Estimation of the water content of needles under stress by

Introduction: Methods: In this work, we combine vegetation indices with machine learning algorithms to estimate the water content of needles at a large scale. Multiple vegetation indices are screened via recursive feature elimination cross validation (RFECV), and then support vector regression (SVR) and back-propagation neural network (BP) models are used to predict the leaf weight content fresh (LWCF) and leaf weight content dry (LWCD) of needles over a large area. The water content ranges were then classified based on the severity of damage derived from actual sampling data. These ranges were used to categorize the estimated water content, thereby assessing the degree of tree damage. The accuracy of the method is verified by comparing the estimation results with field measurements, and the results are combined with the classifications of the leaf loss rate(LLR) to assess the severity of infestation. Results: The results indicate that: 1) When estimating LWCD and LWCF using the SVR and BP models, the SVR model demonstrated superior accuracy and stability (MAE for LWCF = 0.1477, RMSE = 0.17314; MAE for LWCD = 0.10507, RMSE = 0.14760). 2) The classification accuracies of LWCD and LWCF were notably higher in areas with light and medium damage, suggesting that these indices are effective indicators for assessing damage caused by