4.2 Sensitivity Analysis for River Discharge at Annual and Monthly Scales
The sensitivity indices of discharge predictions at the Hengshi hydrological station considering all four uncertainty sources were first estimated by using the four–layer hierarchical framework (Eq. (6)) and summarily evaluated at the annual and monthly timescales. The variability of mean annual discharge by averaging hydrological projections from the GGESs, GCMs, hydrological models and parameters is shown in Figure 4, which represents the general effects of individual uncertainty source. The variations of GCMs, hydrological model and parameters are generally large, whereas the fluctuation of GGESs is relatively low. Figure 5 displays the annual times series (2020–2050) of sensitivity indices of GGESs, GCMs, hydrological models and parameters to the discharge projections. As shown, the contributions of GCMs and hydrological parameters (GGESs) to the uncertainty vary gently (strongly) at the annual scale. The results demonstrate that the hydrological parameters and GCMs are the first and second largest uncertainty sources, respectively, with the uncertainty contribution up to 47%, implying the large predictive uncertainty caused by different GCMs as well as hydrological parameters (Figure 5b and d). The sensitivity indices of GCMs tend to firstly increase then decrease during the projection period. Compared with the GCMs and hydrological parameters, the GGESs are the least important source of uncertainty for the discharge predictions with the mean value of 4.1% (Figure 5a). This is well supported by Figure 3c, which indicates that the simulations show very limited differences among the three different GGESs. Hydrological models are the third largest uncertainties, with the contribution no more than 30% (Figure 5c). The sensitivity indices of mean annual discharge from individual uncertainty source (Figure 5a–d) is consistent with the total variance (Figure 5e).
For the monthly discharge, the averaging hydrographs are shown in Figure 6. For example, the dispersion of GGESs in January to May is larger than in other months (Figure 6a), and the dispersion of GCMs in May to August is larger than in other months (Figure 6b). Figure 7 displays the intra–annual variability of the sensitivity indices of discharge projections. The results show that the sensitivity indices of GCMs tend to increase from January to June and decrease from June to December (Figure 7b), whereas the opposite trend patterns are identified for the sensitivity indices of hydrological model and parameters (Figure 7c and d). GCMs contribute the largest uncertainty during summer season (78.88%) and show decreased uncertainty during winter season (62.80%). The largest uncertainty of GCMs in summer is probably due to the high variability of rainfall in this reason. The contribution of hydrological models and hydrological parameters to the uncertainty is lower in spring and summer than in autumn and winter. The contribution of GGESs to discharge uncertainty firstly increase from January to March and decrease from April to December (Figure 5a), with the large contribution during winter (4.36%) and spring seasons (4.64%), and the limited contribution during summer (2.74%) and autumn seasons (2.12%). The corresponding total variance is shown in Figure 7e, demonstrating that the variance tends to increase from January to May and decrease from June to December. The distinct characteristics of the total variance in dry season and rainy season are interesting.