
with Xpro from FGI. Xpro-processing. Sun-elevation&azimuth computations,
Cloud screening in LiDAR.
September-October Kulloo-weekend-activity-center.
Tassu is turning out to be a fine lady...
Viljo & Perttu at a Birthday party. Took my camera with me
to try some "sports photography" and document our "Viikki-Sunday-sähly"
-activity, ongoing since Autumn 1995.
Nordic RS days in Kumpula. T. Mielonen, R. Ilves, L. Markelin
& L. Korhonen kindly visiting the peat-poster. Hannu, OH1XX &
K9AY-loop in Pertteli, Sat Oct 31.
Oct 16
The analysis on the response of the tree classification and X-vars
in LiDAR to the varying parameters of the range- and AGC comepansation
continued. Most of our hypothesis turned out to hold and the earlier
theoretical work with the radar equation by Ahokas et al., Wagner et
al. (2006) seems to hold true. The results suggest that there are diffecerences
between species and echoe types. The latter is easy to consider in optimizing
the range normalization, as it can be done separately for single- and
multi-return pulses. The gain from doing that is however rather small,
max. 1% in classification accuracy of the Finnish main tree species. The
derivation of the observation matrices was painful as it took sometimes
>10 hours to produce feature sets for a certain grid of the parameters.
Tried to get the SAS to run in batch-mode, to read e.g. 500 different input
files and store results of classifications. I was helped by the SAS-expert
at the university. It turns out that, of course, that there are several
alternative ways of doing the same thing. The solution was to make a program
that writes the SAS-code, with ODS -statements to store the tables that
are normally seen in the results-window, and then to use proc export to
write the solutions to ASCII files that have standard structure allowing
them to be read again by a program that collects all the results.
These figs give support to the theory that range normalization
of intensity dependes in the "interaction" i.e. the scatterer geometry
(incl. gap size distribution, 'clumpedness') and silhouette area.
The behaviour of individual features in F-test was logical for the best
predictors of species and the only echoes. Range normalization reduces
per tree CV, and shifts the intensity distributions towards a mean "at
(R/Rref)^a", so it acts in two ways: noise reduction (for overlapping
strips) and shifting (towards mean topography/elevation).
(Review Pro Gradu, NSF, CJRS, NN-article, language review, FR
arrival (paperwork etc.), KKK-2009 (Thu-Fri), Proposal-5a, Report UH-3a,
Aug-23-2008-May-31-2009 datasets)
Oct 9
Finished by Tue the LiDAR extraction for the LAI / VALERI-plots
in Hyytiälä. On Monday, the tower-XP-PC would not boot. Carefully
checked every disk in another win2k machine, and they were ok. It eventually
turned out the a SATA-cable between the system disk and the m-board
was broken, presumably this cable had also been responsible for all
earlier slowness and other hastle in that PC. Wed started to extract
the LiDAR data for the 14000 tree crowns such that the intensity normalization
parameters were tried in a grid of values. Well, it wasn't until Friday
that I believe that a bug-free version of the exctractor-routine was
available. As it takes 6-7 hours to compute the 20 x 14000 x 960
observations (1.4 GB of ASCII), bebugging was painful. We tried if there
is a minimum in the CV of per tree intensities along the parameter range
(Fig below). Also, tried if there is a peak in the F-value in an ANOVA of a feature between species. Still unknown
if there is a peak in the SP-detection performance somewhere among the
intensity normalization parameters, also looks as if the optimums are
dependent on the tree species (reflector) and the echo type (intersection
type). LK was succesfully using the 2007 agc&range -normalized intensities
for LAI-estimation, and the 2008 data needed normalization, which was
done on Friday afternoon using again the F-statistics as criteria to be
maximized. We had seven types of natural and man-made intensity targets,
and the idea was to find parameters of the normalization such that in
the ANOVA of the targets, F was maximized. Similar coefficients were found
as those in the 2007 data, which was flown also with the als50-ii, but
different version. Next week remains the preparation of the LIDAR and
field data for a study where we look at the possibility of explaining between-tree
variation in increment using LiDAR features. The laborious peer review
was done on Sunday when it rained all day. Correspondence around the idea
of using LPS in forest, where trilateration is based on acoustic distance
measurements. There is activity in Scandinavia. Oh, and Monday I found a
bug in the Bilinear interpolation routine of KUVAMITT, but it luckily
did not have a major effect in DEM/CHM evaluation (Fig). Very exciting week
with long awaited research Qs getting answers and sprouting new Qs, as
usual.
1) Local minima in range correction coefficient for Maple (Acer
platanoides). 3) The SATA cable. 3) The DEM differences before and after
correction of the BUG in bilinear interpolation. 4) Intensity tarps used
for calibrating the agc-compensation model in als50-ii-2008 data.
Oct 2
Writing applications for funding in 2010. Got one sent out on
Tue. Peer reviewing an interesting but laborious paper that has some
unfamiliar statistical methodology applied. Such cases always result
in extra work, if you really wish to know what went on. Got some results
concerning 2008 AGC-compensation tests. On Wed-Fri started to help LK
in the retrieval of the LiDAR data for the LAI-plots in Hyytiälä,
some of which reside outside the 2004-TIN. This meant that I needed to
tune the raster-ti-TIN-filtering DEM-algorithm to use also the 2008 data
and build local 200 x 200-m DEMs that cover the plots. Then we looked
at the ASCII data samples containing 2004, 2006, 2007, 2008 data in r=40m
circles, and noted that in 2008 the echo order was wrong. I had mangled
with the LiDARRecord -struct in May, and swapped the order for the 2nd
and 3rd echoes. So needed to program a patch that read all 9039 1-ha binary
als2008 files (46 GB) and swapped the 2nd/3rd pulses and fixed the last
echo Byte positions. The VB-application run for 8 hours, but now the 2008
LiDAR, which wasn't used untill now, should be in order as well. Lot's of
correspondence around the ongoing/starting analysis with co-op partners.
20 x 20 km area (2510-2530E, 6850-6870N) with LiDAR density in
2008. By product of the pulse-swapping process. Shows the 15 flight
lines 1-4 km agl.
Sept 25
Perttu was ill the whole week (8 days) so stayed home and worked
75% as the flu was trying to knock me out too. Planning and reading
articles in radiometric calibration processes and doing one peer review
for a remote sensing journal. The three plans were: i) The Hyytiälä
2008-2009 aerial imagery project, ii) LiDAR in growth estimation and
iii) tuning of the intensity normalization project. Some fun correspondence
around the als50-ii's AGC. It turned out that response time is tunable and
many researchers have used the als50 data believing it is a 0/1 case. I
felt really stupid since I've also used the data and assumed that there's
just an ON/OFF switch for the AGC and had been lucky in the sense that the
AGC settings in the 2007 data from Hyytiälä had slow AGC-response
times. This way the intensity may get to be saturated, corrupting the range
observations slightly, or some echoes are lost. When we looked last week
at the 2008 data, flight involving also FGI, the instrument had extremely
fast AGC, which is good for the range data (for the DEM estimation), but
the dynamix range of the intensity signal is reduced. The als50-ii system
is either favoring radiometry or geometry, depending on the speed of the
AGC. Friday I was in Kilo @ FGI to discuss details about and on how to turn
the aerial images into radiometer observations. The in situ measurements
in Hyytiälä, both during the ADS40 and the DMC campaigns, in 2008
and 2009, are really versatile. Lauri had found that in addition to the aerosol
optical thickness, the SMEAR II station is also equipped with a device that
measures the horizontal visibility. These data will allow a more accurate
description of the atmosphere, thus improving the estimates of at-target
radiance (incoming) and at-target radiance (reflected, scattered). Lot's
of things to in the next weeks, but it is exciting and new stuff.
The als50-ii scanning In August 2008 was giving minor headache,
as for the AGC in the device and its settings, I thought for some reason
that the AGC was set at fast response, i.e. to react immeadiately to
the signal. Then I had a proper look at the data. The AGC byte was >245
for the 3-km and 4-km flights giving an overall impressions that it was
just basically at the maximum value. As for the 1-km and 2-km data, the
values were also different from those in 2007, but the range was similar,
4-5 bits, and the response was slow. In a flightline, where the AGC was
swithed off, data was lost at times, because the sensitivy of the system
was deteriorated?
Sept 18
(Sun-Tue) On Monday we were in Hyytiälä to see the
trees that we used in the tree species classification in LiDAR research.
Hans Ole, Aarne, Timo, Antti, Sanna and Silja were present. We discussed
the field experiment, maintenance of the plot network, management of
the forests in the 2x6-km AIO, LiDAR data acquisition etc. We had
a LiDAR-promenade and looked at the LiDAR intensity maps, such as this
pdf
to learn about the intensity signal "in situ". On Monday
afternoon we made an excurision to see the plots measured in 2009. We
also looked at some "extreme trees or cases", extreme in LiDAR metrics.
Interestingly we could easily see the differences in the trees as measured
by the LiDAR intensity. It also looked as if the LiDAR signal might explain
some of the classification errors (species) made in spectral optical data.
Wed-Fri was hectic as I needed to prepare for my first ever lecture in
Swedish. To complicate things I said yes to some research application writing,
and I ended up doing two proposals, which were mailed on Friday. The best
part of the lecture in Swedish is that it is now done. Our university is,
in principle, bilingual, however, you seldom hear anyone use Swedish. Here's
the ppt-file
to my lecture notes. Our field excursion-pdf
is here and the presentation about 2009 field measurements is in this
pdf.
Demonstrating
the Lapinkangas trees and measurements in Hyytiälä. H-O Ørka in the front. Foto T. Tokola.
Sept 11
Finished and submitted the article on tree species classification
in LiDAR. Correction by co-authors and some last-minute analysis were
done. Prepared the excursion for Mon
next week in Hyytiälä. Hired AH for another month to
finish the field work project. Started the laborious preparation of
a lecture to be held next week in Swedish. My first ever
and the vocabulary is somewhat unfamiliar. Will be exciting to
see how it goes. Also made some preparartions for FR visit from ETH
in Oct. Stressfull as always. Highlight
of the week was when Iwe arrived on Tue mng to the kindergarden,
and the Audi 80 engine was behaving like a sauna owen. Steam everywhere.
The manifold between the
carburettor and the motor had created a leak of cooler
liquid. End of that story... Luckily there was a car for rent in Salo.
--------------
We decided to change the old oven to a new one. It had served
for 25 or more years, but was misbehaving recently. Built a "tent" around
the oven and put a fan on top of the chimney.
Believe it or not but a company would have charged us almost
2 kEuR for the work... Here's a view to the northern wall that got new
ladders this summer. Fresh new paint as well.
It seems as if owning an old house means that you are surrounded
by unfinished projects all through your life.
Sept, 4: Mon-Wednesday was spent with finishing of the article.
Some practical arrangements were made concerning Nov-Feb, and on Friday
EP visited and we looked at the Hyytiälä data. The CHM-routine
in Kuvamitt needs to be improved. Now there were problems if the raster
size wasn't an integer. Looks as if September needs to be spent with money
applications, what a blast! I need to remember to help LK with his efforts.
