Adaptive ROI-Based Tree Health Assessment Using UAV Multispectral and LiDAR Data

Accurate individual tree health assessment is essential for effective forest monitoring, particularly in complex forest stands where neighboring crowns introduce shadow effects and spectral mixing. This study proposes an adaptive Region of Interest (ROI) approach dynamically scaled according to individual tree crown metrics, integrating UAV-derived multispectral imagery and LiDAR point clouds acquired from a Scots pine forest in Kozağaç village, Gümüşhane, Türkiye. Individual trees were automatically detected using a Canopy Height Model (CHM), and crown width, crown area, and spatial parameters were extracted to generate dynamically scaled ROI windows. The Normalized Difference Vegetation Index (NDVI) and Normalized Difference Red Edge Index (NDRE) were then calculated within these windows to classify individual trees as unhealthy, moderately healthy, or healthy. Compared with fixed crown area windows, the proposed adaptive ROI approach improved all classification metrics for both NDVI and NDRE. Correctness increased from 79% to 81% for NDVI and from 63% to 66% for NDRE. Similarly, completeness increased from 68% to 81% and from 59% to 64%, IoU from 52% to 68% and from 42% to 48%, and F1 scores from 68% to 80% and from 59% to 65%, respectively. These findings demonstrate the potential of adaptive ROI sampling to improve individual tree health classification and support UAV-based forest health monitoring.