Existing near-real-time (NRT) SAR-based flood mapping systems are typically designed for individual SAR sensors and often face challenges in achieving reliable flood extent delineation across diverse environments. For instance, the self-supervised Radar-based Inundation Daily (RAPID) system (Shen et al., 2019a) struggled with sensitivity to snow and ice. In this study, we enhanced RAPID by integrating an Adaptive Initialization (AdaI) module for refining initial water samples. This enhanced system is designed to work with multiple SAR imagery in an NRT manner, adapting to various configurations, including variable microwave wavelength. This flexibility significantly improves the system's ability to detect submerged floods in complex conditions,especially in complex regions such as mixed snow cover and arid dryland areas, so can enhance the temporal coverage and accuracy. We demonstrate the system's efficacy through a comparative analysis against baseline methods, including a simple bimodality-based segmentation and the original RAPID, showing balanced performance across diverse environments and consistent detection capabilities. Moreover, the system can seamlessly work across various SAR sensors. The input SAR imagery ranges from Sentinel-1, Radar Constellation Mission (RCM), Radarsat-2 (RS2), Capella, and Advanced Land Observing Satellite-2 (ALOS-2), which include C-band, X-band, and L-band, respectively. Results align well with very high-resolution PlanetScope images and reference floodwater masks, highlighting that the robustness of SAR flood mapping critically depends on the initial sample generation, as misleading initializations can significantly compromise detection accuracy to the extent that post-processing cannot adequately correct. Furthermore, we outline future developments for NRT flood mapping systems using multi-SAR sensors, emphasizing the integration of advanced change-detection algorithms and the expansion of satellite tasking to improve detection performance and response times.
