4.4.2 Sensitivity Analysis for Annual Peak Surface Runoff
The sensitivity indices of the GCMs, GGESs, and VIC model parameters for annual peak runoff predictions were estimated and the results are displayed in Figure 13. As seen, the most important source of uncertainty is GCM, followed by the hydrological model parameters and GGESs. The contribution of uncertainty from GCMs is in the range of 63%–84.4%, which is significantly larger than that of annual runoff (Figure 10). The uncertainty caused by hydrological parameters and GGESs are in 14.03%–36.73% and 0.14%–5.88%, respectively, and tends to show a large variation over the study period. The total variance of annual peak runoff is shown in Figure 13d with the phenomenon of the slight fluctuations of hydrological parameters and the strong fluctuations of GCMs.
The spatial distribution of sensitivity indices for the annual peak runoff depth based on the three–layer hierarchical framework is presented in Figure 14. Similar to the sensitivity analysis results of annual surface runoff predictions (Figure 12), the spatial heterogeneity shows little effect on the importance of the uncertainty sources. The most important source of uncertainty over the study basin is GCM (80%–90%), followed by the hydrological model parameters (8%–18%). In contrast, the GGESs contribute the least uncertainty (0.2%–4%) to the annual peak runoff predictions. At the spatial scale, the uncertainty due to the hydrological parameter is larger in the western and northeast regions (>18%) than in the southern regions (<12%) at all seasonal scales. However, the uncertainty caused by GCMs is comparatively lower in the western and northeast regions (<86%) at all seasonal scales.