How well do hydrological models simulate streamflow extremes and drought-to-flood transitions?

Abstract

Flood impacts can be enhanced when they occur shortly after droughts. Hydrological models are useful tools to better understand the underlying processes and mechanisms driving the response of floods occurring in close succession to streamflow drought. However, it is yet unclear how well hydrological models capture these compound extreme events and which modeling decisions are most important for model performance. To address this research gap, we calibrated four conceptual bucket-style hydrological models with different structures (GR4J, GR5J, GR6J, and TUW) for 63 catchments located in Chile and Switzerland using different calibration strategies. Specifically, we assessed the relative importance of different methodological choices in simulating and detecting observed drought-to-flood transitions, including model structure, streamflow transformation, and the Kling–Gupta efficiency (KGE) formulation and weights used to calibrate the model parameters. We demonstrate that model performance, as expressed by the KGE, does not guarantee a good performance in terms of detecting streamflow extremes and their transitions. Further, we show that a model's performance with respect to capturing extreme events primarily depends on how well it captures streamflow timing. Our results also highlight that model structure, catchment characteristics and meteorological forcings play a key role in the detection of transitions. Overall, we find that model representation of drought-to-flood transitions is generally poor, especially in semi-arid and high-mountain catchments compared to humid low-elevation catchments. Ultimately, our study provides insights for further model improvements to simulate and better understand drought-to-flood transitions and to identify regions prone to this hazard.