Résumé
•Difficult modelling of water availability in semi-arid poorly gauged small lakes.•Long-term Landsat surface water observations provide valid data on reservoir state.•Ensemble Kalman Filter combines 15 years of Landsat data in 7 hydrological models.•Errors from hydrological modelling & Landsat’s limited temporal resolution are reduced.•Surface water volumes RMSE reduce by 54% according to long-term field monitoring.
Small reservoirs represent a critical water supply to millions of farmers across semi-arid regions, but their hydrological modelling suffers from data scarcity and highly variable and localised rainfall intensities. Increased availability of satellite imagery provide substantial opportunities but the monitoring of surface water resources is constrained by the small size and rapid flood declines in small reservoirs. To overcome remote sensing and hydrological modelling difficulties, the benefits of combining field data, numerical modelling and satellite observations to monitor small reservoirs were investigated. Building on substantial field data, coupled daily rainfall-runoff and water balance models were developed for 7 small reservoirs (1–10 ha) in semi arid Tunisia over 1999–2014. Surface water observations from MNDWI classifications on 546 Landsat TM, ETM+ and OLI sensors were used to update model outputs through an Ensemble (n = 100) Kalman Filter over the 15 year period. The Ensemble Kalman Filter, providing near-real time corrections, reduced runoff errors by modulating incorrectly modelled rainfall events, while compensating for Landsat’s limited temporal resolution and correcting classification outliers. Validated against long term hydrometric field data, daily volume root mean square errors (RMSE) decreased by 54% to 31200 m3 across 7 lakes compared to the initial model forecast. The method reproduced the amplitude and timing of major floods and their decline phases, providing a valuable approach to improve hydrological monitoring (NSE increase from 0.64 up to 0.94) of flood dynamics in small water bodies. In the smallest and data-scarce lakes, higher temporal and spatial resolution time series are essential to improve monitoring accuracy.