Dependence of rainfall–runoff model performance on calibration conditions under changing climatic conditions

Calibration period selection is crucial for rainfall–runoff (RR) models in stationary and non-stationary climates. In this study, we assessed the impact of four calibration conditions on RR model performance: calibration duration, temporal lag between calibration and simulation periods, variation in rainfall volume, and rainfall patterns similarity. To ensure the generalizability of our findings, we used five RR models (GR4J, Simhyd, WASMOD, HBV, and XAJ) to simulate streamflow across two catchments in Iran with stationary and non-stationary climate conditions. Our results show that catchments exhibiting non-stationary RR relationships necessitate longer calibration periods to obtain stable simulations. Model performance gradually drops as the temporal lag between the calibration and simulation periods increases. This drop in performance is even more pronounced in catchments experiencing non-stationary climate conditions. Similarity in rainfall volume is more important than rainfall pattern similarity for predicting future streamflow in non-stationary RR relationships.