Linking canopy structure to brown spot needle blight severity in mixed pine-hardwood forests using UAV LiDAR and machine learning

Brown spot needle blight (BSNB), caused by Lecanosticta acicola , is an emerging threat to loblolly pine ( Pinus taeda L.) forests in the southeastern United States. The disease causes needle chlorosis, necrosis, premature defoliation, and canopy thinning, with consequences for productivity and timber yield. Most unmanned aerial vehicle (UAV)-based studies of BSNB have focused on spectral imagery or Structure-from-Motion metrics, and the role of integrated three-dimensional canopy structure in multiclass severity assessment remains unclear. This study evaluates a UAV-based RGB and light detection and ranging (LiDAR) feature-fusion framework using a DJI Matrice 300 Real-Time Kinematic (RTK) platform equipped with a Zenmuse L1 sensor for multiclass BSNB severity classification. Field observations and UAV-based RGB-LiDAR data were collected across 14 forest health monitoring plots covering 39.2 ha in Alabama, USA. Loblolly pine trees were assigned to four severity classes: Healthy, Early, Moderate, and Severe. RGB bands and vegetation indices were combined with LiDAR-derived canopy height metrics, canopy cover, and vertically stratified canopy density metrics. Random forest (RF), extreme gradient boosting (XGBoost), support vector machine (SVM), and artificial neural network (ANN) classifiers were trained and tested separately within each plot. XGBoost provided the best performance in 13 of 14 plots, with within-plot overall accuracies between 90.6% and 99.7%. LiDAR-derived canopy height, canopy cover, and strata-based canopy density metrics were the most informative predictors. These results demonstrate the potential of RGB-LiDAR fusion for within-plot BSNB severity classification, and validation at independent sites will support its broader application. Canopy height, cover, and vertical canopy density are useful structural indicators of BSNB severity, and RGB-LiDAR integration can support spatially explicit severity mapping and targeted field surveillance in loblolly pine forests.