Filling the gaps: Calibrating a rainfall-runoff model using satellite-derived surface water extent

Calibration is a crucial step in the application of hydrological models and is typically performed using in situ streamflow data. However, many rivers on the globe are ungauged or poorly gauged, or the gauged data are not readily available. In this study, we used remotely-sensed surface water extent from the Global Flood Detection System (GFDS) as a proxy for streamflow, and tested its value for calibration of the distributed rainfall-runoff routing model LISFLOOD. In a first step, we identified 30 streamflow gauging sites with a high likelihood of reliable GFDS data. Next, for each of these 30 sites, the model parameters related to groundwater and routing were independently calibrated against in situ and GFDS-derived streamflow time series, and against the raw GFDS surface water extent time series. We compared the performance of the three calibrated and the uncalibrated model simulations in terms of reproducing the in situ streamflow time series. Furthermore, we calculated the gain achieved by each scenario that used satellite-derived information relative to the reference uncalibrated scenario and the one that used in situ data. Results show that using the raw GFDS data as a proxy for streamflow for calibration improved the skill of the simulated streamflow (in particular the high flows) for 21 of the 30 sites using correlation as a metric. Furthermore, we discuss a calibration strategy using a combination of in situ and satellite data for global hydrological models. •Limited availability of streamflow data confounds hydrological model calibration.•We tested the value of satellite-based surface water extent data for calibration.•Skill improvements (r) were obtained for 21 of the 30 tested stations.•GFDS data provide promising opportunities for calibration in fully ungauged regions.•However, the dimensionless character of the GFDS data poses some limitations.