*3.2. Selection of General Circulation Models*

The general circulation model (GCM) is widely used to estimate the impacts of future climate conditions on hydrological cycles [26,49–52]. The GCMs used in this study (Table 2) were available in the Intergovernmental Panel on Climate Change (IPCC) data archive (https://pcmdi.llnl.gov/mips/cmip5/, accessed on 16 October 2016). Based on monthly precipitation data from 40 GCMs for two representative concentration scenarios (RCP4.5 and RCP8.5) and the future climate scenario period based on CMIP5, we divided GCM data into two sections. The 45 years from 1960–2004 (historical climate period, HCP) were considered the baseline period, and the 45 years from 2010–2054 were the future climate period (FCP).

**Table 2.** Summary of 40 general circulation models (GCM) selected in this study.



**Table 2.** *Cont.*

In the context of climate change, three GCMs (i.e., the dry, moderate, and wet effects) were chosen to represent the future climate conditions. Then, daily precipitation and temperature derived from GCMs were used as forcing data to project streamflow in the FCP. The representativeness of the ensemble GCMs is considerably improved in the projection of climate variables [53]. Among the 40 GCMs under the two scenarios in CMIP5, the numbers of GCMs predicting increasing and decreasing future precipitation were 36 and 4, respectively. To choose representative models and reduce uncertainties, three models were selected to simulate future climate conditions, i.e., CSIRO-Mk3-6-0 (predicting dry conditions with the largest precipitation declines), MIROC5 (wet conditions with the largest precipitation increases), and FGOALSg2 (median conditions with a median change in precipitation) (Figure 2).

To generate the mean climate conditions, the GCMs' climate projections were biascorrected with the delta-change method (for details, see Navarro-Racines, et al. [54]), which simply superimpose the mean monthly anomalies between the GCMs-simulated baseline and the future period on the observed historical precipitation and temperature to represent future climate. Specifically, first, we calculated the ratio between the observed and simulation precipitation data of the three selected GCMs in the historical period (1960–2004). Second, we multiplied or added the precipitation and temperature data of the three GCMs in the future period (2010–2054) with this ratio to obtain simulation data for the FCP. Finally, we used the simulation data as forcing input data for running SWAT (Soil and Water Assessment Tool) to estimate daily streamflow.

#### *3.3. SWAT Model*

The SWAT hydrological model is a continuous-time, computationally efficient, and semi-distributed catchment-scale hydrologic model [55]. The catchment was divided into hydrological response units (HRUs), and surface runoff volumes were simulated for each HRU. SWAT has been widely used in different catchments worldwide and proved to be an effective tool to examine hydrological responses to land use and climate changes [56]. More details on SWAT are given in Easton, et al. [57], Guo, et al. [58].

This study used daily meteorological data (precipitation, maximum and minimum temperature, mean wind speed, radiation, mean relative humidity) from 1960–2012 as forcing data to simulate daily runoff in the WRB. The performance of predicted runoff was assessed against observed daily streamflow data in the same period. In the SWAT simulation, 1983–2012 was the calibration period (warm-up period: 1983–1993), and 1960–1982 was the validation period (warm-up period: 1960–1969). Comparing the simulated runoff between the calibration and validation period, the simulation of monthly runoff using the SWAT model had a good performance in WRB (Figure 3).

**Figure 3.** The streamflow simulation using SWAT model in calibration (**a**) and validation (**b**). The blue line is the 1:1 line. The rug represents the data distribution density.

#### *3.4. Trend Analysis*

Trend analysis can provide effective and useful information on possible tendencies in the future [59]. The nonparametric Mann Kendall test was used to identify trends and trend significance in baseflow in this study. This test provides two parameters, i.e., the significance level and slope magnitude [60]. *p* values ≤ 0.05 were considered significant. The Z (derived from a certain climate element sequence) and S are the trend and order column and are used to detect the significance test. This test method has been widely employed to detect significant monotonic increasing or decreasing trends in long-term time-series data [8]. Method details can be found in previous studies [61,62].
