Leveraging Satellite-Derived Reservoir Data for Enhanced Hydrological Model Calibration: Towards Advanced Flood Prediction in Dam-Regulated Basins

Reservoirs play a critical role in shaping hydrological regimes within watershed systems, presenting challenges for accurate flood modeling. While many studies have developed reservoir operation schemes to enhance downstream discharge predictions, the impact of reservoir representation on the calibration of distributed hydrological models remains unclear. Moreover, the limited availability of upstream reservoir regulation data complicates the inclusion of dams in flood modeling. This study introduced a synergistic framework designed to improve flood predictions in data-scarce, dam-regulated basins, with the Nandu River Basin in Hainan, China, as a case study. By integrating 30m FABDEM and multi-source satellite altimetry, we reconstructed daily storage variations of the Songtao Reservoir, optimizing reservoir scheme parameters, for incorporation into the DRIVE hydrological model (DRIVE-Dam). Two calibration strategies—reservoir-inclusive and reservoir-excluded—were tested using streamflow data from the basin outlet. Satellite-derived data effectively captured high-frequency reservoir water level and storage dynamics (CC=0.95), enabling long-term simulations without management information. Evaluations incorporating hydro-stations within the watershed demonstrated that the reservoir-enabled calibration produced more accurate event hydrographs, reduced peak flow errors, and yielded realistic spatial patterns of N-year flood thresholds. This strategy also lowered flood false alarm rates (FAR) from 0.42 to 0.20 and improved the critical success index (CSI) from 0.50 to 0.54. In contrast, the reservoir-excluded calibration exhibited an overly active baseflow and subdued runoff during rainfall events, and slow flood recession, leading to overestimation of minor floods. These discrepancies arose from the mismatch between the model's naturalized assumptions and its attempt to fit observed human-influenced flood pulses, resulting in a delayed response throughout the entire basin drainage network. Our findings underscore the shortcomings of traditional calibration paradigms for watershed flood estimation and highlight the strategic value of Earth observation in enhancing hydrological and flood modeling.