Multi-site calibration, validation, and sensitivity analysis of the MIKE SHE Model for a large watershed in northern China

S. Wang, Z. Zhang, G. Sun, P. Strauss, J. Guo, Y. Tang, A. Yao
2012 Hydrology and Earth System Sciences  
<p><strong>Abstract.</strong> Model calibration is essential for hydrologic modeling of large watersheds in a heterogeneous mountain environment. Little guidance is available for model calibration protocols for distributed models that aim at capturing the spatial variability of hydrologic processes. This study used the physically-based distributed hydrologic model, MIKE SHE, to contrast a lumped calibration protocol that used streamflow measured at one single watershed outlet to a multi-site
more » ... to a multi-site calibration method which employed streamflow measurements at three stations within the large Chaohe River basin in northern China. Simulation results showed that the single-site calibrated model was able to sufficiently simulate the hydrographs for two of the three stations (Nash-Sutcliffe coefficient of 0.65–0.75, and correlation coefficient 0.81–0.87 during the testing period), but the model performed poorly for the third station (Nash-Sutcliffe coefficient only 0.44). Sensitivity analysis suggested that streamflow of upstream area of the watershed was dominated by slow groundwater, whilst streamflow of middle- and down- stream areas by relatively quick interflow. Therefore, a multi-site calibration protocol was deemed necessary. Due to the potential errors and uncertainties with respect to the representation of spatial variability, performance measures from the multi-site calibration protocol slightly decreased for two of the three stations, whereas it was improved greatly for the third station. We concluded that multi-site calibration protocol reached a compromise in term of model performance for the three stations, reasonably representing the hydrographs of all three stations with Nash-Sutcliffe coefficient ranging from 0.59–072. The multi-site calibration protocol applied in the analysis generally has advantages to the single site calibration protocol.</p>
doi:10.5194/hess-16-4621-2012 fatcat:tmaojy2cgjf57cur5zaa6qw6ai