Also got help with logistic regression. Need to look at that next week.s
Academic life, ... I got really disappointed again this week
when I realized that so many focus on things that support their career
and not on problems that need to be solved. As if solving the problem
"in one article" is really stupid - if instead you can do 7 (items in
your list of publicatios) seven and still not solve the problem but "provide
with good results". "I'm taking this course in academic A and B, because
it might add ods of getting a tenure position - not because I really need
improvement in A or B". Think of that. This explains why it is much easier
to do co-op over country borders and disciplines. "Lean forward, that's
my only advice", said once a senior academic to me. That helps in taking
the hits - mental ones also I guess. Anyway, if you close your eyes on many
things, working with resrach even at the university can be exciting. I'm
really privileged, I should remember.
August, 28. Last week was spent with the species discrimination
with LiDAR project. There was a bug in the computation of the sensor-to-sensor
normalization, which resulted in the agc-noise being amplified... Missing
outer () parentheses in the longish equation where the reason. Recomputed
the (13318+570) x (960 vars) observation matrices. The differences
in kNN/LDA/RF were due to the use of priors. Interesting effects observed
as the intensity signal was dependent on both the absolute and relative
size of birch trees, which partly explains why birch was hard do discrimianate.
Lot's of writing to do in the article, since it could be split into 3-4,
which I'm not going to do although it would maximize academic output
and the length of the CV/LOP.
August, 21. Preparing sites for the field team's remaining
3-week campaign. Compiling the LiDAR-SP manuscript with results from
Joensuu & Norway. Some obscurities still with the data that need
to be chacked: 1) The sensor fusion and computation of sensor-to-sensor
normalized intensities resulted in inferior accuracy compared to intensity
data that was only normalized for the range and agc. 2) Site - age distribution
looks partly odd. Need to check the plot-level age data once more. 3) RF,
kNN and LDA in R all produced different classifications when compared to
SAS-software's kNN/LDA and Minitab's LDA, which were in line with k-MSN
classifications.
Excursion to Hyytiälä Sun-Tue, Sept 13-15 seems to
be feasible.
Prepared and send an abstract in support of the activities of
KKK, remote sensing club in Finland. Another one awaits.
The mapping accuracy of the understory trees (field campaing
2009) was done in 3-stage analysis somewhat similar to image orientation:
exterior orientation of the control points, between strip (relative)
and within strip (interior) orientation by taking distance measurements.
The accuracy inside strips is 3-4 cm, but the exterior orientation of
the strips has an accuracy of apprx. 13-15 cm, which can still be improved
somewhat by using additional observations and constraints.
Aug 6-8. We had a nice 20-yr
reunion @ Hyytiälä of the-1989-forestry-students. Tears and
laughter in right proportions.
July-August
The LiDAR-based tree-sp detection paper is due next. HO and MM
have been active and compilation of results awaits now.
Disapppointing but the FW-LiDAR campaign did not see daylight
in 2009 as I was not able to raise a consortium (during the July holiday
season). FW-LiDAR escapes Hyytiälä.
PEER reviewed 2 articles and reviewed applications for 3-yr PhD-positions.
The latter is very difficult but helpful in that it is easy to recognize
what is essential, and what happens when
you are not an expert excatly in the field, which is presumably
the case in most evaluations. Your readers are not so aware of the field,
and to convince the reader takes something else.
A good aspect of vacation is that when one returns back to the
office the mind is empty. I was fortunate as the holiday wasn't interrupted
by too many work-related events.
In early August started to work by visiting the team in Hyytiälä.
Here pictured in Lapinkangas mapping understory trees in a 0.4-ha plot.
We apply the photogrammetric-geodetic method with end points of (wire)lines(strips)
positioned very accurately. All trees have an exterior accuracy of
approximately 0.15 m in KKJ and an accuracy of 5-10 cm in strips. We
are also taking bore core samples and remeasuring trees after 4-7 years.
Holiday-trips: Miehikkälä Salpalinja-Bunker, Viljo
@ Velkua-beach (Naantali), Lappi TL old stone piles (UNESCO-site) with
Grandfather.
Start of vacation, Sun July 5, 2009.
June MARV1
Showed up in Hyytiälä on Mon, June 15 to get the 3-month
field campaign started for Timo & co and myself. 7 hectic days
spent in getting the 5000+ trees in 8 plots measured for the MARV1-course. The
18 3-student groups worked in 5 x 4 plots. Pauliina and Aarne filled
in the 2 plots that were left without student- surveyors. In all, the
plots cover 2.68 hectares of Hyytiälä forests and some 3100
maped and measured trees were added to the experiment.
Some practice before actual forest mapping.
Finding the omission trees. Some trees were measured,
some burned in 2009. Tuula & co.
Improved RESECTION to have an illustration of the geometry at
the start of the iteration. But only after all computations were done.
This kind of feature is helpful in determining azimuth blunders. Something
similar is needed for the interpoint distance observations.
The DMC images arrived from NLS and were quickly converted into
BIP format from the different 16/8-bit, 1-4 channel tiled TIFFs. The
8-bit versions lack colors (narrow histograms). Made some orthoimages for
the MARV1-course.
Pauliina & Aarne were given their assignments, the
mapping of the understory vegetation is slow, 0.04-0.06 ha per day,
but the accuracy is good, < 10 cm for XY.
May-June: "ISPRS Hannover-NOVA
Mekrijärvi course-FORECO manuscript revision-PEER reviewing-Hyytiälä
measurement campaign & MARV1-course"
It seems that the end of May was too busy since I had no time
for keeping this blog. Quite a bit of effort was put in getting the LiDAR-feature
data sets
ready for the tree-sp detection project involving also other
researchers. There were 12 x 80 variables in the requested form + point
clouds for JV.
Then the trustworthy laptop stopped working which gave extra
headache for a week or so. It's amazing how dependent we are on our
computers and how
I had developed this "emotional relationship" with my nw8240
that had accompanied me fr nearly 4 years.
Hired two field workers to measure understory and growth of trees.
This went smoothly despite the fact that we were so late. There is
very little work for
the students this summer.
CJRS and IJRS journals got their reviews. I even got some feedback
for the review after I requested it. It's good to know if you are doing
the review properly or not...
NLS flew their DMC in Hyytiälä Sunday
May, 31. I was there with Teemu H. and it was interesting to get to
use the spectrogoniometer and see the FGI-UAV. The flights occurred in
the mng, with low 25-degree solar angle and later in the day, when the
sun was at 45-degree elevation. 2-, 3-, and 4-km heights were included
in the image blocks. The campaign completes the test set of aerial cameras
in Hyytiälä: now I have UltraCAM D, DMC and ADS40 tested. UltraCAM
XP should be tested at some point as it has smaller pixels, and probably
better lenses than the D version.
Ilkka Korpela & Jouni Peltoniemi
FGI's UAV & reflectance tarps.
measuring the white reference.
Hannover ISPRS workshop 2.6.-5.6. again turned out to be a fine
event. Lot's of new and old friends. I had just a poster there.
After Hannover, I was very busy preparing for the NOVA 3D course
11.6.-12.6.. I decided to have a full day of program with 3 x 45 min
lectures and 2 x 90 min
demos/computer labs. Getting the Mekrijärvi PCs to run the
VisBasic 6 luckily worked out quite easily (just one day of fighting
with the computer lab), but in all, I invested 6-7 working days to
get that 1-day act together. Well, in the future it will be less
pain if I get a chance to do it again.
We started field work campain in Hyytiälä on Monday
June, 15. I was not much in the field as I needed to get the MARV1-excersice
ready. That involved some 14-hour working days and some work still remains
for the last two days before the course.
Arttu and Perttu celebrated the end of the school year, which
was memorable and you do it only once.
The intensity tarps with LiDAR obs (intensity in HSV-color),
point size depicts flying height 1, 2, 3, 4 km.
The elevation maps shows that the XY accuracy is worse for the
higher flying heights.
15.5. Still unloading the ELS-files of the 2008 LiDAR. A better
program structure is definately needed to speed up things when upto 10
flight lines cross (overlap) a certain area (hectare). The 67%
smaller file size of the ADS40 PAN02F images got its explanation; a
different
jpeg-level was applied unnecessarily in these NADIR PAN data,
unfortunately, the jpeg-level of PAN02F (10% level) was the same all
through the campaign, B14 and F27 sensors were compressed with
the 25-% level..
PAV warnings not necessarily mean that the pitch-angle, which
was taking large values due to low speed of the ES-MAA, would corrupt
the
omega-values in ADS40 (or ALS50).
Got the monoplotting / ray-casting-with-LiDAR working in the
ADS40 data such that the LSR-UTM-KKJ transformations work.
13.-14.5. Putting the als50-2008 data into pulse-records and
per-hectare binary files. With several overlapping flight lines from
the different
altitude, the C-program resulted in heavy file-fragmentation
with lots of fopen() and fwrite() fclose() calls for the hectare files.
Upto 1000 fragments
for BIN-files with 50000 pulses per ha. Trying to solve this
by allowing for 100kB empty tails that are always reserved when space
runs out. This
way the OS should be able to track what's going on and less fragmented
files should come out.
The
altm3100 data of 2006 and the als50 data of 2008 are giving same heights
near Lakkasuo on road 66.
Also the intensity data shows that the XY-match
is good for the als50 1-km data
Programmmed the ray-casting using the ads40-images; it needs
to be debugged still properly because it involved orietation and coordinate
system transformations between KKJ, UTM and the Local Cartesian system
of the ADS to enable shooting of the ray through the telecentric lens
of ADS40 to the KKJ-world below.
11.-12.5. Started to program the phase-II LiDAR feature extractor
that separates observations by sensor, and provides features
also for the non-agc and non-range normalized cases in addition
to the "every-sensor-fused-and-all-intensities-normalized" case. This
version also output the LiDAR data to separate BINARY files.
Tue was spent at FGI in a joint "LEICA - CO-OP AROUND HYYTIÄLÄ
EXPERIMENT" seminar. A. Rohrbach and R. Roth presented
the latest in ADS-camera and ALS-LiDARs. An interesting detail
was the option of having FW-features in a discrete-return LiDAR
obsevation, hardware-computed on-the-fly. MPiA is nowadays implemented
in 30 sensor models. Effect of AGC is non-linear in
ALS50/60, yet the exact response (form) is not available for
those who wish to compensate for its effect. The ADS can now record
uncompressed at very high "refresh-rates". NLS guys were present
and we discussed the recent advances in leaf-on DEM estimation.
I presented (pdf) the background
and contents of the Hyytiälä RS experiment; past, ongoing and
planned projects involving LiDAR and aerial image.
Timo T went through the aims
of Academy project at Joensuu Univ. Results in camera-simulations and
Alpha-shape computations. The ADS40
campaign of 2008 is a part of this project / acquired for this
project.
Eetu P. presented his SP-detection method that relies on the
use of a CHM for tree detection and sampling of image values in the
sunlit
and shaded sides of a crown. There is a need for another test
set (LiDAR and ref.trees) from Hyytiälä.
Eija H presented the EUROsdr project; the ADS40 iamgery will
need to be transferred soon to the research organizations; TT&IK
will participate
in the evaluation using SP-detection for benchmarking.
4.-8.5. Debugging LiDAR-SP project's obs-file / re-programmed
some of the trickiest stuff. This 2-3-day subproject was now on its 3rd
week, and this resulted in delaying of all the plans. Also, 2009-project
plans were reformulated as MK starts another project by June 1.
MMVAR-Colloquim (pdf) on Friday.
