Abstract: The continuous degradation of coastal wetland ecosystems and the dramatic shrinkage of Suaeda salsa (S. salsa) have created an urgent need for high-precision ecological restoration monitoring techniques. In this study, a multi-source remote sensing inversion framework integrating the PROSAIL-D radiative transfer model and the particle swarm optimized random forest (PSO-RF) model was proposed with betacyanin, which has an indicative role in the process of S. salsa’s adversity acclimatization, as the core research object. The PROSAIL-D model was optimized by designing field experiments in different years and replacing the specific absorption coefficient (SAC) of anthocyanin in the original model with the measured SAC of betacyanin. Based on the optimized PROSAIL-D model, this study utilized three methods, such as the three-band model, the competitive adaptive reweighted sampling (CARS), and simulating the reflectance curves under different pigment content gradients, in order to retrieve 16 spectral feature variables sensitive to betacyanin content. Then, these 16 variables were validated through field measurement spectra, finally, 10 variables were chosen to estimate betacyanin content prediction models, followed by a systematic comparison of performance differences among PSO-RF, convolutional neural network (CNN), and partial least squares regression (PLSR) under three measured data input ratios (2%, 5%, and 8%). The results indicated that ① the optimized PROSAIL-D model showed significantly better accuracy in matching the measured data in the simulation of reflectance curves of green and red phenotypes of S. salsa (RMSE_red=0.005 23 μg/cm², RMSE_green=0.007 52 μg/cm²). Moreover, the gap between the retrieved betacyanin content and the measured values has significantly narrowed, with the R² of the green and red phenotypes of S. salsa increasing by 32% and 51%, respectively, compared to before optimization. ② Ten constructed spectral feature variables showed the best response to betacyanin and were verified by multiple indicators of measured data (R²＞0.7). Among them, the Chlorophyll-Insensitive Betacyanin Index (CIBI) and the Normalized Difference Spectral Index corresponding to wavelengths of 612 nm and 662 nm had the highest importance, both reaching 0.7 or above.③ PSO-RFR model has the best inversion accuracy (R²=0.96, RMSE=1.27 μg/cm², RPD=4.86) with 8% of measured data, which was higher than that of CNN (R²=0.93) and PLSR (R²=0.86). This further indicates that a high proportion of measured data can effectively correct the spectral response deviation between the physical model and the real scene; ④ The PSO-RF model has significant advantages in the inversion of betacyanin at different concentrations. It performs particularly well in the low-concentration range (0-10 μg/cm²), with R²=0.92, RMSE=2.14 μg/cm², and RPD=3.12. In the medium-to-high-concentration range, it still maintains an effective prediction level (R²＞0.75). The optimal model constructed at the canopy scale was applied to the UAV hyperspectral images. The MAE, RMSE and MAPE between the inversion results and the measured values were 1.95, 2.20 μg/cm² and 17% respectively. The study shows that it is feasible to retrieve the canopy betacyanin content using UAV imagery with the PROSAIL-D and PSO-RF models and feature variables. This study provides a high - precision remote sensing inversion method for the physiological monitoring and ecological restoration assessment of Suaeda salsa in coastal wetlands.