Abstract

Quantifying the uncertainty sources in assessment of climate change impacts on hydrological processes is helpful for local water management decision-making. This paper investigated the impact of the general circulation model (GCM) structural uncertainty on hydrological processes in the Kaidu River Basin. Outputs of 21 GCMs from the Coupled Model Intercomparison Project Phase 5 (CMIP5) under two representative concentration pathway (RCP) scenarios (i.e., RCP4.5 and RCP8.5), representing future climate change under uncertainty, were first bias-corrected using four precipitation and three temperature methods and then used to force a well-calibrated hydrological model (the Soil and Water Assessment Tool, SWAT) in the study area. Results show that the precipitation will increase by 3.1%–18% and 7.0%–22.5%, the temperature will increase by 2.0 °C–3.3 °C and 4.2 °C–5.5 °C and the streamflow will change by −26% to 3.4% and −38% to −7% under RCP4.5 and RCP8.5, respectively. Timing of snowmelt will shift forward by approximately 1–2 months for both scenarios. Compared to RCPs and bias correction methods, GCM structural uncertainty contributes most to streamflow uncertainty based on the standard deviation method (55.3%) while it is dominant based on the analysis of variance approach (94.1%).

INTRODUCTION

The Intergovernmental Panel on Climate Change (IPCC 2014) stated that the precipitation and temperature patterns would significantly change by the end of the 21st century. Climate change has a wide and profound impact on hydrological processes. Changes in hydrological processes and increased extreme events (e.g., drought and flood) have been detected and have exerted significant impacts on ecological and social systems (IPCC 2014). Therefore, understanding the hydro-climatic effects of future climate change is critical to local water management, especially for arid regions, where hydrological changes are more sensitive to climate change than those humid regions.

When accessing the impact of future climate change on hydrology, climate models are often coupled with hydrological models (HMs) to predict future changes in hydrological processes (Liu et al. 2010; Ficklin et al. 2013; IPCC 2014). Climate change projected by state-of-the-art general circulation models (GCMs) suggested a warming temperature trend along with seasonally and spatially varying precipitations for the 21st century (Reyers et al. 2013). As uncertainty is inherited in modelling, it is necessary to consider the uncertainties from GCMs, climate variable downscaling and hydrological modelling (Graham et al. 2007; Jiang et al. 2007; Chen et al. 2011, 2012) in local impact studies. Previous studies have investigated different uncertainty sources, and most of them noted that GCMs are one of the greatest sources of uncertainty in assessing climate change impact on hydrological processes (Horton et al. 2006; Wilby & Harris 2006; Graham et al. 2007; Chen et al. 2011, 2016; Dobler et al. 2012; Bosshard et al. 2013; Lung et al. 2013; Exbrayat et al. 2014), although uncertainty from HMs was also important over many areas of the world (Hagemann et al. 2013). More specifically, major contributors to uncertainty depend on the assessed hydrological variables (Booij 2005; Chen et al. 2011; Gampe et al. 2016; Shrestha et al. 2016) and the assessed watersheds (Finger et al. 2012; Lutz et al. 2013; Ragettli et al. 2013; Addor et al. 2014; Huss et al. 2014; Vidal et al. 2016). Methodology used for uncertainty decomposition includes qualitative visual interpretation of the prediction spread (Wilby & Harris 2006; Graham et al. 2007; Chen et al. 2011) and the quantitative standard deviation method (Xu & Xu 2012) and analysis of variance (ANOVA) approach (Bosshard et al. 2013; Addor et al. 2014; Duethmann et al. 2016). Although these studies have been conducted in many regions throughout the world, the differences between these uncertainty decomposition methods have been seldom compared.

