Unraveling spatiotemporal dynamics of pine wilt disease via time‐series UAV monitoring and deep learning

Abstract BACKGROUND This study investigates the spatiotemporal dynamics of pine wilt disease (PWD) to inform data‐driven management. We implemented a time‐series monitoring framework in a township in Zhejiang Province, China, acquiring seven sequences of unmanned aerial vehicle (UAV) orthomosaics between 2022 and 2024. An enhanced YOLOX‐based change detection model was developed and trained on 300 000 samples. This model exploits phenological variations to automatically identify PWD‐discolored pines, effectively filtering confounding objects. RESULTS The model demonstrated robust performance, achieving an Average Precision (AP) of 0.89, with Precision and Recall exceeding 85%. Analysis revealed a consistent westward expansion and a progressive increase in disease hotspots. Crucially, winter surveys detected substantial delayed‐symptom pines missed in autumn, roughly equivalent to the autumn baseline. Consequently, the annual cumulative mortality caused by PWD was nearly double (2×) the autumn count. Over 90% of trees newly identified in autumn were located within 300 m of infections detected the previous spring, indicating strong spatial clustering. Furthermore, 80% of infected trees occurred at elevations < 400 m and slopes < 29°, aligning with prevailing easterly winds. CONCLUSION This research establishes a validated framework bridging remote sensing and on‐the‐ground sanitation. By quantifying the symptom lag effect (which doubles mortality estimates relative to traditional autumn surveys) and elucidating environmentally driven spread mechanisms, we provide a scientific basis for correcting census biases and optimizing resource allocation for precise PWD management. © 2026 Society of Chemical Industry.