Effects of characteristics of calibration periods on building hydrologic models in the upper basins of South Korea

Study region: Republic of Korea, concentrating on the three upper basins of the largest three rivers, including multipurpose reservoirs. These three study basins are crucial to the water resources system, and their runoff records have not been affected by anthropogenic impacts. Study focus: Recent severe droughts highlight the importance of improving water resource monitoring through rainfall-runoff modeling. The effects of calibration periods have been studied to develop more precise hydrologic models. This research investigated how calibration period features impact model performance and uncertainty, utilizing ANOVA analysis. SWAT and GR4J were calibrated for three South Korean basins during dry, normal, and wet calibration periods with lengths ranging from 1 to 20 years. New hydrological insights for the region: Among the characteristics of calibration periods, hydrologic conditions had a site-dependent impact on both model performance and the contribution to uncertainty, while the effect of calibration lengths was site-independent. The benefit of including more information with a longer calibration length was irrelevant to the sites, while the distributions and severity of hydrologic conditions varied across sites. Model performance reached its maximum with longer calibration lengths, regardless of the model and basin, with performance improvements of 6.8 %, 0.3 %, and 19.9 % under calibration models conditioned on the same hydrologic conditions compared to other conditions. A similar pattern emerged: the contribution to uncertainty from hydrologic conditions was site-dependent, ranging from 1.3 % to 22.6 %, while the contribution from calibration length was not, ranging from 4.1 % to 8.1 %. These findings emphasize the importance of the calibration period's data length and hydrologic conditions for rainfall-runoff model simulation.