Hydrological model skills change with drought severity; insights from multi-variable evaluation

Abstract

Listen

Translate article

Hydrological models often do not properly simulate streamflow (Q) during extreme events, including droughts. Limited abilities in simulating Q during droughts may arise from a misrepresentation of Q generating processes during these periods, but little research has focused on distributed, process-based models over human-affected areas and extreme events. To shed more light into model consistency during these periods, we evaluated the ability of the hydrological model Continuum in simulating Q over the human-affected Po river basin in Italy during droughts of different severity over the last 13 years, including the severe 2022 event. To investigate the causes for potential model deterioration during severe droughts, we assessed the simulation of evapotranspiration (ET) and Terrestrial Water Storage (TWS) against independent remote sensing-based benchmarks, and possible inconsistencies in forcing and benchmark data. Finally, we included a moderate drought in the calibration period, as potential strategy to improve model performances during severe droughts. The model represented well Q (KGE = 0.81 for the outlet of the basin), ET (r = 0.94) and TWS (r = 0.76) over the whole study period. Focusing on Q and specific sub-periods, model performances were comparable during wet years (2014 and 2020) and moderate droughts (2012 and 2017), with KGE across the 38 study sub-catchments of 0.59±0.32 (mean ±standard deviation) during wet years and 0.55±0.25 during moderate droughts. The model simulated Q well for the outlet section of the basin also during the severe 2022 drought (KGE = 0.82). However, performances across the sub-catchments declined in 2022 (KGE = 0.18±0.69). For the severe drought, we detected a decrease in model performances for ET, in particular over human-affected croplands (mean decrease in r by 105% and mean increase in nRMSE by 86%). Furthermore, calibrating during a moderate drought did not improve model performances in 2022 (KGE = 0.18±0.63), pointing to the fairly unique conditions of this period in terms of hydrological processes and human interference on them. Our study highlighted decreased model skills specifically during a severe drought and identified the neglection of irrigation as the most plausible cause for this. Given projected increases in severe droughts and the frequent modelling simplification of human activities, despite their heavy interference in many regions, our findings are highly relevant to move towards more robust hydrological modelling in a changing climate and the Anthropogenic era, to support management and adaptation strategies.