Not as great success as we had just 1+4 people in the audience. Presented
the results of the ISPRS
Hannover June conference. In the middle of all, I programmed
KUVAMITT and debugged the ADS40-data to see that the geometry and
radiometry are ok.
3.5. Peer reviewing for IntJRemSens (This one took only 10 hours
in total, which is progress), CanJRemSens remains 50%.
Planning timetables for May-August to include some field teaching
in yytiälä, ISPRS Hannover trip, PhD-course at Mekrijärvi,
work on the LiDAR-SP and ADS40-SP papers and field work for data
collecting for winter 2009-2010 (detection and measurement
of understorey trees; estimation of per tree foliage, with LiDAR).
May 1 was a particularly sad in Kulloo. RIP R.
April
28.4. I was stupid and said yes to another peer review project,
now two overlapping - as if there's not enough work.
I went through the ads40-flight and enummerated the files that
need to be considered in converting all data to KUVAMITT
format. Altered the c++ program PanTiffToRaw.exe to work
with argv[] arguments for file I/O. Then worked with KUVAMITT
to see how the object-to-image calculations function. Superimposing
LiDAR data is much slower in L0/L1 images because
of the iterative method. There must be ways of a) making
it faster, b) being sure that the solution in image is
unique.
Superimposing some 6000 points in ADS40 images took 250% more
time in comparison to frame images.
Forward-ray intersection works separately for the pushbroom and
frame images, but mixing image rays
(observations) in one estimation needs further work.
27.4. MK was working with the additional plot and tree level
variables, to complete the observation matrix for the LiDAR_In_Tree_Species_Identification
study. Started to prepare the ads40-images.
MARV1 measurement campaign will see daylight. Started with planning of the areas.
Lot's of correspondence around the 2008 ADS40 campaign, DMC 2009
campaign, and MARV1 measurement campaign.
Working on a data matrix that has all plot level variables for
the 117 Hyytiälä plots.
Computers are good at making hard-to-detect errors which cause
a lot of extra trouble, when you need to re-do things.
Got the Mire article ready April 16, 2008. Co-author responses
were added. It is about 15 pages in journal format, which is quite long,
but the article covers three topics, which were put in a single research
article.
Working hard to get the ISPRS Hannover workshop
paper done that was due April 15, 2008. Interesting findings in the
LiDAR intensity response of different tree species and trees of varying
age. The study has 13000 test cases, which is 20-fold to any other that
I'm aware of. Finally the hard work in Hyytiälä is "paying off"
and the experiment allows for reliable testing of many hypotheses.
Winter was good and we were skiing to school/kindergarden untill
April 3. Easter in Kulloo, Viljo wanted to be Darth Vader, so you see
that we are still living in the world of Star Wars. If all goes as planned,
we will have a new family member in May.
CQ WPX SSB in Arkala. Pre-contest planning took
a bit of effort. We worked in M2, and had a great time. Aurora was moderate,
and 20 meters stayed open nearly 20 hours each day. Lot's of new prefixes
around, probably due to the fact that telecommunication administration
all around are getting 'liberal' and finding new ways for additional funding.
Anyhow, I enjoyed myself calling "CQ Contest, this is radio Arkala, QRZ?".
Anaglyph
stereo image of Arka-Jussi, the 160/80-m beam up 100 m. Red-blue or Red-Green glasses needed
for 3D effect.
OH8NC, OH6KN and OH1TV operating on Saturday.
Before
the contest we had some elementary training. OH6KN and OH1TV discussing
antennas. Our contest wine was Tollo, in the spririt of WPX.
Working on the LiDAR_In_Mire manuscript.
22.3. PileUP! 13(1) was released. Some 70+ hours were spend with
this issue. Would not have been possible without help by OH7WV/BY4 and
OH6BG. Again I found myself writing articles, which isn't what
I had in mind when I accpted the editor's post. Next one is in
September, so there's time to recover.
Continued working with the monoplotting observations for the
SP-recognition in LiDAR / ADS40 work. Lot's of work, as the total number
of trees was over 17,000 that were checked measured. 14,000 trees were
finally included for SP-indentification tests.
Finalizing the LiDAR_In_Mire study with MK. Some problems with
the feature data resulting in recomputations of classification results.
11.3. Preparing a large observation file. Derived a local model
dcrm = f(sp, d, h) for the expected crown width to be used in collecting
the per tree LiDAR points.
9.3. Continued verifying the 2002-2005 field measurements in
the 2007 images and LiDAR. Positioned the top positions to correct for
missing height increment and errors in the field data. This ref. data
in addition to 2006-2008 ref. data will serve for both LiDAR-SP and
ADS40-DMC-Vexcel-SP -tests.
Compiling of PileUP! 13(1) commenced.
6.3. Reading some interesting papers in FW-LiDAR and normlization
of LiDAR intensities by B. Höfle & N. Pfeifer and Wagner et
al. 2008. It seems that LiDAR is not going to be a toold for retrieving
reflectance of trees / canopy elements. LiDAR-TreeSpecies and LiDAR peatland
classification projects going further. Going thru the 2002-2005 field measurements
to see that the trees are still in place in 2007, at the time of the LiDAR.
Single-tree detection detection using LiDAR pseudopoints,
ground points and points hitting other trees.
3.3. The frequency of weblog notes is low... Highlight of February
was the 9-day holiday to Pelkosenniemi at the end of the month. It
was a bit cold up there, but we had good company and enjoyed ourselves
with the winter activities. We also had a little party to celebrate
finishing of the house extension. Plenty of Sauna activieties in Kulloonkylä
in February. The new sauna surely makes a difference. We now have a
sauna.
Work with LiDAR in area based vegetation analysis continued in
February as well as getting acquinted with new classification tools,
especially the SVM with MK. A few lectures given at MARV4-course.
Best part of Sauna in the wintertime, Arttu
30.1. Finally there with Lakkasuo LiDAR-features. Some applications
also prepared at the end of Jan.
This is the hummock-hollow-binary model of Lakkasuo. A shows
ditches from 1961, B is in the are that forestry student drained in
the 1920s and C shows the ridge-hollow (Kermiräme) part of the
bog.
The XYZ-accuracy seems to allow derivation of the mire surface.
19.1. Haven't been active here. The LiDAR tree species project
is proceeding in parallel with the Lakkasuo project that is very interesting
since mire ecosystems have thus far escaped all research. Working with
MK who's programming the LiDAR feature extraction in Python / R using
Linux. The house project cntd after new years, and the sauna is about
to be finished by Jan 24. PileUP! 12(4) turned
into a 52-page issue, and it kept me rather busy for about a month. So
many little things all the time that it feels as if nothing is really
progressing. Must start with the ADS40 work by Feb 1.
16.12. With the ads40 sensor now in order, started to program
a routine that takes as input a reference tree (field measurement), then
maps it into the images that see it, fits a crown model to the LiDAR data,
computes the angles to the camera and sun, samples the images for pixels
on different sides of the crown envelope, computes LiDAR-features from
the points that make the crown (model) - height and intensity distribution
metrics, then stores everything (R,G,B,IR,PAN, Intensity, angles) into
one record per tree.
The first results in tree species recognition using LiDAR intensity
only indicate that pine separates with 87% accuracy, but spruce and birch
only at 75%. With
some new variables the accuracies were 88% for pine and spruce
and 75% for birch in a test set of 2700 trees, age 30-140.
FINALLY solved an issue with a memory leak in dll-functions that
call MATLAB C-library functions. The leak was roughly 250kBytes each
time a linear algebra solver was called, which made simulations with the
code impossible. The solution was simple, I concentrated on reading the
instructions and did exactly what it said. Declaring all mxArray's first,
then using of mlfAssign() in every assignment insted of "=" is required
for the memory management to work.
Lakkasuo-project is proceeding. Active also with next issue of
PileUP!
10.12. Understood finally the L1-geometry, a bug with the LINES
and NUMBER_OF_SCAN_LINES options. Ray intersertion and back projection
routines now implemented. Need to optimize the ground2image routines for
fast superimposing of vector data such as LiDAR points. Also, need to
check how to implement pair and triplet matching using the L1 imagery.
8.12. The level-0 geometry of the ADS40 imagery is starting to
work. Many routines needed as well as case() statements to choose between
frame imagery and 3-line-sensor in KUVAMITT - the objective is
that both traditional frame images and ADS-images can be used at the
same time for the basic photogrammetric operations.
28.11., 1.-4.12. Working on the ads40 support for KUVAMITT. Wednesday
worked on the LAKKASUO LiDAR data / DEM. Thu: Now got the geometry for
the LEVEL 1 EPIPOLAR IMAGERY somewhat going. Also found c-code
for the WGS84 XYZ - UTM conversions and compiled that into kuvamittGeotrans.dll. This is needed for transformations
between the ADS40 local Cartesian photogrammetric system and the global
system (UTM35 and KKJ) in which the field data is.
DEM of the 2006 Lakkasuo campaign. 
Depending the flying height the intensity values change.
29.11.-30.11. 
@OH2BH. Operated 44 hours out of the 48 possible. 2800 contacts,
140 countries and 38 zones. It was on telegraphy. Listen
to what it sounds like (mp3
recording by K5ZD).
27.11.2008 Working on ads40 support for kuvamitt. Programmed
read-tools for *.cam, *.sup and *.odf files. VBs 64kByte struct limit
does not allow for storing the xy-positions of the 12000 pixels in a
CCD-line (> 96kBytes).
25.11.2008
Winter
arrived with 40 cm of snow and reminded us of xmas.
View to the new "laundry room" and bedroom.
Construction; a bit of painting remains to be done next summer.
Arttu & Perttu doing homework.
24.11. Tue. Had a two-month break in keeping this "diary". Just
too much happening and energy levels low... House project proceeded well
in early Oct, we were 3 working to get things ready for winter. In November
we were finishing the interior. The 4-8-cm thick cement layer has not
dried in 10 weeks to allow for finishing the floors in the Sauna and shower
room. Well, there's time to wait and no reason to start too early and
then later regret. Work: Silva Fennica -article about seedling stand remote
sensing was approved and the final manuscript with alterations in figures
got done. In Oct visited Stockholm for a 3-day trip: evaluation of research
proposals by the environmental protection agency. That was interesting
as I needed to get acquited with underwater remote sensing, benthic video
mosaicking, bathymetric LiDAR etc. ADS-40 imagery and ALS50 -data
from August arrived in late Oct and started to examine them. Made an IJRS
review. Participated an interesting LiDAR-seminar in Evo with most participants
being non-academics. Here's my 45-minute presentation in pdf. Quite a bit of effort in November
by many was put to PileUP! 12(3) which was released on Nov 23rd (link). 40 hours
from its release, Google Analytics shows 1420 pageviews and 910 visitors
from 66 countries.
Visited OH5Z for CQ WW SSB.
25.9. THU. LVTS application writing. The Silva Fennica reviews
had arrived, minor corrections + lots of language corrections from
the earlier review by JiTh. : Recalculated the Monte-Carlo simulations
for estiumating the back-projection accuracy - now using the var-cov
matrices of the aerial triangulation, the correlations between omega,
phi and X0, Y0 had caused a trifold overestimation (14 um vs. 4.5 um)
of the errors (earklier assumed uncorrelated additive Gaussian errors
in the six exterior orientation parameters, which is wrong), which was
corrected when error terms were simulated using the var-cov matrices of
individual images and Matlab's functions (modified mvg-function), the 2007a
release function did not work with the R12 from 2000 that I have but parsing
the contents ofthe mvg.m helped. House: the miniature digger visited in
the afternoon.
