ESTIMATION OF TREE VOLUME USING UAV LIDAR DATA

Abstract

 Remote sensing is rapidly establishing itself in forestry as a means of determining forest resources more efficiently and in greater detail, so it is necessary to assess the extent to which this technology can replace or supplement traditional field measurements. This study evaluated the suitability of UAV LiDAR data for determining the volume of Scots pine (Pinus sylvestris) trees.

Field measurements were taken in six pine stands with different characteristics, and high-density LiDAR point clouds were used to create a canopy height model (CHM). After comparing several canopy height metrics, the 90th percentile of canopy height (P90) most accurately matched the field measurements and was therefore used to select the most suitable buffer zone. Subsequently, a broader set of metrics derived from the CHM was applied to model tree volume.

Based on the CHM metrics, several regression volume models were tested, and their accuracy was evaluated by cross-validation using plots, applying a training scheme where one plot was left out of the training sample during each cycle (leave-one-plot-out). The most accurate results were obtained using a Neural network, with an average volume error (RMSE) of about 0.30 m³. The most significant systematic underestimation was found in the youngest stand with the smallest volume. Meanwhile, in more mature stands, the model errors were smaller and did not have such a clear directionality. The analysis also identified isolated outliers that had a greater impact on the overall distribution of errors, but their removal reduced the overall volume deviation. After combining the data at the stand level, the errors decreased and the overall volume deviation remained small.

The results show that UAV LiDAR data can accurately reproduce tree height and provide reliable area-level volume estimates, but it is recommended to perform a local accuracy check before applying the method in new conditions.