The Tienshan Mountains, ‘water tower’ of central Asia, are the main water sources and ecological barriers, and very typical in terms of the dry and alpine continental climate characteristics together with data scarcity. The Kaidu River, a typical watershed located in the south slope of the Tienshan Mountains, is one of the headwaters of the Tarim River, the largest endorheic basin in China. Understanding future hydrological processes and their related uncertainty help the sustainable development of countries along the ‘Silk Road’ (Li et al. 2015). In the literature, most studies in this area have focused on historical hydrological events (Shi et al. 2007; Liu et al. 2010; Piao et al. 2010; Chen 2014; Rumbaur et al. 2015). Although there have been some investigations for future scenarios (Liu et al. 2010, 2011; Sorg et al. 2012; Fang et al. 2015a; Xu et al. 2016), few of them (Duethmann et al. 2016) quantified the uncertainty sources. Characterizing the uncertainty sources is of high importance for a valid interpretation of the results.

This paper aims to investigate the impact of climate change on the hydrological system and assess the impact of GCM structural uncertainty on hydrological processes. To this end, a cascade of a GCM ensemble, downscaling methods and a HM were used to simulate future hydrological processes. Three main questions are addressed: (1) How will the future climate and hydrological processes change in this arid mountainous region? (2) Which one of the following issues contributes most to future hydrological processes: GCM structure uncertainty, representative concentration pathways (RCPs) or downscaling methods? (3) Do different uncertainty decomposition methods produce different results? Understanding these issues will enable us to better assess future hydrological changes and related uncertainties. The paper is organized as follows: the section below introduces the study area; the next section describes the cascade of the GCM ensemble, bias correction methods, HM and the methodology on how to decompose the uncertainty sources; a results and discussion section follows and the final section drawing conclusions.

STUDY AREA

The Kaidu River Basin (Figure 1), with a drainage area of 18,634 km2 above the Dashankou hydrological station, is one of the four headwaters of the Tarim River. It originates in the Tienshan Mountains. Recharged mainly by rainfall and snowmelt (SM), the Kaidu River provides 42 × 108 m3 amount of water for agricultural irrigation and ecological water conveyance for the lower reaches of the Tarim River, which is crucial to the local eco-environmental and economic development. The altitude ranges from 1,340 m to 4,796 m above sea level (a.s.l.) with an average elevation of 2,995 m and average slope of 23%. This watershed has a temperate continental climate with alpine characteristics. The average annual temperature at the Bayanbulak meteorological station is −4.1 °C and annual precipitation is 278 mm; precipitation generally falls as rain from May to September and as snow from October to April of the next year. The average daily flow at the Dashankou hydrological station is around 120 m3/s (equivalent to 201 mm runoff), ranging from 15 m3/s to 973 m3/s.

Figure 1

The topography, river system, meteorological stations (Bayanbulak and Baluntai) and hydrological station (Dashankou) of the Kaidu River Basin together with its location in the Tarim River Basin (bottom right) and China (top right).

Figure 1

The topography, river system, meteorological stations (Bayanbulak and Baluntai) and hydrological station (Dashankou) of the Kaidu River Basin together with its location in the Tarim River Basin (bottom right) and China (top right).

DATA AND METHODOLOGY

Figure 2 presents the framework for the hydrological modelling under future climate change. First, daily climate predictions from 21 GCM models from CMIP5 (Coupled Model Intercomparison Project Phase 5) under RCP4.5 and RCP8.5 were downloaded (http://cmip-pcmdi.llnl.gov/cmip5/; IPCC 2013), and then these grid-based climate predictions were downscaled/bias-corrected to the station scale using four precipitation (BCpcp) and three temperature (BCtmp) bias correction methods. These bias-corrected climatic variables were used to force the well-calibrated HM of the Soil and Water Assessment Tool (SWAT). Compared to other sources (e.g., model parameter, model structure), meteorological input contributes to the largest part of uncertainties in hydrological modelling (Wilby & Harris 2006; Graham et al. 2007; Chen et al. 2011, 2016; Bosshard et al. 2013), therefore, we only used one HM in this study. In total, there were 504 hydrological simulations from 504 different combined meteorological inputs (i.e., different combinations of 21 GCMs, 2 RCPs, 4 BCpcps and 3 BCtmps), as shown in Figure 2.

Figure 2

Framework to study the future climate impact on hydrological processes. The dashed box shows different RCPs, GCMs, bias correction methods for precipitation and temperature, and the HM.

Figure 2

Framework to study the future climate impact on hydrological processes. The dashed box shows different RCPs, GCMs, bias correction methods for precipitation and temperature, and the HM.