In flight 06214 (S-N flight lines), parameters Y0 and omega were
highly correlated. Error in one is compensated by error in another.
24.9. WED Writing the application for LVTS. In METLA in the afternoon.House-project
Insulating the chimney for the required 10-cm thick layer. Cleaning
for tmr's insulation with "used newspaper".
23.9. TUE Looking at the LiDAR features for sp-recognition. Construction:
The roof-makers arrived finally. The very sad news from Kauhajoki arrived
in the afternoon.
22.9. MON Programming of KUVAMITT's Feature extraction.
Per-tree species backscatter / 1st return intensity profiles
down from tree top show interesting species-spesific patterns that need
to be examined thoroughly.
21.9. SUN. Another day with the chain saw. Send Heerburg reference
surfaces for Z-control.
20.9. SAT. Working on the firewood.
19.9. FRI. Programming KUVAMITT for the retrieval of per tree
image features.
Crown surface normals projected in six views. The normals are
along points that are either in shadow ®®
(on the opposite side to the incoming sun rays), in direct light ®® (on that side of the crown envelope
that is sun-lit), or towards the camera. The idea is to sample the
images for pixels in "appropriate positions" on the crown surface.
While I was having lunch I looked out the window and thought
that working from home is not good for mental health. An elk family
(mom + twin boys), were eating apples in the garden! Funny enough, she
did not care at all when I was photographing her from a distance of 10
meters. The other calve passed by me, 2m separated us. This lady-elk
is well known in the village of Kulloonkylä, every year she has twins
and survives by staying close to houses where hunters can't shoot her....
18.9. THU @ HUT for GPS/GNSS-lectures.
17.9. WED. Programming KUVAMITT for the retrieval of per tree
LiDAR and Image features.
16.9. TUE. @ HUT for the photogrammetry lectures by Fabio Ramondino.
15.9. MON. Save copies (since June!). Cleaned the e-mail. Planned
Sept-Oct timetable. Perttu was ill, so was with him. Evening spent at
the construction, we finished the inner wall supports. Got the plumber
for Fri, after that the insulation s.b. ordered. Let's see if the roof
guy shows up this week. Soon it'll be winter. Quite some news from NA; let's
see just deep the markets will dive.
14.9. SUN. Inner walls & supports for the el-heating below
the wooden floor was built for a few hours. Electr. came and worked
for a couple of hours on the wirings.
13.9. SAT. Trying to make some progress at the construction after
week's break. Started to build inner walls. To realize that the "ventilation
expert" had forgotten to connect one 75-mm tube to up and through the
roofs. Made some firewood & putting those wet ones that had been
in the rain all summer in a pile. I hate building: it is costly, no one
really cares but for the €s (experts in this line of business suck), very
few actually base the "knowhow" on true knowhow, which is applicable in
"new situations", i.e. in something they've not exactly seen/done before.
Around 1000 visists @ PU-site that GA has been able to register; don't
really know how many people disable Java or Cookies.
12.9. FRI. In Evo to see beavers. Some OH7-phonecalls. There
in Oct 10. TODO: figure out how all this organized.
11.9. THU. Checking what sort of post-GPS-calculations might
be needed, if so with experts in OH2. E-mail exchange to officials in
SM0 for early Oct. Afternoon: Faculty board meeting.
10.9. WED at HUT for meetings. ADS40 orientation - some obscurities
in GPS Base data? Sorting that out. TODO figure out if Lakkasuo-2006
LiDAR would be processed in 10 x 10 m, or 20 x 20 m cells for Peatland
site / crown cover analysis, discussions at MMEKO. TODO, organize PC, Kuvamitt,
LiDAR & images for LK. PileUP! 12(2) is ready. Not recovered yet.
08:08 UTC onwards, GPS-solution is unstable.
9.9. TUE Completed the 10-page application. Still a bit of fever.
Construction: Heikki came and we discussed the el-wiring & fuse-boxes.
He saved us 2000€ in convincing that we don't need a new outdoor "kWh-measuring-box".
PileUP!-editing in the evening.
8.9. MON Still ill, but needed ti write the UH-3-yr grant application
with its DL being 10.9. Construction: More koolaus-work (roofs) in the
evening. PileUP! editing towards morning hours.
7.9. SUN. In fever / flu. Some activity towards Joensuu.
6.9. SAT "Koolaus"-work. TH3. Shaveling to find the el-wires
to the house (Bosse), made the hole in the old foundations for the water
pipes.
Bosse and the boys found the wire! "Reglar" are straight. Room
for water pipes.
5.9. Fri. In Turku. Concrete covered with plastic. TH came for
a visit. "Koolaus"-work.
4.9. Thu. ADS40-stuff. Floor-concrete work. Off to Turku.
3.9. Wed. Meetings at Dept. Cleaning e-mails. TODO, check what
to do with prof. N's papers & old photogrammetric instruments at
dept. TODO check titles for BSc-projects. Construction: ventilation built
(6h), prepared everything for tmr's floor-work. in the evening with Jussi.
2.9. Wondering how to do the
image i/o as the images can have 85000 lines, divided in sub-blocks
of tiled TIFFs. VB-integer is signed, just like VB-long. L0 and L1
are differently oriented. Need to implement good "tile-arithmetics"
to efficiently manage the images. Untiling L0 images is also an option.
Construction: tried to figure out where the waste water would be going...
Did some leveling for that. We have new legislatation on the handling
of waste waters and it is really difficult to determine what systems is
acceptable (and works) and what is not. Some 10000 eur instruments
that cost to run 300 eur a year (electricity, chemicals, emptying) are
actively marketed here, but other, more low-cost systems do equally well.
You can also plan it by yourself (copy manufacturer's plans from the www)
and thereby save 700 eurin planning costs. TH payed a short visit.
Viljo's supper: herring & chocolate cereals....
1.9. Mon. Learning the ADS40 Info Kit's contents to include the
sensor model in KUVAMITT, eventually. Jussi came in the afternoon and
we installed the remaining "bituliitti" in attic and some eave-boards.
Electr. installed the last heating cables.
31.8. Sun. In Nikkilä for the christening. Pause at the
construction, instead, activities with A, P & V + some editing
of PileUP! 12(2).
30.8. Sat. Another full day at the construction. With Jussi &
Alpo, luckily. Painted the eave-boards, now that it was dryish for once,
laid a round cement block under the "cylinder-oven", isolated the bedroom
floor. Electrician put the remaining heating cables to the floor. Then
we spent 2 hours carrying & repiling the windows that the delivery
guy had dropped nearly in the middle of the gate / road to the house, what
a service by them! Ellu put some pre-painting chemical in the walls so they'd
survive the winter. Viljoa, Arttu and Perttu helped the best they could.
Jussi worked on something up on the roof. Day 4 after surgery...
Jussi and the 72.5-cm, round cement block for the oven's base.
What is to be sauna and shower room floor.
Ladders on the roof. When I'll be heating the oven, it's easy
to find a set of wood that cost me 15€ but ended up here...
The corner with old wall, blocked window, new, 40-cm thick (yes
40 cm) wall, and to-be-floor. Cable for warming the thin cement-layer
in the floor. Let's see how that works out.
29.8. Fri. Cam test, SMEAR II & Lakkasuo e-mails. Oct 21
preparations. Construction: put the "steam-stop-paper" + 50-mm wood
on walls, built 2nd, permanent ladder to the roof, panels to wall, el.
heating cables to sauna cement floor, etc. with Jussi.
28.8. Thu. Cntd working on the summer 2008 field data. Construction:
it rained untill 6 p.m., put some boards on the wall, went shopping
because of the rain. Jussi helped.
27.8. Wed. Worked on the documentation of the summer 2008 field
& RS data. Cntd building the outer wall around the new window + finished
ladders on the roof, with Jussi, again helping.
26.8. Tue. GCP data for 59 signals valid in August 2008 (xls). Discussions about
Hyytiälä as a RS test site, role of the different land owners
/ managers. Construction: Installation of a window, a smaller one put
in place of the older -> new wall. Building of ladder's to the roof.
Safety "bridges" around the house were histed up to 2.5 m - it's safe
now to work on the roof. Iron's for the 4-cm concrete layer on floors
were laid. Ordered concrete for floor & men & ventilation installation.
Jussi came to help.
25.8. Mon. Board meeting of Dept. Meeting with UB. GPS-reference
data for Sat was sent. GCP UTM-KKJ transformations by NLS/JL. Built
a hat for the chimney with cement with Jussi.
24.8. Sun. Perttu's day with some celebration. Construction:
cleaned the extra "uretaani" fillings, isolated the chimney and put
insect nests around the wall-roof connection. The roof-guy promised
(for the 4th time) to come and take measurements and give instructions.
Jussi was there. Wx forecast for Hyytiälä is good, so hoping
fr some DMC-flights in there. Southern Finland looks good on Tue, Hyytiälä
both Mon, Tue.
23.8. Sat. Woke up at 7 to watch EUMETSAT hourly images by SMHI, allmetsat
EF##
wx data etc.
The weather was a bit cloudy at 7, but cleared and ES-MAA was above
Hyytiälä 09:50 when cloud cover was apprx. 5% and stayed like
that for the next 2 hours while ADS-40 was flown to produce 10, 20, 30
and 40 cm data. FGI tarps, people and instruments were in place as well
as Uli's instruments. Everything was a bit moist after night's rain. Spend
a lot of time on phone with cloud-observers that I had on the site...
ALS50 was operated as well. We seemed to have lots of luck in terms of
weather although it wasn't perfect, cloud-free. Let's hope for more wx
tmr and next days. Anyway, a great effort by many people in OH2, OH3,
ES, and HB9.
Tarps and instruments - Hyytiälä football field (Photos
Uli Biesl). Uli's spectrofotometer and reference.
Tarp, a close look. Siemens star (see wiki) and Hyytiälä
1910-buildings in the bkgnd. Photos Ulrich Biesl.
22.8. Fri. ADS is installed and ready. Weather is very unpredictable,
1008 hPa and clouds. Porvoo shows clear sky in the evening, and I'm looking
at allmetsat + all possible sources.Let's see what happens. Very little
to do and help. Construction is not progressing a lot, which is tough
to tolerate.
21.8. Thu. Tried to organize summer's paperwork. ADS40 traffic.
Cleaning at the construction site.
20.8. Back home to wait for better wx in Fri / Sat.
19.8. Hyytiälä. Mapped the new GCPs together with Ulrich
Biesl.(DATA in XLS) Weather
is not looking favorable, some chances over the weekend presumably.
18.8. Mon. ADS-40 arrangements @ HKI/VANTAA & MASALA. Off
to Hyytiälä.
17.8. Sun. Another long, warm and very humid day at the construction.
The southern eave-bords were planned and supports build. Before that,
rest of the old wall-bords were demolished and the bitulit-woord-fiber-coating
was put on. All this was possible with Jussi being around to help. He
has again "brought some hope" with him. Hpefully the wx allows finishing
of the chimney tmr. ADS40 wetaher for next week is not looking too good.
16.8. Sat. Put the eave-boards (below) to the East & North
sides together with Jussi and we'll continue tmr. It rained all day.
