Search

The work was performed in the old-growth linden-spruce forest of the Kologrivsky Forest Nature Reserve (Kostroma Oblast, Russia) based on aerial photography with a quadcopter. Automatic detection algorithms made it possible to detect most of the trees in the forest canopy. Tree detection by orthomosaic using neural network algorithm ‘DeepForest’ turned out to be of better quality than detection based on the canopy height model using an algorithm based on the sliding window method. As a rule, both methods showed better results for conifers compared to deciduous trees. Comparison of the average heights of trees estimated from remote data and measured by ground survey did not reveal significant differences. Additional ground surveys to assess the quality of undergrowth detection are needed.

In this paper, we investigated three pine (Pinus sylvestris) forest plots (each of 50 × 50 m), different by age and composition, located in the Prioksko-Terrasny State Nature Biosphere Reserve (Moscow Region, Russia). This study was aimed to evaluate the forest stand attributes based on the photogrammetric point clouds and canopy height models (CHM). For aerial photography, we used the unmanned aerial vehicle (UAV) quadrocopter DJI Phantom 4. At the first step, we used Agisoft Metashape software for the building of dense photogrammetric point clouds and orthophotoplans. Then we used the lidR package in the R environment for processing of dense point clouds. We used a cloth simulation filter for classification of ground points, spatial interpolation algorithm tin for creating a normalised dataset, and the algorithm lmf (local maximum filter) for individual tree detection and tree height assessment. For accuracy assessment, we collected field-based data, and calculated recall (r), precision (p), and F-score (F). Finally, we calculated CHMs (30 cm/pixel) derived from dense point clouds using the pit-free algorithm. To address the value of UAV data for delineating tree crowns, we compared the outputs of CHM data using two common algorithms (watershed and region-growing), and the result of manual orthophotoplans vectorisation. We obtained a high accuracy of individual tree detection. The algorithm found 46.7% to 87.5% of trees accounted on the sample plots by the field-based surveys. The recall (r) value varied from 0.5 to 0.9. The value of p varied from 0.9 to 1.0. The F-score, considering both factors (p and r), varied from 0.7 to 0.9. The highest accuracy was obtained in the site with a single-layer stand, where large trees with well-distinct tree crowns dominated. Spatial heterogeneity of tree stands reduces the accuracy of tree detection. We also found that tree heights estimated on the dense clouds were well matched with tree heights measured in the field. This dependency was described by the linear regression of y = 0.99x, R2 = 0.99. With both the watershed and region-growing algorithms, the total crown area estimation often exceeded the results of orthophotoplans manual vectorisation, where differences reached 25.1%. Differences between two delineation algorithms varied from 0.2% to 19.7% for the same sites. More accurate results were obtained for plots with lesser density of tree stands. Overall, our results have shown the potential of using photogrammetric point clouds for estimating tree attributes (heights and density) in single-layer pine stands. Widely used tree crown segmentation algorithms do not provide reliable estimates of the crown projection area, and more accurate results could be obtained after further improvement of the technique.