UAV-multispectral imaging and machine learning for Brown Spot Needle Blight severity assessment in southeastern U.S. pine forests

Brown Spot Needle Blight (BSNB) poses a significant threat to loblolly pine (Pinus taeda) forests in the southeastern United States, reducing timber yields, biodiversity, and overall forest health. Traditional detection methods rely on field-based assessments, which are time-intensive and impractical for large areas. Unmanned aerial vehicle (UAV)-based multispectral imaging presents a potential alternative, but its effectiveness for BSNB detection remains largely unexplored. To address this gap, we developed a UAV-based remote sensing framework to detect and map BSNB severity using multispectral imagery and machine learning. Specifically, we aimed to (i) classify and map BSNB severity using Support Vector Machines (SVM) and Artificial Neural Networks (ANN) and (ii) quantify the density of healthy and BSNB-infected trees using point cloud-derived metrics from UAV-based Structure from Motion (SfM). Field-based assessments across fourteen loblolly pine-dominated sites in the state of Alabama, provided BSNB-verified observations for model training and testing, and high-resolution UAV-based multispectral imagery were acquired using a DJI Mavic 3M. Spectral analysis of processed image bands and derived indices identified the Normalized Difference Vegetation Index (NDVI) and Soil-Adjusted Vegetation Index (SAVI) as optimal predictors of BSNB presence. Classification models achieved high accuracy, with SVM and ANN reaching 94.79% and 94.00% accuracy in Washington County, and 94.94% and 94.89% in Cullman County, respectively. Kappa coefficients ranged from 0.80 to 0.92 for SVM and 0.80 to 0.89 for ANN. This study provides one of the first systematic UAV-based approaches for BSNB detection and severity mapping, demonstrating the potential of combining multispectral imagery, SfM-derived metrics, and machine learning for operational forest health assessment.