The rain and missing roof cover is making me a little worried. Electrian
came and we opened the roof-lock to find a route for cables.
15.8. Fri ADS40 test flight arrangements. Terass floor boards
and pillars were put in place. Chimney maker made it to 4 m, i.e. 1 m
is missing. Roof was covered with a large "blanquet". TH 0.
14.8. THu. Send the job-application. Calls on the ADS40 test
flight. Got everything ready for the chimney-maker. TH 3. Tmr, check
military & extension to flight permissions!
13.8. Wed. Gave the HSS-lecture on RS. Arranged ADS40-flight
permissions.
House: Outer wall-boards now done, floor support / boards in
places. Bought material. TH 8.
12.8. Tue. Arttu and Perttu started school!
Lots of traffic concerning flight and photography permissions
for the scanner flight in Hyytiälä next week. After about 15
calls & help by many people things started to sort out. House project
proceeded: boards on the wall, 10 150-mm holes in the foundations for
ventilation were bored, "isolation paper" was ordered. Isolation (wood
fiber) should arrive tomorrow. In p.m. received a call saying I have a
lecture tmr. Luckily the subject is familiar. The SB-220 linear amp was
fetched. TH 8.
Project in the morning of 12.8. Boys ready for schoolday nr 1.
Perttu and Arttu.
11.8. Mon. Chimney material came. Supports & panels were
sawn and put in place, front-side done. Must call for the "isolation
paper", check the plumber. TH 9+8.
10.8. Sun. A day to recover a little and plan floor sructures
& heating solutions (we checked if it is possible to put both el.
wiring & water pipes in the floors), clean the site & put the
sawn wood in better, rain proof piles.
9.8. Sat. Working on getting the wood-fiber-boards to cover the
walls and bottom of the floor. TH 0.
8.8. Fri. The first really rainy day, and the roof-connection
leaked water, actually water poured in. Well, it stopped raining in the
afternoon and the connection was finished with the bitum felting. It
should hold as the 1st insulation will arrive on Wednesday next week. And
before that, we need to have the boards to cover the walls and the floor,
from below. The tinsmith payed a visit and we agreed on things. The old
roof needs to be redone in the next 5 or so years he said. We also found
a spot that possibly leaks. Material for the chimney-maker was ordered
and he promised to arrive at some point. Roof plates should arrive on the
19th. A bit late, but that's because we thought we'd get them for a better
price elsewhere, which wasn't the case. 1st pieces of woodfiber board found
its way to the walls and floor today, when it rained heavily. Boy was
I tired, day started at 7 a.m. when shopping for supplies. TH 8 + 10.
7.8. Thu. TH 6+6. We cntd building the roof-support-boards and
the connection to the old roof, for which T found a solution that hopefully
works: we do not demolish the old roof 's eaves apart from the old "undulating
plate", a complete boarding, "bituminous felting" and on top of that
the "sculptured tin" to connect the old 31-degree roof with the 16-degree
new roof and its tin plate coating. Two "tin chambers" need to be built
by the tinsmith. Tmr we start at 7 a.m. All day we watched tomorrow's
wx forecast changing every hour or so, guess it's hard to predict the
wx this summer. The rain radar and on-line maps by FMI are great. Ou,
yesterday's wonderment was solved, we change the floor structure, our planner
has used a safety coefficient allowing a 4-cm-thick layer of concrete
to be installed.
The carpenter sculpturing the roof connection. The new roof crest
hits the old window, 15 cm below it, as planned.
Not much of the extension is visible from the garden, but, what's
vertical?
The autumn is gradually coming. It was +15C today, with one 30-minute
shower.
Kulloo has now lots of Centaurea jacea, Verbascum nigrum
and Epilobium angustifolium.
6.8. Wed. TH/6. The The roof is now 96% waterproof after installation
of the plastic roofing was installed. T was a real creeper! Building
of the eaves started the day. Actually a visit to a local timber firm started
the mng. Then we build some scaffolds around the extension. The day ended
in a bit of wonderment - how are we going to be able the build the floor
and underfloor heating on top of a plywood surface? A bit late for these
types of Qs, but that's life and it's solvable - I guess. The tinsmith never
came to visit. And we got a tender from the chimney-maker, the pricing -
shock horror! Absolutely no highlights today despite the fact that I wore
my pink goggles.
5.8. Tue. NLS's new DMCcamera flew over Hyytiälä and
exposed the CCDs... Under a cloud cover. More cloud-free imagery will
be tried. That was good news. The weather is not favoring aerial photography
now and looks as if that the next 5 days are bad as well. Made some purchases
(A-ladders) in the mng, prepared everything for the rooftruss-lift operation
that started at 5 p.m. Terass bottom was built. Tmr need to buy 48 x
120 mm wood as it was missing from the material lists.Today I felt really
worn out. I hope that we can finish the roof a.s.a.p. - after everything
is on the safe side. The highlisght of the day was when I saw the boys
for 15 minutes in the late evening...TH/6.
The crane visited for 35 minutes while T the saw to it that all
rooftrusses were in place.
The 16-degree steep roof needs to be made waterproof next with
a "connection" with the old 31-degree roof.
4.8.Mon. ADS40 correspondence to HB9 and ES. 1Hz-data from base
stns for GPS support now ok'ed.Got the language check for the seedling
stand manu. Working at the construction: last m3's of rubble, sand and
dirt now done. No more shaveling, I hope. Rooftrusses came. Shopping for
supplies. Tmr rooftrusses should go up. Highlight today was when played
Memory & puzzles with Viljo. A & P have learned to swim at the
'learn-to-swim-in-2-weeks' school.
3.8. Sun. Last two GCPs to visit and carry stuff to in Hyytiälä.
On the way back stopped in Orivesi and Hauho. In the evening we went
to Kulloo-stadium to play baseball with the new sets.
A new GCP by Route 66. Botrychium sp.? B. lunaria.
It's fall as Botrychium multifidum was "in flower". Trifolium
aureum in Ilmoila.
Sedum annuum (long time no seen, last time in Unkila/Siitama/Orivesi
in 1995) and mailboxes in Ilmoila/Hauho.
2.8. Sat. Cntd work on the signals all day in Hyytiälä.
Highlight of the day was when I was driving on the Väärinmaja-road
to reach a NLS-geodetic point 88M5041 - I was followed by Sebastian
Loeb!! - on his way to Talo-Mäkelä to the start SS21
of Neste Rally. I also met with an OT (b. 1919) living some 5 km from
Hyytiälä - he had some excellent knowledge of Hyytiälä
forests since 1934. I wish I had more time to talk to him. Lowlight
was at a point in Jokihaarankangas which I searched for an hour or so.
No elk flies, but mosquitos were abundant. Hyytiälä is now
ready for camera tests&demo flights.
1.8. Fri. Checking, building new ones and painting old signals
for the August camera test/demo-flights in Hyytiälä. So another
day with hammer and a saw in my hands. Did the nmorthern part of the
AOI first because tmr Neste-rally will block the roads.
31.7. Thu. Another warm day at the construction. We got everything
ready for the rooftrusses. Left for Hyytiälä in the evening
to join the fun.
30.7. Wed. Father's birthday - he's 79 now. How time flies. I
look at Viljo at the construction and remember a similar 'setup' in 1971
in Pyötsaari, Vehkalahti SE Finland. It was another 13-hour day at
the house with some telephone QRM fm NLS, FGI & Heerburg. All looks
good now. The small-GSD-issue etc. We finished the floor, put some temporary
plywood as floor to make the site safer when building the walls, which
started all right. The 8-m 51x300 mm pieces
were heavy but up they went. Relocated sawn wood between piles.
Got the last drainage pipes connected and tmr hope to fill
in everything as rain is now at sight. Of course as the roof was meant
to built next week...Still a few cubics of rubble to be washed and carried.
Building of the signals started in Hyytiälä, where I need to
go at some point. TH/6.
Final check on the 'wells', they've survived well since 2001
when installed. Wednesday evening and the walls "started to see daylight".
In 1952 everything that was available was used, e.g. there were 20-cm
diameter stones in the foundations, inside the concrete.
29.7. Tue. Email exchange & calls to Heerburg, Kirkkonummi
& Joensuu. The flow of the test is now clear. Radiometric issues
regarding the ability to register stereo with full bit rate at a low
altitude remains. House extension project cntd from 9 to 9: breaking
down the wod surface of the old side, finished all the floor supports,
made the big entrance 60x80 -cm "hole" to carage so that the "new cellar"
is reachable. TH/5.
28.7. Mon. We needed to built a pillar to support the shorter
I-bar as its other end was supported by the 50-year old concrete wall.
Just in case. The I-bars were "glued" with a special, 20 € / 25 kg concrete.
Some 198x50 mm floor supports wereput to place. A lot of rubble was washed
and carried to fill the surroundings, which is now 67% done. Estimated
that around 40 m3 of filling is needed and it is all manual work...TBD
Order the roof elements, check the chimney-maker, plumber! Window measurements
and placement. TH/2. Figured that we can't build the floor quite as planned
as it would mean that the isolation (floor itself) would be damaged in
rain when built before the roof.
27.7. Sun. A day to recover and shop for more supplies - which
you can do on a Sunday, summertime. Returned the RSE proof with 2 comments.
26.7. Sat. +27 and sunny. The site was hot, again. First pieces
of wood for the floor. 6 men carried the 190-kg and 150-kg 200-mm I-bars.
Cntd washing the rubble and filling the surroundings. Another 5 m3 or
so. Isolation put to cover the pipes that are low and subject to
freezing. Bosse came to help. TH/8 as well.
Instruments for cutting the wood. I-bars supporting the 200-mm
thick first layer. Fill with washed rubble, 50-mm isolation.
25.7. +27C clear wx all day. Returned the bitumen-pan and gas
bottles. Washed the rubble, small diameter and the 16-32 mm rubble and
carried it and sand to fill the surroundings. Some 10 m3 (15 tn ?) got
moved by hand in buckets. All warn out in the evening. Tmr starts building
of the floor and other structures, finally, TH.
24.7. Thu The bars came in a van! Digging of the deep-canal cntd.
The chimney-builder came. Washing the site for the bitumen work, which
we did in the evening. TH/4 & Jussi came to help and run the +300C
degrees warm pan. (TODO RSE proofs
arrived.)
The bitumen was laid 50 cm both ways, up the wall and on the
washed rock. 60 kg's of it, 3-5 mm thick layer.
23.7. Wed. Washing the rubble to get rid of the clay & dust,
which would block the underdrain pipes if left. Fetched the pan and
gas bottles. Laid more pipes. 12 m3 of 30 mm stones / rubble arrived.
Digging of the 2-m-wide, 2 m deep canal for the pipes. (ADS40 info arrived,
possibilities for August exist.)
22.7. Tue. Made the concrete "slope" (bevel) at the base of the
moulding so that water would find its way away from the foundations.
This bevel and rock in front of it will be covered by bitumen. 2nd coating
on the bricks. 1st underdrain pipe was laid in the rock-depression and covered
with rubble. A water system pipe was laid on top of it. Tmr need to fetch
the 3000 mm and 2400 mm "metal bars" that support the floor. They weight
61.3 kg/m. The plan is to lay the bitumen on Thu evening. TH/4.
