Impacts of pine species, infection response, and data type on the detection of Bursaphelenchus xylophilus using close-range hyperspectral remote sensing

Abstract

The early detection of forest pests and diseases is a primary focus of remote sensing applications for forest health monitoring. Pine Wilt Disease (PWD), which causes significant damage to pine resources in many countries and regions, has been a key area where the close-range hyperspectral remote sensing has demonstrated its advantages for early diagnosis. However, it remains unclear whether PWD can be detected during the pre-visual stage and, if so, how to achieve hyperspectral detection. This study aimed to investigate the impacts of pine species, infection responses, and data types on hyperspectral detection of PWD, particularly in the pre-visual stage. Artificial inoculation experiments were conducted across three locations with 76 sample trees of two pine species, and hyperspectral data were collected regularly using ground-based non-imaging and UAV imaging spectrometers Five infection responses were identified: keep healthy (KH), quick infection (QI), slow recovery (SR), quick recovery (QR), and slow infection (SI). Spectral analysis revealed dynamic changes in the indices RVI (680–550,750) and NDVI (560,680), corresponding well with the spectral characteristics of the five infection responses. The infected trees with QI response could be spectrally detected starting from day 14, with over 50 % accuracy. Importance analysis using RF identified RVI (554,677) and NDVI (531,570) as consistent in detecting pre-visual stages. In contrast, the six VIs determined by PCA-S (RARSb, RVI (900, 680), RVI (800, 680), RVI (760, 500), RVI (800, 635), and REP) exhibited high consistency and played a crucial role in identifying pre-visual stage infected trees. These VIs combined with specific color bands, enabled the creation of false-color images highlighting infected trees starting from day 14 post-inoculation. The study highlighted the importance of recognizing infection response patterns for accurate PWD detection, with only QI response trees showed a stable infection cycle, making day 14 post-infection a meaningful starting point for spectral detection. Additionally, imaging and non-imaging data types did not significantly affect the detection process, and the impact of spectral resolution variations between 1 nm and 3.5 nm was negligible. Further research is required to determine the threshold for larger differences of spectral resolution and to explore detection across various pine species and growth environments.