GCMs and RCPs

The state-of-the-art climate change projections of 21 GCMs under two emission scenarios (RCP4.5 and RCP8.5) were used as climatic data in this study (Table 1). These models are from over 20 institutes or universities. Of all GCM simulated climate variables, daily precipitation, and maximum and minimum temperatures from 1975 to 2099 were used.

Table 1

Information about the GCM ensemble used in this study

No. Modelling centre Model Institution 
BCC BCC-CSM1.1-m Beijing Climate Center, China Meteorological Administration 
CCCma CanESM2 Canadian Centre for Climate Modelling and Analysis 
CMCC CMCC-CM Euro-Mediterranean Centre on Climate Change 
CNRM-CERFACS CNRM-CM5 CNRM (National Centre for Meteorological Research), CERFACS (European Center for Research and Advanced Training in Scientific Computation) 
CSIRO-BOM ACCESS1.3 CSIRO (Commonwealth Scientific and Industrial Research Organisation, Australia), and BOM (Bureau of Meteorology, Australia) 
CSIRO-QCCCE CSIRO-Mk3.6 Commonwealth Scientific and Industrial Research Organisation/Queensland Climate Change Centre of Excellence 
GCESS BNU-ESM College of Global Change and Earth System Science, Beijing Normal University 
INM INM-CM4 Institute for Numerical Mathematics 
IPSL IPSL-CM5B-LR Institute Pierre-Simon Laplace 
10 LASG-CESS FGOALS-g2 LASG, Institute of Atmospheric Physics, Chinese Academy of Sciences; and CESS, Tsinghua University 
11 MIROC MIROC5 Atmosphere and Ocean Research Institute (The University of Tokyo), National Institute for Environmental Studies, and Japan Agency for Marine-Earth Science and Technology 
12 MIROC MIROC-ESM Atmosphere and Ocean Research Institute (The University of Tokyo), National Institute for Environmental Studies, and Japan Agency for Marine-Earth Science and Technology 
13 MOHC HadGEM2-ES Met Office Hadley Centre 
14 MPI-M MPI-ESM-LR Max Planck Institute for Meteorology 
15 MRI MRI-ESM1 Meteorological Research Institute 
16 NASA GISS GISS-E2-R NASA Goddard Institute for Space Studies 
17 NCAR CCSM4 National Center for Atmospheric Research 
18 NCC NorESM1-M Norwegian Climate Centre 
19 NOAA GFDL GFDL-CM3 Geophysical Fluid Dynamics Laboratory 
20 NOAA GFDL GFDL-ESM2G Geophysical Fluid Dynamics Laboratory 
21 NSF-DOE-NCAR CESM1(BGC) National Science Foundation, Department of Energy, National Center for Atmospheric Research 
No. Modelling centre Model Institution 
BCC BCC-CSM1.1-m Beijing Climate Center, China Meteorological Administration 
CCCma CanESM2 Canadian Centre for Climate Modelling and Analysis 
CMCC CMCC-CM Euro-Mediterranean Centre on Climate Change 
CNRM-CERFACS CNRM-CM5 CNRM (National Centre for Meteorological Research), CERFACS (European Center for Research and Advanced Training in Scientific Computation) 
CSIRO-BOM ACCESS1.3 CSIRO (Commonwealth Scientific and Industrial Research Organisation, Australia), and BOM (Bureau of Meteorology, Australia) 
CSIRO-QCCCE CSIRO-Mk3.6 Commonwealth Scientific and Industrial Research Organisation/Queensland Climate Change Centre of Excellence 
GCESS BNU-ESM College of Global Change and Earth System Science, Beijing Normal University 
INM INM-CM4 Institute for Numerical Mathematics 
IPSL IPSL-CM5B-LR Institute Pierre-Simon Laplace 
10 LASG-CESS FGOALS-g2 LASG, Institute of Atmospheric Physics, Chinese Academy of Sciences; and CESS, Tsinghua University 
11 MIROC MIROC5 Atmosphere and Ocean Research Institute (The University of Tokyo), National Institute for Environmental Studies, and Japan Agency for Marine-Earth Science and Technology 
12 MIROC MIROC-ESM Atmosphere and Ocean Research Institute (The University of Tokyo), National Institute for Environmental Studies, and Japan Agency for Marine-Earth Science and Technology 
13 MOHC HadGEM2-ES Met Office Hadley Centre 
14 MPI-M MPI-ESM-LR Max Planck Institute for Meteorology 
15 MRI MRI-ESM1 Meteorological Research Institute 
16 NASA GISS GISS-E2-R NASA Goddard Institute for Space Studies 
17 NCAR CCSM4 National Center for Atmospheric Research 
18 NCC NorESM1-M Norwegian Climate Centre 
19 NOAA GFDL GFDL-CM3 Geophysical Fluid Dynamics Laboratory 
20 NOAA GFDL GFDL-ESM2G Geophysical Fluid Dynamics Laboratory 
21 NSF-DOE-NCAR CESM1(BGC) National Science Foundation, Department of Energy, National Center for Atmospheric Research 

