Using multivariate analysis to detect the hyperspectral response of Chinese fir to acid stress.

A useful method was developed to establish a diagnostic model using hyperspectral remote sensing to predict and monitor acid stress on plants. We analysed the hyperspectral response of Chinese fir to acid rain by measuring the spectral reflectance of the seedling leaves, sprayed by simulated acid rain (pH, 2.5, 4.0, and 5.6), for three periods. The sensitive bands were located and the rules for predicting classes of simulated acid stress on Chinese fir were established using a classification and regression tree (CART) approach. The acid-sensitive bands of Chinese fir were nearly all located between 380 and 410 nm, 460 and 560 nm, and 640 and 750 nm. CART predictor variables, which were selected from sensitive bands, reduce data dimensionality significantly. The misclassification errors of the CART training process in correctly attributing variables to respective target classes are 7.78%, 6.67%, and 11.67% respectively, at each measurement period, and the cross-validation misclassification errors are 16.6%, 11.1%, and 23.3%, respectively. Our results show that the spectral reference bands, which are related to chlorophyll-aandbaround 670 and 450 nm, as well as the slight peak in the green around 550 nm, significantly affected the classification accuracy on acid stress. These provide useful optical response to acid stress. [ABSTRACT FROM AUTHOR]