Pine shoot beetle (PSB, Tomicus spp.) outbreaks cause widespread Yunnan pine (Pinus yunnanensis Franch) mortality in southwestern China. Early identification of PSB attacks could help forest managers mitigate the infestation before it turns into an outbreak. However, the subtle spectral changes and complex process of PSB-induced crown discoloration make the remote sensing approach difficult.This study employed a manipulated insect infestation experiment to reveal suitable monitoring indicators. Healthy Yunnan pine crowns were infested with PSB adults at different preset densities (light, moderate, and severe) using the bagging method. The crown damage parameters, physiological properties, and corresponding spectral data were systematically measured in time series. Partial least squares regression (PLSR) was used to retrieve crown chlorophylla+b (Cab) and relative water content (RWC) via band reflectance. Random forest (RF) was used to determine the optimum spectral variables capable of capturing dynamic variations in crown shoot damage ratio (SDR).Results showed that (1) after four weeks’ PSB shoot feeding attack, SDR reached saturated (26%-50%) for all damage levels, indicating a crown discoloration “switch point”. (2) A continuous decline was found in Cab and RWC at both shoot and crown levels during the whole infestation process, and the crown level decrease was smaller than the shoot level due to crown heterogeneous discoloration. (3) For PLSR retrieving crown Cab and RWC, estimation accuracy of healthy crowns could reach R2 = 0.70, RMSEcv = 14.60 mg/m2 for Cab, and R2 = 0.82, RMSEcv = 1.25% for RWC, respectively. However, PSB early attack impacted these retrieval accuracies. (4) Spectral differences were evident in the visible region (530–600 nm) and near-infrared (NIR) plateau (780–1300 nm) among the preset damage levels. Significant differences in spectral variables (AMP, CI, OSAVI, RD, and SR) were observed between healthy and moderately damaged crowns. (5) In RF modeling, the importance of spectral variables for SDR estimation varied throughout the shoot feeding process. OSAVI, CI, MSI, PSRI, and SR2 were the top indicators that were suitable for identifying SDR ranging from 5% to 50%. SDR could be estimated with an accuracy of R2 = 0.71, RMSEcv = 8.92%, after 4 weeks’ infestation. Simultaneously, with the RF model, 4 preset damage levels could be differentiated with an out-of-bag error of 14.94% and a kappa coefficient of 0.8835.In conclusion, this study mainly examined the physiological and spectral signatures of PSB early attacked crowns, and provided optimum spectral variables and models which will further help unmanned aerial vehicle (UAV) and satellite remote sensing identify PSB infestation at the early stages.
Read full abstract