2.2 GCMs and GGESs
GGESs provide plausible descriptions of how the world might evolve during the 21st century with respect to a range of variables, such as technological change, socioeconomic change, energy and land use, and emissions of greenhouse gases and air pollutants (Vuuren et al., 2011). GCMs driven by a time series of GGESs are considered as the most essential and feasible tools for projections of future climate change and have been widely applied to evaluate the hydrological impacts of climate change (Kay et al., 2009; Habets et al., 2013; Xu et al., 2013; Alfieri et al., 2015; Wu et al., 2014, 2015). In this study, the GCM simulations were obtained from the Coupled Model Intercomparison Project Phase 5 (CMIP5) database. The CMIP5 climate change projections are driven by a time series of emissions and concentrations from the representative concentration pathways (RCPs), consistent with a high energy–intensive scenario RCP8.5, a midrange mitigation emission scenario RCP4.5, and a low energy–intensive scenario RCP2.6 (IPCC, 2013). The high–resolution (0.25°×0.25°) downscaling multi–model ensemble averages of 13 CMIP5 GCMs for the baseline 1970–2000 and future period 2020–2050 with 3 different GGESs (i.e., RCP2.6, RCP4.5 and RCP8.5) were used at the time of this analysis (as shown in Table 1). For the baseline 1970–2000 and future period 2020–2050 under each GGES, a total of 50 simulation samples of daily temperature and precipitation were generated based on the following statistical downscaling process (Wu et al., 2014): (1) The GCM outputs (i.e., temperature and precipitation) were interpolated to the sites of the study basin using the bilinear interpolation method; (2) On the basis of the observed monthly temperature and precipitation at multiple sites, generalized additive model was fitted to generate local–scale simulated sequences of the baseline and future climate change scenarios for each GCM; (3) Local–scale simulated sequences from the selected GCMs were weighted averaged using Bayesian model averaging method; (4) The monthly weighted averaged climate simulations were temporally disaggregated into daily weather forcings based on the stochastic weather generation method. Each GGES consists of 50 simulated samples of daily temperature and precipitation; (5) The 50 simulated samples of daily temperature and precipitation were finally interpolated to a high–resolution grid (0.25°× 0.25°) of the study basin using the bilinear interpolation method.
2.3 Hydrological Models and Uncertain Parameters
The model uncertainty of this research consists of two different hydrological models: the variable infiltration capacity (VIC) model and Xinanjiang model. These two widely used but totally distinct models have some similar outputs and are suitable for comparison.
2.3.1 VIC Model
The VIC model is a semidistributed, grid–based, hydrological model that was developed by the University of Washington and Princeton University. The model can simulate the physical exchange of water and energy among the atmosphere, soil and vegetation in a surface vegetation–atmospheric transfer scheme (Liang et al., 1994; Lohmann et al., 1998). In this study, version 4.1.2d of the VIC model (available at www.hydro.washington.edu/Lettenmaier/Models/VIC/index.shtml) was run to simulate the water balance over 50 grid points (0.25°× 0.25°) of the study catchment. The Dag Lohmann model (Nijssen et al., 1997) was used for transporting the grid cell surface runoff and baseflow simulated by the VIC model within each grid cell to the outlet of that grid cell and then into the river system.
There are seven uncertain hydrological parameters in the VIC model: the infiltration curve shape parameter B ; the soil depth of layers 1 (d 1), 2 (d 2), and 3 (d 3); and the three base flow–related parametersDm , DS , andWS . In this study, we chose the parametersd 2 and B , since they are commonly considered as the most sensitive parameters (Demaria et al., 2007).B represents the relative area ratio of the average water content of the grid to the maximum water content of the grid. Larger Bvalues indicate greater inhomogeneity of the spatial distribution of moisture content and more surface runoff. Meanwhile, changes ind 2 can significantly increase the evaporation loss and decrease the seasonal peak discharge. The ranges of parameters were chosen based on the minimum and maximum parameter values in the Global Land Data Assimilation Systems (GLDAS) data set for China. In this research, 20 samples were generated following Latin hypercube sampling (LHS) method for these uncertain parameters. We assumed that uncertain parameters follow uniform or normal distributions (Table 2).