Erannis jacobsoni disturbance detection based on unmanned aerial vehicle red edge spectral features

Boreal coniferous forests play important roles in global ecological and economic processes. Mongolia, rich in forest resources and part of the boreal ecosystem, faces significant deforestation due to Erannis jacobsoni Djak (Lepidoptera: Geometridae), a rapidly spreading needle pest in coniferous forests. This study aims to provide with rapid and precise pest occurrence data, enabling timely and effective control measures to preserve and enhance the agroforestry ecological environment. In vegetation disturbance detection, UAV remote sensing exhibits operational performance with unique spatiotemporal advantages (notably cm-resolution data acquisition and flexible revisit cycles) unattainable through traditional ground surveys or satellite platforms. Therefore, we used unmanned aerial vehicle (UAV) imagery from representative areas affected by E. jacobsoni, calculated conventional and red edge spectral indices, extracted features sensitive to pest infestation levels, detected disturbances using machine-learning algorithms, and analyzed the pest's spatial distribution. The sequential forward selection (SFS) and successive projection algorithms (SPA) can effectively extract features sensitive to the response to pest disturbance, in which the red edge (RE) features have a greater effect than the conventional (CONV) features in detecting the pest. The detection models developed using machine learning all achieved accuracy rates above 82%, with the Back Propagation Neural Network (BPNN) performing the best. Spatial distribution analysis based on the model revealed that E. jacobsoni primarily exhibited a pattern of outward diffusion from the center of aggregation during the outbreak period.