RCP4.5 (lower emission scenario) is a stabilization scenario with the total radiative forcing rising until 2070, which will remain stable at 4.5 W/m2. In contrast, RCP8.5 (higher emission scenario) is a continuously rising radiative forcing pathway (at a target of 8.5 W/m2 in 2,100) with a further enhanced residual circulation and significant CH4 increase (Van Vuuren et al. 2011). RCP4.5 and RCP8.5 are equivalent to B1 and A2 of the Special Report on Emission Scenarios (SRES).

Downscaling/bias correction methods

To account for the low spatial resolution in GCM outputs, four BCpcp and three BCtmp methods were used to downscale the grid-based GCM outputs to the station scale (where there is a meteorological station). These correction methods are local intensity scaling (LOCI), power transformation (PT), distribution mapping (DM) and quantile mapping (QM) for precipitation as well as linear scaling (LS), variance scaling (VARI) and DM for temperature. These methods can be classified into mean-based (LS and LOCI), variance-based (PT and VARI) and distribution-based approaches (DM and QM). Table 2 briefly describes the characteristics of each method. The methods have been widely used in downscaling and bias correcting the climate model outputs (e.g., Schmidli et al. 2006; Fang et al. 2015b).

Table 2

Characteristics of bias correction methods for temperature and precipitation

Approach Characteristics Reference 
Precipitation 
 LOCI It corrects the wet-day frequencies and intensities by setting all precipitation values less than a wet-day threshold to zeros Schmidli et al. (2006) and Fang et al. (2015b)  
 PT It adjusts the standard deviation of the precipitation series by producting a factor for each month Teutschbein & Seibert (2012)  
 DM To match the assumed distribution function of the raw data to that of the observations by assuming the raw and the observed precipitation follow the gamma distribution Block et al. (2009) and Piani et al. (2010)  
 QM It is non-parametric and is generally applicable for all possible distributions of precipitation without any assumption on its distribution Themeßl et al. (2012) and Chen et al. (2013)  
Temperature 
 LS To perfectly match the monthly average of corrected values with that of observed ones by adding a factor Lenderink et al. (2007)  
 VARI To correct both the mean and variance of temperature Terink et al. (2010) and Teutschbein & Seibert (2012)  
 DM To match the assumed distribution function of the raw data to that of the observations by assuming the raw and the observed temperatures follow normal distributions Teutschbein & Seibert (2012)  
Approach Characteristics Reference 
Precipitation 
 LOCI It corrects the wet-day frequencies and intensities by setting all precipitation values less than a wet-day threshold to zeros Schmidli et al. (2006) and Fang et al. (2015b)  
 PT It adjusts the standard deviation of the precipitation series by producting a factor for each month Teutschbein & Seibert (2012)  
 DM To match the assumed distribution function of the raw data to that of the observations by assuming the raw and the observed precipitation follow the gamma distribution Block et al. (2009) and Piani et al. (2010)  
 QM It is non-parametric and is generally applicable for all possible distributions of precipitation without any assumption on its distribution Themeßl et al. (2012) and Chen et al. (2013)  
Temperature 
 LS To perfectly match the monthly average of corrected values with that of observed ones by adding a factor Lenderink et al. (2007)  
 VARI To correct both the mean and variance of temperature Terink et al. (2010) and Teutschbein & Seibert (2012)  
 DM To match the assumed distribution function of the raw data to that of the observations by assuming the raw and the observed temperatures follow normal distributions Teutschbein & Seibert (2012)  

