Assessing the impact of Earth Observation data-driven calibration of the melting coefficient on the LISFLOOD snow module

Abstract

Abstract. LISFLOOD is a continental, operational hydrological model widely used in Europe. Among various hydrological processes, it simulates snowmelt using a degree-day approach, where the snowmelt coefficient is typically calibrated against discharge data. This study evaluates LISFLOOD’s current snow module and investigates the effects of a post-replacement of the snowmelt coefficient across nine European basins with varying snow influence. The parameter is calibrated using Earth Observation (EO) snow cover fraction (SCF). The outcomes illustrate the extent to which a parsimonious calibration of a single model component, intended to improve a specific module, affects hydrological model performance. To this purpose, we integrate Sentinel-2 and MODIS data to address issues related to gaps and misclassifications in snow detection over complex terrain and obtain an improved reference snow cover. Using EO SCF, we estimate a spatially distributed snowmelt coefficient, in contrast to the uniform values currently used in LISFLOOD. The coefficients are optimized by matching modelled and observed SCF, and their hydrological impacts are assessed while keeping all other model parameters unchanged. This enables us to test whether modifying only the snowmelt coefficient affects discharge performance, and whether the standard calibration adequately represents both snow dynamics and streamflow. Compared with EO SCF, the standard calibration showed biases from −1 % to 22 % and RMSE values from 20 % to 55 %. The EO-based proposed approach improved both bias and RMSE by up to 8 %. In general, the optimized coefficients did not significantly change the simulated discharge at the basin level in terms of KGE, but their application led to noticeable divergences in discharge within smaller upstream catchments. Moreover, the improved representation of snow cover led in some cases to shifts in the timing and magnitude of snowmelt and total runoff. These findings highlight the potential of integrating EO data to calibrate the snowmelt coefficient without changing other calibration parameters. This approach may offer practical advantages in situations that require accurate snow cover representation without the need for a complete re-calibration, which may be overly onerous. Nevertheless, our results indicate that standard calibration procedures already provide an acceptable representation of snow dynamics.