Parameter transferability of a distributed hydrological model to droughts

Abstract

Hydrological models often do not simulate properly streamflow (Q) during droughts, because of a poor representation of the interactions among precipitation deficits, actual evapotranspiration (ET), and terrestrial water storage anomalies (TWSA) during these periods. However, there is little research comprehensively evaluating model skills during droughts of varying intensity in a spatially distributed way. To shed further light into these drops in model skills and step toward more robust models in an anthropogenic era and a changing climate, we evaluated Q, ET, and TWSA simulations during moderate and severe droughts, and we tested if calibrating during a moderate drought could enhance model performances during a severe one. We applied the distributed hydrological model Continuum over the heavily human-affected Po river basin in northern Italy and the period 2010 – 2022. Moreover, we exploited independent ground- and remote sensing-based datasets to evaluate the temporal and spatial variability of Q, ET, and TWSA monthly simulations across the whole basin and 38 sub-catchments. Model performances for Q across the study sub-catchments were comparable during both wet years (2014 and 2020, mean KGE = 0.59±0.32) and moderate droughts (2012 and 2017, mean KGE = 0.55±0.25). Further, Continuum simulated well Q for the basin outlet even during a severe drought (KGE = 0.82 in 2022), while its performances generally decreased among the sub-catchments (mean KGE = 0.18±0.69 in 2022). In general, the model well represented ET and TWSA seasonality over the study area, and a decline in TWSA over the more recent years. Yet, during the severe 2022 drought we detected an increased uncertainty in ET anomalies, especially in human-affected croplands, that could explain the Q performance drop along with an increased anthropogenic disturbance. Including a moderate drought (2017) in the calibration period did not lead to a significant improvement in model skills during the severe event (mean KGE = 0.18±0.63 for Q during 2022), meaning that the severe 2022 drought was fairly unique for the study area both in terms of hydrological processes and human disturbance on them. By unveiling an increase in model uncertainty during a severe drought and possible causes for it, our findings are relevant to assess and possibly enhance model robustness in a changing climate and the anthropogenic era for adequate water management, disaster risk reduction, and climate change adaptation.