HM and model setup

SWAT (Arnold et al. 1998), developed at the Agriculture Research Service of the United States Department of Agriculture, has been widely used for comprehensive modelling of the impacts of management practices and climate change on hydrological processes at a watershed scale (e.g., Jayakrishnan et al. 2005; Singh et al. 2015; Awan et al. 2016; Tamm et al. 2016). To represent the spatial variability, a watershed is first disaggregated into subbasins and each subbasin is further divided into hydrological response units based on soil and land use data. For more details, refer to SWAT manuals (http://www.brc.tamus.edu/).

The SWAT model was successfully applied in the Kaidu River Basin (Fang et al. 2015c). The SWAT model was first set up with digital elevation model (DEM) (www2.jpl.nasa.gov/srtm/), land use (from the Environmental and Ecological Science Data Centre for West China), soil map (from Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences), and observed meteorological data (at two meteorological stations Bayanbulak and Baluntai, Figure 1; China Meteorological Data Sharing Service System), which form the base for estimating meteorological input for each subbasin with the elevation band method. Then, model parameters were calibrated with the observed streamflow data at Dashankou station (hydrological station in Figure 1) and good model performances were achieved for both calibration and validation periods, as shown in Table 3. More details of the SWAT model setup and calibration can be found in Fang et al. (2015c).

Table 3

Performance of SWAT model forced by the observed meteorological data

Statistics NS PBIAS R2 
Daily streamflow (calibration period 1986–1989) 0.80 0.01% 0.80 
Daily streamflow (validation period 1990–2002) 0.81 2.94% 0.81 
3DMHF (validation period 1990–2002) 0.43 −19.6% 0.98 
7DMLF (validation period 1990–2002) 0.72 7.5% 0.87 
Statistics NS PBIAS R2 
Daily streamflow (calibration period 1986–1989) 0.80 0.01% 0.80 
Daily streamflow (validation period 1990–2002) 0.81 2.94% 0.81 
3DMHF (validation period 1990–2002) 0.43 −19.6% 0.98 
7DMLF (validation period 1990–2002) 0.72 7.5% 0.87 

When studying the impact of climate change on flow extremes, we used average annual 3-day maximum high flow (3DMHF) and average annual 7-day minimum low flow (7DMLF), as used in Sanborn & Bledsoe (2006).

Quantification of uncertainties from GCMs, RCPs, BCpcp and BCtmp

To analyse the impact of climate change on hydrological processes, five periods are defined: 1986–2005 (control period), 2020–2039, 2040–2059, 2060–2079 and 2080–2099. For future changes of the precipitation, temperature and streamflow, we calculated the ensemble median and 25th and 75th percentiles instead of the mean value because these statistics have a lower sensitivity to outliers than the ensemble mean (Benestad 2004).

There have been several uncertainty quantification methods applied in climate change impact studies, as proposed in the ‘Introduction’. Standard deviation and ANOVA approach are two quantitative methods commonly used in assessing the contributions of GCMs, downscaling methods, RCPs and HMs to the total variance of hydrological variables within an ensemble of climate models (Yip et al. 2011; Déqué et al. 2012; Gampe et al. 2016; Vidal et al. 2016). Here, we used these two methods to decompose the contributions of GCMs, RCPs, BCpcp and BCtmp to prediction uncertainty.

The model ensemble consists of 504 model runs (combinations of 21 GCMs, 2 RCPs, 4 BCpcps and 3 BCtmps, as in Figure 2). For each model run, the hydrological signal Y (i.e., relative change of a hydrological variable) of the future periods (i.e., 2020–2039, 2040–2059, 2060–2079 and 2080–2099) was calculated as:  
formula
(1)
where QFUT and QCTL are values of a given hydrological variable at the future period and control period. The decomposition of its total uncertainty can be conducted by either the standard deviation or ANOVA.