It rained for 2 hours today.. Trustworthy neighbor's good old
(ca. 1990) cement "mill". 1st pipes, the underdrain pipe goes in a sawn
corridor / ditch and here, go under the pillar's base..
Pillars, 1-2 m tall, with 1st coating of plaster. Chimney's foundations.
Sawn ditch in granite.
The (TH's) tool. Bevel. Energy drink.
21.7. Mon. Fetched drainage pipes etc. Returned the broken "grinding
machine" and got my money back as they had no replacement. Ditches
done, quite a job to clean the site from the rock-bits and mud and prevent
the mud from blocking the existing underground drainage. Filled the pillars
with concrete (leveling them to -0.265 m) and started to cover the wall-bricks
with a thin layer of cement. Relocated stuff to make room for the 2nd
delivery of sawn goods tmr.
20.7. Cntd making the ditches - to the bedrock - first 5 hours
with the 60-Euro Honk-Kong-week-ago-bought "grinding machine" untill
its bearings started to make odd noise and I did not dare to use it anymore.
Need to return it tmr to Porvoo Hong Kong (place of quality stuff) as
it has a 2-yr-guarantee! (When I got it last monday, the salesman actually
reminded to come and fetch a new one should it break down). Then fetched
in the aftrenoon, from Hamari, the real-man's tool, a Husqvarna 30-cm
diamond disc with a two-stroke engine. Broom! Made some noise (and mud)
untill 8 pm, when realized it is a Sunday.Even with this 20-kg masculine
tool it was slow. And I no longer wonder why a meter of low ditch in granite
costs 100 eur. The two-day sunshine and me dressed up in a "carpenter's
vest" has made me a genuine redneck, I noticed, haa!
19.7. Sat. The chimney was brought to level -0.39 m (the floor-structure
is 39 cm thick) and the pillar bricks were put to place, with pillar
tops now -3 .. - 1cm below the req. level. Need to level with cement
that goes inside the pillar-bricks. Washed the rock. Sawing of the bricks,
dust, took place far away from the site today. Lesson learned.
18.7. Fri. Finished the wall and first layer of bricks to the
chimney. Bought the bitumen 90 kg, there's a pan in Porvoo for 21 EUR
a day. TH/4.
17.7. Thu. Some 100++ stones to the brick-wall. Delivery of sawn
goods in the mng. Needed to pay it right away... Guess times are getting
rougher. TH/4.
16.7. Wed. The mould was demolished and got rid of the plywood.
Took the nails. Carried the first 70+ bricks. Tmr starting to build the
wall. TBD Recall to order the "location
check" of the house!. First delivery of sawn goods tmr.
.
Late evening photos of the site.
15.7. Tue Bought the needed wooden material - locally. Got it
for 100 eur less than the previous offer. Started to break the mould.
Still had a bit of fever.
Worked 5 hours on a Hyytiälä experiment: "MARV1-2008-Plot-6".
Made the maps & labels for the field team in Hyytiälä.
14.7. Mon. Inspector was here. The cement-truck came and just
before it started to rain we emptied all 5 m3. Started making the ditch
to the rock with the new tools. A big cloud of stone-dust, if anything.
Jussi came to help us. TH/3.
100 iron hooks. Had a better read the blueprint, I would have
needed only 50% in places where the mould was higher than 30 cm.
13.7. Sun. In Turku for a knight excursion.
Arttu the knight.
12.7. Sat. Peer-reviewed the due paper. Fetched the "cement-vibra".
Off to Salo.
11.7. Fri. Made the 4 moulds for the pillar supporting the terrace.
Got the cement-lorry + inspector for Mon.
10.7. Thu. Finished the "iron-work". Cleaned (vacuumed) the rock.
Filled some holes in the mould with cement. 100 hours of put to the mould.
9.7. Wed. Iron-work, 80% done. Ordered first load of bricks &
cement to be delivered on Fri. TH/3 helped a few hours in the evening.
8.7. Tue. Started to put the 4 x 10 mm iron + iron nets in place.
There's a "hook" every 20 cm that's made of 8 mm iron and each one has
different height because of the curvy rock surface. This seems like a
laborious and not-so-nice work phase.
7.7. Day 6. Pertti came & we worked on the chimney's mould.
Made deal about the material & fetched more iron & plywood.
Captain underwear strikes back!
6.7. Day 5. Fetched some sawn wood from Pornainen & tools
from Hamari & a bit of work with the mould.
5.7. Day 4. Still working on the chimney mould, which is on a
20-degree slope. Run out of supplies. Tmr getting some 50 x 100 mm sawn
wood and on monday need to buy the irons, a proper table-saw and more
mould-plywood. We all cleaned the site in the evening. Viljo was fun,
we'd hope he stayed as 3.5-yr forever.
4.7.Day 3 of building the mould.Today it was really warm and
moist, ++25C and the rock reflected all the sun's heat. After 8 hours
felt all worn out. July 4, 1998 we got married in Juupajoki / Hyytiälä.
A bit of celebration with Ellu and the Boys. Ten years and lots of exciting
things in between. Actually quite amazing when you think of it.
Mould on the rock.
3.7. Cntd building the mould for the footing that stands on the
rock. Takes time as the rock-relief is curvy and there are many stairs
to be made. 9 hours and felt really tired, guess office-work is not best
physical exercise.
2.7. Started the house project with the foundations - footing.
What a blast...
1.7. In Hyytiälä, students made last checks in the
field and gave the updated files & sheets. Everything was done
by 12 a.m. Some 2400 trees were measured by the students. Consulted
MiPu for a 2007-orthoimage & old marv1-code. On the way back home
in the evening stopped in Orivesi for some botany. Time for summer holiday.
Had time to visit the Hyytiälä's Artificial-Erkamo-lush-site
near Muistokuusikko: Lonicera xylosteum and Mercurialis perennis.
Glyceria lithuanica & Brachypodium pinnatum in SW Orivesi.
Dryopteris cristata & Carex elongata.
Botrychium lanceolata & B. lunaria.
B. multiflorum, old leaves. Lathyrus silvatica.
30.6. In Hyytiälä, checked from 09 a.m. untill 09 p.m.
data by the students. All went smoothly, minor bugs only to be checked
tmr. All groups were
within 6 hours, after 2.5 days in the field. tmr need to collect
sheets and go home.
29.6. On the way to Hyytiälä, payed a visit to a eutrofic
mire in Karkkila.
I had hoped to see Cypripedium calceolus in flower,
but was too late and it was too early for Epipactis palustris in flower,
so missed both.
Carex panicea & E. latifolium were abundant, inter alia.
28.-29.6.
In Pernå archipelago.
Fishing in Våtskär. "Siikas" from Hamskär, outermost
reef, 20 km from Loviisa. Siika was still in the shallow waters.
Hamskär had rich fauna and flora. Meadows had all colors.
UC-57 submarine visited Hamskär in November 1917. It never
made it back home. Our transportation in 2008, m/s Vera.
Glaux maritima. Lichens on the rock.
27.6. Fri in Hyytiälä. Students start to calculate
tree positions. Updated the www-site to include a "status page". Jari
can start at plot 1.
26.6. Thu in Hyytiälä. Students practiced basic measurements
and learned the concepts in Muistokuusikko and tried resection.exe.
Resection was modified to write just one header line in results.txt
and output of Baarda-outlier-test-statistics was included in the screen
i/o.
Geranium bohemicum in a clear-cut near site 3, a view north of
site 3.
Measurements got smoothly started. In the evening paid a visit
to Lakkasuo in Orivesi.
Salix lapponum, Dactylorhiza incarnata.
Carex livida, Menyanthes, Urticularia. Sphagnum subsecundum.
Carex helenoides. Trichoporum alpinum.
Pedicularis palustris. Montia fontana.
25.6. In Hyytiälä, prepared Thursday's program. Got
things ready for the afternoon demo/lecture. Prepared material for delivery
on Thu.
3D-stuff in the opening lecture.
Makkarakallio's Silene rupestris and Sphagnum compactum
were in the old loci.
1994 sheep pasture, Alnus glutinosa seedlings in places
- or likely glutinosa x incana.
Carex rhynchophysa and Salix myrsinifolia
in their old places in the wet part of the pasture.
Tapiolanmäki was again filled with the fumes of Platanthera
bifolium.
24.6. Send the ADS40-report. House project in p.m. - made some
purchase and comparisons of bids. Got things ready for Hyytiälä
- travelled there in the evening.
Stopped in Pälkäne
for some Fabaceae-botany. Anthyllis vulneraria
survises at the sad sandpit along with Astragalus alpinus and Lotus
corniculatus.
The new camera has got me excited in photography again.
Boy is this different from the old day's of Canon F1 & film.
23.6. Seedling stand paper, added Tuukka's input in Intro + MultVarNormality
test results and send the manuscript to the Journal. Provided an AOI/spec
for a possible DMC-test in August - yet open, DMC is different, PAN-ratio
4.5, 12 um pixels, 69-degree cross-track FOV, f = 12 cm. TODO: compile
the information about the radiometric test/instrumentation for ADS40
(TBD!). Got bids for material, need to choose soon to get the
stuff. Rain, could not work on the project.
22.6. Rainy day so worked on the seedling stand manuscript on
a Sunday.
21.6. Ikaalinen - Tampere - Kulloo.
20.6. In Ikaalinen - Jämijärvi. Juhannus in Finland.
Puonti,
Tevaniemi. Lepola, Jämijärvi.
Jussi
& Boys..
Lauha, the Finnish horse.
19.6.
Finished and sent the RSE-Lichen revision 1. Send, finally, the AOIs
for the possible UAV flights. TODO by Tue; send the confirmation about
the radiometric measurements during the possible ADS40 flight. EiHo will
have the details by then. Agreed that JiTh will revise the seedling-stand
manuscript. Surveyors came in the morning and marked the rock for the house
corners. Send a blueprint of the carage to the construction company in
Mäntsälä. Took full set of backups. Off to Ikaalinen.
Real-time
GPS positioning under a sparse pine canopy. An offset of 4 m.
18.6. Correcting the RSE-Lichen-manu. Called for bids for the
house project materials.
17.6. Started correcting the RSE-manu, visited Viikki to deliver
photos for forestry students for the 100-yr-jubileum. Visited
RakVi in Porvoo to approve a 35 cm offset of the house extension.
New prints needed, another waste of €'s.
16.6. Made the field sheets. TODO specify AOI for UAV!. Worked
on the house project to figure out if the roof can be made (to last)
in the blueprinted position. Needs an offset of apprx. 35 cm.
13.6. Ordered the maps and prints (folder).
Delivery on Tue or Wed. Final editing of the course instructions for
MiPu. TODO field sheets with column ordered serials. TODO prepare pine
plots P1-P6 for the field group of Hyytiälä forest station.
These guys turn 7 today!
11.6. In the field to position poles, 9 per plot, for MARV1-plots
1-3. Plot 4 remained for Thu.Programmed KUVAMITT-
HYDE to be used @ the field station over a remote IP-HDD. Transfer
over the network was slowish, maybe 2-5 MBytes/sec, which is too slow
for the 20-30 Mbyte binary LiDAR files but reasonable for the 1-2 Mbyte
screen image I/O.
10.6. Recovered and left for Hyytiälä. Prepared maps
and labels for LauK's field campaign & MARV1-stuff.
