This image illustrates various components (trees, tree clusters, old logging trails, etc.) that can be detected prior to harvesting operations. The idea is to support harvester operators work to execute environmentally sound operations. Image by Omid Abdi.
Individual Tree Detection (ITD) automatically identify and extract single-tree features, such as tree locations, species and crown characteristics. We have achieved significant process in ITD by fusing data from various sources, such as Airborne Laser Scanning Data (ALS), RGB orthophotos and multi-spectral images.
Our recent paper proposes a Super-Resolution (SR)-based ITD method to predict individual tree locations, delineate crowns, and classify species from open-source half-meter multi-modal aerial data (Mao et al. 2025) . In Boreal forests, roughly 70 % of individual trees can be detected with open-source data.
In forestry, information of the soil properties would enable forest managers to prevent soil damages, improve forest growth, biodiversity and soil healthy by adjusting the correct timing for operations, choosing the right regeneration method and most suitable silvicultural system for the given soil type.
Our research group has taken significant breakthrough in digital soil modelling. Our recent study introduces an advanced GeoAI approach that combines graph neural networks (GNNs), LiDAR, and satellite data (Sentinel-2) to predict fine-grained soil particle size distributions (PSDs) with high accuracy. We are currently expanding this methodology to include other key soil properties, such as the thickness of the humus layer and the distribution of coarse-grained soils, in the new GeoAISoil-project funded by Finnish Impact Foundation (Abdi et al. 2025).
Logging trails are a vital component of modern forestry since they provide the crucial avenue to extract wood from forest to the roadside. We have been investigating this dilemma from numerous aspects. We have introduced a method in which old logging trails are segmented using a novel U-Net Convolutional Neural Network and high-density laser scanning data. Moreover, we have developed methods on how GNNS track records from forest machines can be exploited to define centerline of the logging trails, number machine passes and goodness of the logging trail network. This info linked with environmental considerations will construct foundation for future logging trail design solutions.