4.1 Hydrological Simulations and Predictions
We first evaluated the hydrological model simulation by dividing the recorded data series into two subperiods: 1970–1990 for model calibration and 1991–2000 for model validation. The efficacy of the model simulation was evaluated using the Nash–Sutcliffe efficiency (NSE) coefficient and relative error (RE). The performance statistics for the simulation results are summarized in Table 3. The simulated discharges generally match well with the observations, especially at the monthly timescales (see Figure S1 in Supplemental Material). For the VIC model, the NSE is greater than 0.82 at both daily and monthly timescales, and the RE is less than 8.72%, suggesting that the VIC model performs relatively well in the study basin. For the Xinanjiang model, the NSEs of the daily and monthly simulations are larger than 0.70 and 0.83, respectively, while the RE is less than 3.6% at both daily and monthly timescales. This result suggests that the Xinanjiang model is able to reasonably reproduce the basin water balance and hence can be applied to investigate the hydrological impacts of future climate change.
Based on the evaluation of the hydrological simulations, we predicted the future river discharge at the Hengshi hydrological station based on the 50 statistical downscaling ensembles of the 13 GCMs under three different GGESs. The VIC model was run on 20 sets of the parametersd2 and B , while the Xinanjiang model was run on 20 sets of parameters KC, CS, and SM . As a result, a total of 1,000 (5020) simulation samples were generated for the VIC and Xinanjiang models under each GGES. The total number of simulations used in this study thus is 6,000 (321000). The annual hydrographic comparisons of the observed discharge and that simulated by the two hydrological models using 50 GCM samples in the baseline period 1970–2000 as well as the future period 2020–2050 are shown in Figure 3a and 3b. Figure 3c–f illustrate the annual hydrographs by averaging hydrological projections from the GGESs, GCMs, hydrological models and parameters, respectively, for the baseline period 1970–2000 and the future period 2020–2050. For example, the hydrographs grouped by each GCM include 3 GGESs, 2 hydrological models and 20 hydrological parameter sets (i.e., a total of 120 hydrographs). As shown, in the baseline period (1970–2000), the simulations of the two hydrological models forced by 50 GCMs samples generally agree well with the observations (a little underestimation in May–June and overestimation in July–September, Figure 3a). In addition, the GCMs samples driven by Xinanjiang model somewhat overestimates the discharge predictions in July–February (Figure 3b). In the future period (2020–2050), the projections of mean annual discharge vary significantly among different simulation samples driven by different GGESs, different GCMs, and different parameter combinations. The results predict small increases in discharge during late autumn and early winter (November–February) and obvious decreases in spring and summer (March–September) (Figure 3a and 3b). The uncertainty of GGESs is generally limited during the 12 months due to the small differences among different GGESs (Figure 3c). The hydrological models show relatively low uncertainty range in predicting discharges during summer, but for other months the uncertainty range is relatively large and cannot be ignored (Figure 3e). The GCMs and hydrological parameters both show considerable uncertainty ranges (Figure 3d and f), suggesting a large uncertainty involved in the projections of hydrological regimes. Consequently, it is necessary to quantify the contribution of uncertainty from individual sources to enhance the projection credibility.