Standard deviation method

In the standard deviation method, the uncertainty from source A can be derived as the mean standard deviation of the model ensemble by varying source A, while keeping other sources (denoted as A∼ in the equation below) constant:  
formula
(2)
where μ(.) and represent the mean and the standard deviation operator. For example, the uncertainty from GCMs can be obtained through Equation (2) by varying GCMs from 1st to 21st while keeping RCP, BCpcp and BCtmp constant.

ANOVA approach

When applying the ANOVA approach, each uncertainty source is taken as an ‘effect’ which has an influence on Y. The total sum of squares (SSQtot) can be decomposed into sums of squares of the individual uncertainty source, their interactions (SSQInteraction) and an error term SSE. In this study, we ignored the interactions and ANOVA is expressed as:  
formula
(3)
where SSQGCM, SSQRCP, SSQBCpcp and SSQBCtmp are sums of squares of GCM, RCP, BCpcp and BCtmp, respectively. As indicated by Bosshard et al. (2013) and Duethmann et al. (2016), variances calculated with Equation (3) are overestimated by a factor of where N is the sample size of the ith source. Here we offset this by multiplying the uncertainty contribution with a factor to keep the sum of SSQGCM, SSQRCP, SSQBCpcp and SSQBCtmp constant.  
formula
(4)
Then, we can get the contribution of the ith uncertainty source to the total ensemble uncertainty:  
formula
(5)
The signal-to-noise ratio (S/N) was used to quantitively reveal the robustness of the projected streamflow (Zhou & Yu 2006; Addor et al. 2014). Assume that Y(n,t) is the relative changes of future streamflow from the nth simulation (n = 1, 2, … , 252) at tth year for 1986–2005 and 2080–2099 for RCP4.5 or RCP8.5, and the multisimulation mean at year t. The S/N can be represented by:  
formula
(6)
 
formula
(7)
 
formula
(8)

The larger S/N is, the higher the credibility and more significant are the changes in the projected streamflow. Normally, S/N > 1 indicates that the projections are credible to a certain extent while S/N < 1 indicates low credibility or with unsignificant changes (Zhou & Yu 2006).

Further, the consistency of the simulations is estimated as the ratio of the number of simulations with negative projected relative streamflow changes to the number of all simulations (N = 504 in this case).

RESULTS AND DISCUSSION

Performance of the HM

Evaluation statistics in Table 3 indicate good model performances forced by the observed climate variables: daily Nash–Sutcliffe coefficients (NS) (Nash & Sutcliffe 1970) larger than 0.8 and percent biases (PBIAS) within ±10% (Table 3) for both calibration and validation periods. For the high flow and low flow, ‘PBIAS’s of the 3DMHF and 7DMLF are −19.6% and 7.5%, respectively, during 1990–2002. The validation period is three times longer than the calibration period, indicating the calibrated HM is robust, and therefore could be used to study the impact of climate change on local hydrological processes.

Figure 3 compares observed streamflow series, and simulated streamflow series driven by observed meteorological inputs and bias-corrected GCM outputs (combinations of 21 GCMs, 4 ‘BCpcp’s and 3 ‘BCtmp’s) in the control period. The 90% percentile (shaded) of simulated streamflow driven with bias-corrected GCM outputs bracketed both the monthly average observations (left plot) and the observed streamflows at each exceedance (right plot). In addition, ‘PBIAS’s for 3DMHF and 7DMLF ranged from −39.4% to 77.6% and −55.5% to 150.4% with 67% and 34% of these simulations having absolute PBIAS values within 20%. The model performances show generally satisfactory results and could be used to study the impact of climate change.

Figure 3

Observed (dots) and simulated streamflows forced by the observed meteorological data (lines) and bias-corrected GCM outputs (shaded area representing 90% percentile) for the control period: (a) monthly average streamflows and (b) exceedance probability curve of the streamflows.

