Abstract: Chlorophyll a and b content (C_ab) of Suaeda salsa in coastal wetlands is an important indicator of vegetation physiological status and ecological function. However, the differences in chlorophyll retrieval performance across scales remain unclear. In this study, radiative transfer models combined with machine learning methods were employed to systematically compare the retrieval capability of leaf C_ab and canopy C_ab across multiple scales (leaf, canopy, and pixel), as well as their application potential using UAV (Unmanned Aerial Vehicle)-based hyperspectral images. First, multiscale simulated spectra were generated using the PROSPECT-D and PROSAIL-D models. By integrating simulated and measured reflectance spectra, sixteen classical vegetation indices, three-band model, and fractional order derivative-based vegetation indices were calculated to construct training datasets at the leaf, canopy, and pixel scales. Second, spectral feature variables sensitive to Suaeda salsa C_ab at leaf or canopy level were selected using correlation analysis and the minimum redundancy maximum relevance (mRMR) method. Finally, partial least squares regression (PLSR), support vector machine (SVM), and multilayer perceptron (MLP) models were developed for multiscale chlorophyll retrieval, and their performance as well as applicability to UAV hyperspectral imagery were evaluated. The results showed that: ①CIred_edge, TCARI, MTCI, MCARI/OSAVI705,750, and the three-band model exhibited high sensitivity to Suaeda salsa chlorophyll content at the leaf, canopy, and pixel scales; ② among the chlorophyll retrieval models constructed using simulated spectra and 70% of measured spectra, MLP consistently achieved the best performance across all three scales, with R², RMSE, and RPD values of 0.906, 3.772 μg·cm⁻², and 3.318 at the leaf scale; 0.848, 6.021 μg·cm⁻², and 2.612 at the canopy scale; and 0.774, 9.086 μg·cm⁻², and 2.132 at the pixel scale, respectively; ③ when the optimal pixel-scale MLP model was applied to UAV hyperspectral imagery, the RMSE and MAE between predicted and measured values were 1.243 μg·cm⁻² and 0.862 μg·cm⁻², respectively. This study systematically reveals the scale-dependent differences and underlying mechanisms in chlorophyll retrieval of Suaeda salsa, and provides methodological support for high-precision retrieval of vegetation physiological parameters and UAV-based remote sensing applications in coastal wetlands.