Abstract: Mangrove wetlands are one of the most productive ecosystems and important blue carbon sinks. Accurate estimation of aboveground biomass (AGB) of mangroves holds great significance. By integrating field-measured AGB data with Sentinel-1/2 satellite backscatter coefficient, reflectance, vegetation index, and texture feature data, we employed a stepwise multiple regression and partial least squares regression (PLSR) approach to compare the modeling results of different variable combinations and estimated the AGB of mangroves in the Zhenzhu Bay area, Guangxi, SW China, based on the optimal model. Results indicate that: ① The multi-type feature combination model selected by the continuous PLSR algorithm showed the best performance (the coefficient of determination R²=0.88; the root mean square error RMSE=16.07 t/hm²). Among them, the Cor_{p2_3} (the texture variable correlation of the second principal component PC2 of Sentinel-2A in a 3×3 window) contributed the greatest to the modeling; ② The total and average AGB of mangroves in the Zhenzhu Bay area were approximately 45 956.41 t and 48.06 t/hm², respectively. In the spatial distribution of the predicted AGB value, it shows an overall higher level in the central and eastern parts of the area and a slightly lower level in the western part. The high-value areas covered mainly those near human activities such as shrimp farming ponds and oyster rafts, while low-value areas mainly in semi-tidal wetlands or areas with more seedlings; ③ The combination of synthetic aperture radar (SAR) and optical data could effectively enhance the accuracy of AGB inversion. Using just Sentinel-1 backscatter coefficient and derived factors is not feasible to invert biomass, and modeling with texture variables could produce better results than those with reflectance and vegetation index.