Estimation of stand parameters with unmanned aerial vehicles (UAV)

The sustainable management of forest resources requires the accurate, rapid, and cost-effective determination of stand parameters. Due to the temporal, spatial, and economic limitations associated with conventional ground-based measurement methods, remote sensing technologies have emerged as a promising alternative in forestry applications. This study investigates the effectiveness of Unmanned Aerial Vehicle (UAV)-based Light Detection and Ranging (LiDAR) systems for estimating key stand parameters, including diameter at breast height (DBH), tree height, crown width, tree count, basal area, and volume. The research was conducted in even-aged and uneven-aged Scots pine (Pinus sylvestris L.) and Nordmann fir (Abies nordmanniana subsp. bornmulleriana ) stands located within the boundaries of the Bursa Regional Directorate of Forestry, Türkiye. A total of 212 sample plots, ranging in size from 400 m² to 800 m², were established. Ground-based measurements were performed to collect DBH, tree height, crown diameter, and spatial coordinates for each tree. Simultaneously, high-density LiDAR point clouds were acquired using a DJI Matrice 300 RTK UAV equipped with a DJI Zenmuse L1 LiDAR sensor. The datasets were processed using DJI Terra and RStudio, including point cloud normalization, ground classification, generation of Canopy Height Models (CHMs), and individual tree detection through segmentation. Both area-based and individual tree-based LiDAR metrics were extracted from the processed point clouds. Stand parameter estimation models were developed using Partial Least Squares Regression (PLS), k-Nearest Neighbor (k-NN), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) algorithms. Model performances were evaluated based on the determination coefficient (R²), root mean square error (RMSE) and mean absolute error (MAE) metrics. The results indicated that the integration of area-based and individual tree-based metrics significantly improved the estimation accuracies compared to using only area-based metrics. For DBH estimation, the XGBoost algorithm achieved the best performance, yielding R² = 0.97, RMSE = 1.93 cm, and MAE = 1.48 cm. Tree height was best predicted by the RF algorithm with R² = 0.94, while tree count estimation achieved R² = 0.90 with XGBoost. Additionally, improvements of 10% to 20% in error reduction were observed for basal area and volume estimations when integrating individual tree metrics. The operational advantages of the UAV-LiDAR system were also analyzed. Compared to ground-based measurement methods, time savings of 58% and cost savings of approximately 21% were achieved. Additionally, successful data was obtained without data loss in areas that are difficult to access or on steep slopes. If the UAV system is purchased, although there is a small additional cost in the first measurement, savings of up to 73.6% can be achieved in subsequent measurements. In conclusion, UAV-based LiDAR systems provide significant advantages in terms of accuracy, efficiency, and cost-effectiveness over traditional methods for forest inventory and monitoring. The findings of this study emphasize the considerable potential of UAV-supported remote sensing technologies to enhance future forest management practices.