Estimation of Carotenoid Content of Suaeda salsa Leaf in the Coastal Wetland Based on PROSPECT-D

Carotenoid, as one of the important pigments involved in photosynthesis within vegetation, plays a crucial role in the rapid, accurate, and non-destructive prediction of its content, which is essential for dynamically monitoring the physiological status of vegetation. This study focuses on the Dongtai Tiaozini Wetland in Jiangsu, where leaf samples of Suaeda salsa were collected, and the leaf reflectance spectra and biochemical parameters (chlorophyll, carotenoids, betacyanin, equivalent water thickness, and dry matter content) were measured. A three-band carotenoid index (TBCRI) was developed to estimate the carotenoid content in Suaeda salsa leaves, based on reflectance spectra simulated using the PROSPECT-D radiative transfer model. Sensitive feature variables for carotenoid content in Suaeda salsa leaves were selected using sensitivity analysis. Estimation models for carotenoid content were constructed using Partial Least Squares Regression (PLSR), Support Vector Machine (SVM), and Particle Swarm Optimization-Random Forest (PSO-RF) algorithms, followed by accuracy evaluation. The results indicate that: (1) TBCRI, PSSRc, and PSNDc derived from leaf reflectance spectra showed a strong correlation with carotenoid content in Suaeda salsa leaves, with TBCRI exhibiting the highest correlation; (2) The inversion model of Suaeda salsa leaf carotenoid content constructed by coupling the semi-empirical model and the SVM algorithm had the best accuracy, with a coefficient of determination (R²) of 0.941, root mean square error (RMSE) of 0.415 μg·cm^-2, relative prediction deviation (RPD) of 4.301, and standard deviation (SD) of 0.192 μg·cm^-2; (3) TBCRI contributed the most to the carotenoid inversion model (27%), significantly outperforming PSNDc (18%), CRI550 (17.4%), PSSRc (16.8%), PRI (13.9%), and SRcar (6.9%). By simulating the reflectance spectra of Suaeda salsa leaves using the PROSPECT-D radiative transfer model and combining it with machine learning algorithms, this study achieved high-precision inversion of carotenoid content in Suaeda salsa leaves in coastal wetland, providing technical support for assessing the effectiveness of ecological restoration in degraded coastal wetland areas.