Early feature study of Yunnan pine pinewood nematode disease based on hyperspectral remote sensing of ground objects

Pine Wilt Disease (PWD), caused by the pine wood nematode, is a major global forest pathology characterized by the rapid death of pine trees within a span of three months. Forestry management policy requires the eradication of all infected trees during the initial outbreak phase of a disease to contain its spread. This measure substantially relies on the timely identification of diseased trees. Accurate early diagnosis is a critical core component for effective disease control, preventing the spread of the epidemic, and maintaining the integrity of forest ecosystems. Therefore, this study proposes a new approach for early detection of PWD using hyperspectral data combined with measured physiological parameters to obtain diagnostic spectra and optimal biochemical parameters for early detection. This study investigated early-stage PWD by integrating 350-2500 nm hyperspectral data acquired with an ASD FieldSpec 4 and biochemical analysis. Results revealed significant declines in total sugar, reducing sugar, and moisture content during early infection. Study identified the spectral ranges 455-677 nm and 1974-2340 nm as optimal diagnostic windows. Using Competitive Adaptive Reweighted Sampling (CARS) and Successive Projections Algorithm (SPA), we identified diagnostic spectral bands: CARS selected 20 moisture-sensitive bands in red-edge (680-760 nm) and SWIR regions, while SPA pinpointed 4 critical bands (758, 1074, 1124, 1663 nm) across red-edge, NIR, and SWIR. This leaf-scale methodology establishes a technical foundation for regional-scale airborne and satellite hyperspectral monitoring of PWD. The XGBoost classifier achieved 91% accuracy (CARS) and 83% accuracy (SPA) in distinguishing healthy from early-stage infected trees, with AUC > 0.8 for both feature sets, demonstrating reliable spectral discrimination of infection status. This study proposes a novel method for the early detection of PWD based on spectral characteristics, offering valuable insights for the application of hyperspectral remote sensing at a regional scale.