Integrated hydrologic model calibration under non-stationary climates

Abstract

With global climate change, quantifying water availability for management under non-stationary conditions is, and will continue to be, a major challenge. When hydrologic models are calibrated to historic climatic conditions, they may lack the ability to simulate future extreme climates. This research quantified changes in model calibration under non-stationary climate conditions using the Harold L. Disney Training Center (HLDTC) site in Kentucky, USA for demonstration. An integrated hydrologic model of the site was developed using HydroGeoSphere (HGS) and was calibrated using PEST. Hydraulic conductivity (K), specific storage (Ss), and surface friction coefficient parameters were calibrated under four different climate scenarios based on two moderately-extreme precipitation events during the observation period: a. the entire observation record, including the two moderately-extreme precipitation events (base scenario), b. the entire observation record minus the short duration event (April 2017), c. the entire observation record minus the long duration event (February 2018), and d. the observation record without either event. The results demonstrate that the inclusion of observations from extreme precipitation events impact the calibration of the hydrologic model. The variations in K and Ss were the highest between scenarios of all the calibration parameters tested, while the ridge surface friction, topsoil hydraulic conductivity, or clayey sand specific storage remain unchanged. K has the greatest decrease in lateral K (x and y direction) of the clayey sand layers in Scenario D, and greatest increase in lateral K of fractured rock formation in Scenario C. This indicates the importance of lateral flow in the fractured rock during the shorter duration precipitation event. Ss changed in the fractured rock formation in Scenario B, indicating the importance of storage in the fractured rock during the longer duration precipitation event. The model constructed by this study can better capture shorter duration moderately-extreme precipitation events, demonstrated by a better match between observed and simulated hydraulic heads in Scenario C. The results also suggest that not only the presence or absence of these events informs model calibration, but the timing and duration of these events influences the parameters it informs.