9.6. At home, ill. In the morning in Viikki for a MARV1-meeting.
8.6. Porvoo, Laversintie-summer-get-together...
7.6. In Orivesi for the Korppoo family-meeting via Salo.
View
from Korppoonvuori
6.6. Finished MARV1-preparations: Measured plot 5, in Muistokuusikko. Programmed Resection.exe to output
solutions to a ListBox on the main Form. Also, changed its I/O for
control point observations. Programmed to LabelPrinter for LauriKorhonen's
campaign on MARV1-2007 plots. Made the field sheets (files) and some
instructions. All documented and made available here.
5.6. Wrote routine Plot_LiDAR2PDF_inside_a_polygon_Click() that inputs a polyline and finds
LiDAR points inside the polygon
and prints a pdf-map of the intensities or heights of points
in a given height-range.
4.6. Got the labels & maps ready for MARV1. Documented the
work. Preparation of data sheets remains to be done +
updating the written course-instructions
& programs.
3.6. Physense-seminar downtown Helsinki. Interesting talks about
radiometric issues, BRDF & light-canopy interactions.
Lauri & Jari will join forces
for field observations in Hyytiälä.
2.6. Send ADS40-AOI. Finished measuring stand 4 for MARV1. Programmed
the "LabelPrinter" and "MapPrinter". Changes
needed mainly because of the new cluster
plot arrangement (sub-plots) and the new vars on the Labels.
Got plot 4 ready, map looks like this,
measured 60% of trees outside the 100-m x 60-m plot with a 5-m buffer:
In all it took
5 hours to carefully measure 1500
trees on this plot. Still, it wold be better to have an assistant for
stuff like this...
1.6. Measuring stand 4 for MARV1. Made the slides for the Physense
seminar talk on Tue.
May 2008
30.5.
TODO send ADS40-AOI. Contacts on Monday. Measured 2 candidate stands
for plot 4 on MARV1, first
one is too dense / difficult. 2nd one I measured from 2006 images
only to realize afterwards that it has been thinned
in 2007. Try with a 3rd one on Monday. Rcvd Bosse's comments
on the seedling stand paper. Made reservations for Hyytiälä
trip.
P&A's last day at kindergarde.
29.5. Meetings in Viikki, phoned Leica-people, SSR-meeting ni
p.m.
28.5. Measuring stands 2 & 3 for MARV1
27.5. Finished the Gradu. Sent the review.
26.5. Traveling back, Gradu & RSE corrections.
@OG8X
24.5.2008 20:45 UTC in Arkala
Touch-screen antenna control.
SO2R ant switching for 6 towers.
23.5. Traveling to Oulu, reading & commenting the Gradu.
22.5. Programmed KUVAMITT and started measuring the MARV plot 1 (50-yr-old
Pine + understorey, 120 x 120 m).
KUVAMITT now outputs the full record of features per measurement.
Off to Helsinki to get mail & print the thesis on
paper + get the language review by JimTh. House extension project
continued.
21.5. Save-copying / archiving / documenting to the two 750 GByte
USB-HDDs and the 40GB-USB-HDD.
Reading the Pro Gradu-thesis, first 15%.
20.5. Implementation of feature extraction for SP-detection continued.
Awaiting for Bosse's comments before tackling the manuscript. Agreed
to review a RS paper (due in June 18). Received a MSc-thesis to be graded
(ASAP!). TODO, send JaHei the data.
Based on Phase-angle, Off-nadir angle and Azimuthal angle (in
XY, between camera ray and sun ray), the crown is sampled from the sun-lit
or shaded side, or sampling is not done if its a side-view.
Samples of ca. 50-year-old pines (1), spruce (2)
and birch (3), with corresponding LiDAR and Image (Vexcel, 28 cm GSD)
pixel-features (point a meter down, on the surface of the crown model).
When images are taken 2300 m above targets, pixel GSD is ~ 60
cm, the separation in NIR-shaded is no longer as good as it was from
1000 m imaging altitude and 28 cm pixels. Broadleaved trees are seen "a
bit brighter" in the front lit, 2.3-km-view.
19.5.
Continued work on the "monoplotting treemapping" algorithm & at
the same time tried to figure out a way of
doing PAN-sharpening of the Vexcel images for "proper CIR-images".
For SP-recognition tests (to be included in the MARV1-tree records):
Added image and LiDAR feature collection: the images that view
the front-lit and shaded side of the crown are
sampled for R,G,B,NIR,PAN -values, a little down the crown, LiDAR
intensity metrics (range, agc, sensor -normalized) are computed using
the points that were used for crown modeling. The croww modeling is carried
out by giving more weight to positive residuals(default 10 x) -
that way the surface seeks its way to the outer edge of the point cloud.
18.5 Weekend spent with the house extension project.
Photogrammetric relief of the rock on which to build?
16.5. Agreed on UAV in July. TODO provide
maps / orthos of the AOI.Started to debug the PAN-sharpening algorithm,
which is not satisfactory (Vexcel NIR, R, G + PAN) => "CIR". Tried
with the: RGB to IHS Inew from PAN, InewHS to RGB -approach. VexCol2CIR16BIT.exe
After 5 hours...
Many hours later
The histograms are skewed to the right, in the image above, R
and G (G and B) were truncated to Max*85%.
IR (R) is less skewed. A high percentile would be more informative
than min - max.
15.5. Programmed KUVAMITT for needed routines in measuring the
MARV1-trees with different algorithms. Programmed CHM-filter to output
a CHM with combined 2006&2007 data.
Started work with Plot1, documentation is here.
Faculty board meeting in p.m.
14.5.
Continued work on the "monoplotting" treetop positioning algorithm.
A 3rd option to select the candidate is to select a point along the camera
ray that is the same distance away from the camera as the highest LiDAR
point. We know the direction of the LiDAR pulse, it might be usable information
as well.
13.5. Wrote the VB-code into C int _stdcall MyFunc_Cast_Image_ray_and_Find_Closest_LiDAR()
The
execution using malloc-realloc and reading the BIN-files with one call
to fread() became 0.2 sec/tree (45 MBytes of LiDAR data). In VB parts
of array cannot be read with the get-function (no memcpy).
There are always two solutions to the tree top, the highest LiDAR-point
in the "ray-tube" and an XYZ point, that has been lifted by dZ to compensate
for LiDAR underestimation. Here height-growth of trees is seen (Up
and Low -Middle 2002, Low-Right 1997).
There's memory leak of some sort, each call to the dll-function
adds 4K to "Mem Usage" in Task Manager. After a couple of hours, the
leak is isolated to the routine being called, but still get a 4K memory
leak each time a tree is measured...
Maybe some of the options cause it (_cdecl, _stdcall, mixed usage.
This because of Matlab-calls from the dll)
12.5.
Now with the seedling manuscript being reviewed, it's time to test
the old idea of monoplotting. 1st tests:
programmed a "ray-tracer": operator points a tree top in one
image, this ray is cast down to the scene, and LiDAR
points around Z-sol ± range are tested for line to point
3D distance.
Here, LiDAR points inside a 1.5-m wide
"cylinder" around the image ray. Some treetops were pointed in
the left view, some in the right.
Now, it is possible to look for the highest points that is, say
max 0.5 m from the ray's path. The tree top XYZ is along the camera ray,
but the Z of the LiDAR is likely biased down.
We know that
i) the top is along the ray (yellow)
ii) XYZ of the highest LiDAR (red)
iii) d is the 3D distance (perpendicular to the ray)
XYZ has offsets in X, Y and Z (3 unknowns), dX, dY, dZ
Infinite number of combinations (dX,dY,dZ) that bring the
LiDAR
point to the ray's path, where d = 0.
Assume dZ is negative, i.e. need to pull it upwards, say, 0 -
1 m.
This splits the image plane in half (oblique view).
This still does not help a lot...
If the view is near-nadir, then dX, and dY can be computed from
(dx,dy), but dZ remains unknown => we miss the tree top.
Or we fix dZ to -0.5 m (empirical), and compute dX, dY such that
the point is on the ray's path.
14-18 had a break, the tower was dismantled /
more room made for the digger to work next time it comes. Pertti,
2PM offered help.
Here dZ=-0.65 m, and treetop is positioned
at the ray's path in XYZ (white dot) by solving dX, dY. The highest
LiDAR is the HSV-colored dot.
Started coding the Monoplotting routine
in C for a faster implementation, alternatively, the LiDAR I/O is avoided
by bookkeeping of the current position. Reading the 20 + 30 Mbyte, 1-ha junks
of binary LiDAR with VB's get-statement in a loop take about 5 sec
1st time. HDD's own memory helps a bit on later accesses.
in the field. TODO Tue - contact JaHei.
8.5 Planning field campaign 2008, wrote instructions for
July-MARV1-campaign. AU's suggestions came in mail.
at 2 p.m. stopped / got the digger for tomorrow
so needed to work at the house project.
7.5 Gave the CBU lecture 09-12, pdf.
Worked on the manuscript checking references and making corrections
suggested
by TT.
6.5 Finished the CBU-lecture. Finished and sent the Peer-review
report. Mailed to AU in Hyytiälä about the 60x60-m
MARV1-plots. Meeting TuT in Viikki & fetching the
tacheometer for a Hyytiälä-trip around May 15.
5.5. MMVAR-board meeting in the a.m. At home 12 am. Cntd
peer-review for 1 h, boy they are taking time,
I should be less accurate. Prepared the CBU-lecture for
3 + 1 h.
Got finally the carage in pieces.
What remains is demolishing the footing / foundations.
2.5. Finished the raster analysis for section 3.4. Wrote
section 4, discussion of results and conclusions. Proof-
reading. Tried contacting people in Metla/Vantaa. Contacted
Vexcel for details on the radiometric properties of
the UltraCAM D. The antivignetting gain surfaces =>
optical & mechanic antivignetting. Exposure is "simoultanoues"
within 1-2 ms in the used camera, which is 0.1-0.2 m given
airplane velocity. Got the specral responses & other
interesting papers. TODO-list,
see below. R-implementation of BayesNaive (see e-mail). Got answers
for the MTA
1/2008 seedling stand paper.
April 2008
30.4. Received RSE notification and 3 reviews for a manuscript.
JiTh will look at the language and after that
make corrections. "Snail mailed" them (a phrase learned
today).
Reprogrammed the GridFiller, no rotation of the grid anymore,
0.7 m pixels (bugs in the inclusion
tests!). Updated
sections 2.6 and 3.4 in the manu. Looked at the multinormality
tests for LDA by TuT. A good link
for LDA (MANOVA)
assumptions.
TODO 1) Need to come
with a title for the Jubileum seminar talk (Global RS!) by May
15.
2) Check if MARV1 participation
in 2008 makes sence and agree a meeting in Viikki
3) Harrash JaHei about
the possibility of UAV-photography over some sites in Hyytiälä.
4) May 7, CBU lecture,
!!
5) Review - complete
and send by May, 7.
Aerial view + LiDAR, 225 m2 plot in Stand 1
Ref. and classification rasters in Stand 2.
29.4. Learned to know more about this UAV in the a.m. in
Helsinki.
Programmed the LDA tools for the raster analysis + creation
of the reference grid and did some initial
tests in stand1. Low vegetation turns into conifers. TODO
Check cells delineation, starting from the grid,
it does not have to be rotated, just makes everything
complicated in later inclusion testing!
