Impervious urban surfaces critically impact ecological environments, necessitating precise and efficient mapping for sustainable urban planning. While hyperspectral remote sensing is widely used for feature extraction, single-source data often face challenges like homospectral heterogeneity and heterospectral homogeneity in complex urban areas. This study addresses these limitations by integrating Zhuhai-1 hyperspectral imagery with Sentinel-1 radar data, proposing an innovative method to enhance impervious surface mapping accuracy through multi-source remote sensing synergy. Further, we compared four tree-based ensemble-learning algorithms for use with multi-source remote sensing data. The results of the pilot study using this approach can be summarized as follows: (1) The four tree-based ensemble learning methods using multi-source remote sensing features perform better than single-source spectral data extraction. Specifically, the Kappa coefficient for the lightweight gradient boosting tree algorithm (LightGBM) in the impervious surface mapping of built-up areas and urban fringes increased by 0.014 and 0.017, respectively. (2) The LightGBM algorithm using multisource remote sensing features exhibited the best mapping accuracy for extracting impervious surfaces compared to other algorithms, with an accuracy of 93.2% in built-up areas and 92.1% at urban edges. Further, it is also shown to be the most efficient model, with 19.7- and 20.3-second running times in built-up areas and urban edges, respectively. The findings in this study provide a new approach for high-efficiency and high-precision impervious urban surface mapping.