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.