Abstract: On August 7, 2025, a catastrophic flash flood triggered by extreme rainfall struck Yuzhong County, Gansu Province, exhibiting a typical “flash flood–debris flow–flash flood” cascading disaster chain. To address the challenge of reconstructing cascading processes of flash flood disasters under frequent extreme weather, this study takes this event as an example to establish and verify a technical process integrating multi-source remote sensing and hydrodynamic models. By combining observed data from meteorological stations with the ERA5-Land reanalysis dataset, a detailed reconstruction of the rainfall process in the upstream debris flow initiation zone was achieved. Subsequently, the HEC-RAS 2D hydrodynamic model was adopted to reconstruct the inundation process in the downstream floodplain area based on topographic data, land use information, and remote sensing-based disaster extents. The results indicate that: (1) ERA5-Land data effectively captures the spatial distribution and cumulative trend of rainfall, but systematically underestimates short-duration extreme rainfall intensity, with peak rainfall in the study area differing by a factor of 5 to 8 from observed values; (2) the HEC-RAS model constructed with 12.5 m resolution ALOS PALSAR DEM effectively reproduces flood propagation paths and the main inundation extent, and simulation results are highly consistent with actual disaster extents extracted from post-event high-resolution satellite imagery. Local discrepancies are mainly attributed to topographic data accuracy and the influence of human engineering activities on surface parameters. In summary, the integrated “data–simulation–reconstruction” analytical framework proposed in this study demonstrates the dual role of multi-source remote sensing in both driving and validating the reconstruction process of flash flood disasters, providing a technical pathway and scientific reference for early warning and quantitative assessment of small-watershed disasters.