Detecting White Pine Needle Damage through Satellite Remote Sensing

Abstract

Abstract Eastern white pines (Pinus strobus L.) of New England forests have been recently impacted by a fungal disease known as White Pine Needle Damage (WPND), causing widespread needle damage. To complement current WPND monitoring methods based on field and aerial detection surveys, we evaluated the potential of satellite remote sensing technology to detect WPND outbreaks. Using Sentinel-2 spectral vegetation indices (SVIs), we directly visualized change overlapping WPND outbreaks and ran Random Forest machine learning classifiers for feature selection and WPND detection and severity classification. Direct visualization of WPND associated change was most effective through the Normalized Difference Infrared Index (NDII), which captured decreases in vegetation health conditions coinciding with peak WPND symptoms. We obtained good accuracies in binary (WPND vs. Non-WPND) detection (70%) and two-class severity modeling of WPND (75%). The highest accuracies were achieved using imagery from early to late summer. The most selected SVIs for modeling were the Carotenoid Reflectance Index1 (CRI1), the Sentinel-2 Red-Edge Position (S2REP), and the Normalized Difference Vegetation Index (NDVI). Our results suggest detecting severe WPND through fine resolution remote sensing is feasible. However, more work is needed to determine the effects of spatial, spectral, and temporal resolution of remote sensing data for detecting WPND severity levels.