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.