28.4. Mailed the MANU for JimTh. Programmed "GridFiller"
that fills the plot in each seedling stand with
a rotated grid, assigns Zgnd from the DEM. Altered the
"LiDARFeatureExtractor" in KUVAMITT such that
HliDAR is the highest point inside the circle, INTnorm
its intensity and PGR as previously. Modified "Pikseli-
irrotin" to extract image features for the local max LiDAR
points in the grid. Remained programming the
reference grids from the field tree data.
25.4. Prepared the manuscript for JimTh (LangC), Some
work on the seedling stand manuscript but very tired
from all the action - (high fever in the evening). TODO,
TuT send the MultVarNormTest results. Check on Mon.
Tue 29.4. at 08:30 -> Viikki / meeting
Visitors loading the PU.html file in OH.
24.4. Travelled back / PEER-REVIEWING.
23.4. Travelled 05 a.m. - 1 p.m. via Munich-Bremen to
Oldenburg and Bremen.
Hansa LuftBild Award handed by Dr Schroth.
22.4. Continued with LDA calculations & writing of
sections 2 & 3. Got things ready for the Oldenburg
trip. Informed Faculty about the activities there. Correspondence
with TuT.
TODO Send the LiCHEN manuscript for language
review ASAP for JimTh.
21.4. Formulated a Q about the BDR-angles to JoPe.
Found & fixed a bug in the 16-bit 4-channel ortho
output's filepointer statement (LK).
Wrote the structure for section 3 (Results) and continued
work on the manuscript.
18.4. Consulted HaRi on the Fisher LDA. He suggested that
the use of the covariance structure be included & the
optional use of Logit models (Bayesian approaches
JaPel (RKTL)). We will estimate COV-structure from the data.
Copied all data and latest programs to the
\Taimikko\ -folder.
Stand6 classes (conifers / broadleaved / other) x shadow.
image features.
Stand 6, density of LiDAR 6-29 pulses per m2.
Read an interesting document about small, personal UAVs
with digital camera as payload.
Got the SAS ods printer ps... clause to work for proper
Box-Whisker postscript output.
Worked a little on the manuscript.
Phoned JoPe to consult the two obscure planes that are
needed in BRDF representations; be back on MON with a
well structured e-mail.
TODO : Make a final execution plan on what is needed and
if the manuscript is somehow split.
Started another project-blog.
17.4. Examined further how the SAS proc discrim output
(different matrices) can be applied in LDA and QDA.
SAS reports a 96.04 % accuracy for LDA, but when
I took the COV-matrice & Mean vectors and used them in
Excel, the accuracy was 95.7% (without cross-validation).
QDA (the use of per class VarCov matrices)
gave an accuracy of 93.1 % only.
Tested the BINARY "In Shadow, In Direct Light" -classification.
For quick display, made a routine create_vexcel_bmp in
KUVAMITT which takes the 16-bit images and displays
a 3x8-bit RGB image, using the discriminant function it was
quick to illustrate the classification:
RGB 
Shadows
CIR 
3+1 classes
Fused the LiDAR and Image features into one array of observations.
Looked at the per stand height distributions per species
- height separates species - to some degree.
Save-copies.
16.4. In Viikki for UPJ discussions. Discussed with TT/AK
about the possible Bayesin approach to the use
of a priori auxiliary information in the RS task in seedling
stands. Use of models to create standwise "h-sample
trees" to be used with the HLiDAR -feature. Met with LK
=> Image segments per tree => Image Features; LiDAR
features from the LiDAR_Ortho_Images => Classification
in SAS/SPSS.
Programmed the capture of additional "image feature" in
program PIKSELI_IRROTIN, phase-angle (0º for back-
scatter (hot spot), 90 for forward-scatter), it varied
from 33 to 89 in the seedling stand / BFI 06214 imagery and
it seems to affect the DN values of certain species...
DN-values vs. phase-angle, PINE.
SPRUCE.
BIRCH.
RUBUS (note minimum).
Graminaceae & Cyperaceae.
Rock & Soil.
Got familiar with the discriminant analysis in SAS and
TuT's sample code. Made the shadow classification -
6 variables, 96.04% accuracy (kappa 0.92).
The orthoimages by KUVAMITT need flipping in other software,
fixed this "BUG". And send LK new orthoimages.
15.4. Continued work on LiDAR2Image program for
LK.
Hmax (2007 & 2006 data, normalized
to local DEM)= CHM
Max of intensity, 2007 normalized to 2006,
any point in cell
Intensity of point with max H
Orthoimage BFI01214, NIR-channel. "16bit"
Mean of intensity. All LiDAR points in the 1-m cell
SD of intensity all LiDAR points in the Cell
Orthoimage, theme: "Number of LiDAR points in a cell"
Orthoimage, CIR 8 bit
14.4.
Department board meeting; there'll be courses given at MMVAR in the next
years. Let's
see how is the student intake this year, with all
the recent negative media around forestry& forest
industry. 8-11
Agreed to review a paper from uk. 1st readthrough.
Started programming for LK the LiDARPointCloud2Image
-program for LK.
- Rectangular AOI, Raster Size, Hmax, Hmin,
HRange, HSD, IntMax, IntMean, IntSD
Soon realized that there's a need to normalize
the intensities of the two LiDAR data sets. Made
a routine in KUVAMITT that after image matching
of an XYZ point, finds the closest LiDAR2006 return,
and then the closest LiDAR2007 return for
that one and outputs this point pair. R2 of RAW intensities
was 0.54 for a set of 145 pairs with distance
less than .25 m. After normalization for AGC and range,
R2 was 0.66 (IntNormTo2006 = 0.558*IntNorm2007).
However, the range normalization power was 5 for
the ALS50and 2.7 for ALTM3100. The point
set included observations from vegetation, road surfaces,
crowns etc. Was looking at both 2006 and
2007 imagery while trying to find varying targets. The routine in
kuvamitt is GetLiDARObs().
Remember returns(4).z = first return,
returns(3).z
= 2nd return, returns(2).z
= 3rd return, returns(1).z
=
last return in the LiDARRecord -struct (VB
Type).
11.4. Tested what values we get for the TIN nodes (used in TIN2RASTER)
from the 1-m raster models,
the SD of differences was from 0.03 - 0.07
m,that is the "interpolation error".
Finished the sections "Methods - Terrain
modeling" and "Results Accuracy of DEMs". ANOVA
of DEM residuals to explore the inherent
noise in the DEMs.
Chased a bug in the calculation of HLiDAR3-feature.
Weighting of observations was wrong. TuT checks
his algorithm.
Made a simple program that outputs the "canopy-DSM",
realized that the interpolation scheme in
getheight() and getCHMheigh() did not work
if the cell-size deviated from "the usual" 1m resolution.
Fixed this bug and now any cellsize should
do, let's see.
0.5-m raster CHM (DSM-DEM), stand 3.
10.4.
Incremental Delaynay example
in Java (via TKK/GIS), O(sqrt(n)).
Implemented, finally the bilinear interpolation
in KUVAMITTs getheight(), interpretation
of raster surface value. It reduced the SDs of the
seedling-stand DEM residuals by about 20%
=> everything needs to be recalculated! This
affects some earlier interpretations of DEM
accuracy as well.
Faculty board meeting 13- V-P was awarded the best
dissertation 2007 prize. He gave an excellent thank
you speech. Much better than Dean's introduction
was...
Nearest 
Bilinear
9.4.
Meeting in Helsinki / Kumpula @ MK's office: Hyytiälä-100-yr
jubileum. (11-14)
Discussed the "extension" of the Hyytiälä history
(80 yrs) to include last 30 yrs
and a section for forests / lanscape changes. A special
issue with invited papers
for research in Hyytiälä (SMEAR-type of, Peatlands,
Remote sensing, ..) Two optional
journals at this phase.
Organized a little "party" @ MMVAR to celebrate the award and Joensuu
D-thing. (14-16)
ANOVA of DEM errors & accuracy of H_LiDAR for different
targets.
Tested postcript-printing in SAS (Box-whisker, Histograms)
for high-quality prints.
8.4.
Applied the faster TIN-filter to estimate 4 x 6 DEMs for the seedling
stands.
Tried to find examples in TIN data structures and
algorithms, for local point insertion
deletion and TIN-update.
7.4.
Ordered an 8/1 Mbit ADSL connection to replace the 1M/512k and improve
home office infra.
Continued testing the DEMs in the seedling stands
to include 4 different data sets as inputs
for the DEM-estimation: 2004, 2006, 2007 and 2006_2007.
Realized that need to improve
the TIN-filtering's o(n2) complexity.
Implemented a simple lookup table of triangles that the node
belongs to o(n), requiring storage of max_n
* sizeof(int) * n_nodes, which is ~ 60 Mbytes for a TIN
of 1 million points corresponding to 0.5 - 1 km2.
Now the complexity of filtering is o(n) and its
fast.
Tried to figure out why WLS in KUVAMITTs crown model
fitting would not work, i.e. weigths did
not affect results of parameters. 3 hours and then
I realized that I had misspelled "weights.txt",
it was different in the VB's Open -statement
and dll's fopen -call :)
4.4.
Made the stand-level 1-m raster DEM by combining 2006 & 2007 LiDAR
(15 p m2). In comparison
to 2004 LiDAR-DEM (1 p m2), which was modelled with
TerraModeler, there was no significant
improvement, the noise is from 10 to 25 cm, and
the estimation methods may induce offsets in the
order of +/- 5 cm. The PAN_Mean channel alone was
able to separate points in direct light / in shadow
with apprx. 94% accuracy, when all images / stands
were combined.
TT consulted seedling stand modelers on the latest
models. We need to figure out a way of
implementing a priori knowledge into the interpretation
task. Remains to be sorted out next
week.
View from OH8 for EMOTATOR -rotators.
3.4.
Made an experiment with the shadow / light points in the 06214 images
/ seedling stands by
using triple / stereo matching and measuring XYZ-points
in shade / in light. NDVI is insensitive
to shadows. The "type of shadows", measured in spectral
DN-values, varies from stand to stand,
somewhat.
false-color 
(G R NDVI)
Looked at Box-Whisker plots of the Image and LiDAR features.
Tested 2006 LiDAR data in DEM interpolation, results in
-0.12 m bias, 2004 + Terra => +0.11 m.
Same SD / RMSE.
Antenna map
for analog current meters... (altered AzimuthMapper).
Distance = Sqrt(Distance) · constant.
2.4.
Recomputed point - pixel image features & LiDAR features for the
1492 observations. Extended
the list of LiDAR-features (Subroutine in KUVAMITT).
Installed SAS 9.1.3 on the old W2K-machine.
Think about dividing the plots into 2.5 x 2.5 m
squares (~150).
Some minor obscurities in the data that I have from
June 2007.,
How were the heights actually measured, e.g. some
Luzula p. show heights upto 0.6 m?
1.4. Continued
with revising Chapter 2, Material and methods. Received the data for
SAS from TuT.
Now Ellu is sick...PU.html is now accessed from
50 countries, 1811 times since release of PileUP! 1/08.
March 2008
31.3. Started
to work on the seedling stand paper, still unfinished, rewrote the Introduction
to find
a full section of copy-paste contribution...
28.3.
With Viljo who's getting better 8-15. 15-18 Discussed with TuT about
the manuscript. Stand-level
classification of "operational classes" with normalized
intensities: I-norm = I-raw * (R/Rref)^2.5.
|
46
52
89
