Abstract
INTRODUCTION: Leaf water content is a key physiological indicator of plant growth and health status. Constructing leaf water content estimation models based on spectroscopy is an effective method for monitoring plant physiological conditions. METHODS: To improve the accuracy of leaf water content estimation and develop models applicable to different plants, this study collected 1,680 groups of hyperspectral and water content data from peach tree leaves. Estimation models were established using two methods: "constructing vegetation indices" and "selecting characteristic wavelengths." The accuracy and number of wavelengths used in each model were systematically evaluated. The optimal model was used to predict the water content of each pixel in the hyperspectral images, achieving visualization of leaf water distribution. Additionally, 244 groups of hyperspectral and water content data from apple tree and lettuce leaves were collected to validate the generalization ability of the optimal model. RESULTS: Results showed that the optimal models established using the two methods were the linear regression model based on the vegetation index NISDI (3 wavelengths, R(P) (2) = 0.9636, RMSEP=0.0356), and the CARS-RF model (12 wavelengths, R(P) (2) = 0.9861, RMSEP=0.0219). Although the accuracy of the two models was similar, the latter used four times more wavelengths than the former, so the former was chosen as the optimal model. Using the optimal model to estimate the water content of apple tree leaves, the R(P) (2) and RMSEP were 0.9504 and 0.1226, respectively. For lettuce containing only leaf tissue, the R(P) (2) and RMSEP were 0.8211 and 0.1771, respectively. DISCUSSION: These results indicate that the model has some generalization ability and can accurately estimate the water content of leaves of woody plants in the same family, with some performance degradation across different growth forms. The study results achieved accurate estimation of leaf water content for three types of plants and also provided a reference for establishing plant leaf water content estimation models with generalization ability.