Benefit of Multivariate Model Calibration for Different Climatic Regions

Abstract

AbstractHydrological models are traditionally calibrated against observed discharge. However, for a given model, similar performance for discharge simulation can be achieved through a variety of parameter combinations, some of which produce unrealistic simulations of non‐discharge variables. Thus, considering non‐discharge variables in calibration can help to reduce equifinality and give more realistic simulations. Few studies considered non‐discharge variables in calibration and evaluation and across a large sample of catchments. In this study, we calibrate the lumped SIMHYD model with several combinations of different hydrological variables such as discharge, actual evapotranspiration, soil wetness, and total water storage. The various calibration scenarios are conducted across 550 catchments in Australia spanning arid, tropical and temperate regions. We show that the performance of the discharge‐calibrated model for actual evapotranspiration, soil wetness, and total water storage is strongly related to the co‐seasonality of discharge and the respective variable. Considering actual evapotranspiration, soil wetness, or total water storage in addition to discharge in calibration improves the simulation of the added variable on average by 53%, 58%, and 41%, respectively, which comes at the expense of decrease in discharge performance by 5%. Cross‐benefits, that is, the improved simulation of a variable by considering another variable in calibration, mainly occurs between actual evapotranspiration and total water storage and are most pronounced in humid catchments with highly seasonal soil wetness. We also find that when using all four variables in model calibration, the trade‐off in model performance between discharge and the other variables gets more pronounced with increasing catchment wetness.