Latvian Council of Science Grant Agreement No. lzp-2024/1-0484
Project partners: LSFRI "Silava" (leading partner) and Institute of Electronics and Computer Science.
The project is aimed at the development of new methods for the estimation of hemiboreal forest resources at a stand level (species, tree height and diameter at breast height (DBH), undergrowth and forest floor compartments, dead wood, leaf area CO2 removals, evapotranspiration) from very high resolution (VHR) remote sensing (RS) data collected using unmanned aerial vehicles (UAV).
The main objectives of the project are:
- To develop a new classification method of tree species from hyperspectral (HS) data. Species which are common in hemiboreal forests and important for Latvian forestry companies will be targeted: Scots pine (Pinus sylvestris), Norway spruce (Picea abies), birch (Betula pendula Roth and Betula pubescens Ehrh.), grey alder (Alnus incana Moench), black alder (Alnus glutinosa Gaertn.), Eurasian aspen (Populus tremula), pedunculate oak (Quercus robur) and European ash (Fraxinus excelsior) with “dry tree” and “other” classes added for completeness. The method will be based on the 3D convolutional neural network (3D-CNN) using a few-shot learning concept enabling classification with a limited amount of learning data;
- To develop a methodology for incorporating UAV-based imagery and LiDAR data for estimation of forest stand parameters, including accuracy assessment compared to traditional visual stand-level forest inventory;
- To enhance the accuracy of these estimates, address the impacts of species-specific canopy cover density on a tree and stand attribute estimation uncertainty based on effectiveness analysis and differences in the image and LiDAR-based canopy height models (CHM);
- To analyse and compare newly developed methods during this study with the popular methods used for tree species classification and estimation of numerical forest inventory parameters.
Actualities
- 09.2025. So far, 24 forest stands have been selected within the scope of the study, including equal numbers of Latvia's 8 most widespread tree species. In each stand, two sample plots have been established, where reference data collection has begun according to the developed methodology, including information on living trees and deadwood. In all these plots, remote sensing data were collected once during the summer season of 2025 — high-resolution multispectral, LiDAR, and hyperspectral data — as well as their initial processing.
Based on literature studies, LiDAR point cloud classification has been initiated, along with the identification of individual trees, determination of crown area and height, as well as a comparison of different LiDAR data acquisition methods and settings, in order to obtain the most suitable data for this and future studies.
In the second half of 2025, reference data collection, remote sensing data processing, and the testing of initial models will continue.