Figure 3

Observed (dots) and simulated streamflows forced by the observed meteorological data (lines) and bias-corrected GCM outputs (shaded area representing 90% percentile) for the control period: (a) monthly average streamflows and (b) exceedance probability curve of the streamflows.

Projected changes in the precipitation and temperature

The projected precipitation and temperature changes are presented in Figure 4. The medians of annual precipitation change are 8% and 16%, while their 25% and 75% quantiles are 3.1%–18% and 7.0%–22.5% under RCP4.5 and RCP8.5, respectively, for 2080–2099. There is a significant seasonal variation in the precipitation change with a substantial increase in the cold season (November to April of the next year) and small increase during summer (June, July and August) with large differences among GCMs. The uncertainty bands of projected precipitation gradually increases throughout the 21st century.

Figure 4

GCM projected monthly precipitation change , temperature change at the Bayanbulak station, and projected streamflow change under RCP4.5 (grey) and RCP8.5 (black) for the four periods in the 21st century (four boxes sequentially).

Figure 4

GCM projected monthly precipitation change , temperature change at the Bayanbulak station, and projected streamflow change under RCP4.5 (grey) and RCP8.5 (black) for the four periods in the 21st century (four boxes sequentially).

For temperature, all GCMs projected a continuous increasing trend with the median increasements being 3.6 °C and 6.5 °C, and their 25% and 75% quantiles ranging from 2.0 °C to 3.3 °C and 4.2 °C to 5.5 °C, respectively, under RCP4.5 and RCP8.5 for 2080–2099 (Figure 4). Compared to precipitation, uncertainty in temperature projections are considerably smaller.

Projected streamflow change

Changes in streamflow volume and timing

Changes in the precipitation and temperature led to changes in the streamflow. The following results are based on 504 simulations forced by all combinations of RCPs, GCMs, BCtmp and BCpcp. Figure 4 shows the streamflow changes with their 25% and 75% quantiles. This indicates there is no general conclusion that these changes are definitely positive or negative or have the same magnitude. The largest decrease will be likely to occur during 2080–2099 under RCP8.5. The medians of annual streamflow change are −12.5% and −18%, while their 25% and 75% quantiles are −26% to 3.4% and −38% to −7% under RCP4.5 and RCP8.5, respectively, for 2080–2099. Note that most models project decreasing streamflow after the 2060s due to the continuously rising temperature. Seasonally, monthly average streamflow decreases by 15% and 27% for the summer season (June, July and August) while it increases by 3.0% and 3.7% for spring (March, April and May) under RCP4.5 and RCP8.5 in 2080–2099.

In a previous study conducted in the Kaidu River Basin, Liu et al. (2011) reached the conclusion that the qualitative impact results were highly consistent, while they are not in our case. As we used 21 GCMs and different bias correction methods, the climate change scenarios were expected to have larger ranges. This is further discussed based on credibility and consistency indices in the text below. Differences can also be found in the streamflow projections during the snow melt period. Liu et al. (2011) concluded that the flow changes are strongly positive during April–May (17.7%–29.7%) for 2046–2065 using a lumped conceptual model (VHM). Our conclusion is comparable to Liu et al. (2011) and Fang et al. (2015a), but with larger width (−8.3% to 30.6% under RCP4.5 and −18.4% to 23.6% under RCP8.5 during the counterpart period). The reason may be related to the fact that multiple bias correction methods were used in our study while only one (a perturbation approach, equivalent to QM) was used in Liu et al. (2011). The decrease in summer runoff has been supported in many other regions, e.g., the Aksu River in south Tienshan Mountains (Duethmann et al. 2016), a forested Canadian watershed (Chen et al. 2011), the midlatitude alpine regions of the Swiss Alps (Addor et al. 2014) and the Colorado River Basin (Christensen & Lettenmaier 2007).

Changes in high flows and low flows

Future changes in high flows and low flows represented by 3DMHF and 7DMLF are shown in Figure 5. The median changes in the high flows for different exceedances are projected to range from −8.5% to 17.4%, while those in low flow will decrease by 12.4% to 46.7%, which may result in potential drought and hinder agriculture irrigation for the oasis in the lower reaches, especially under RCP8.5. We should be careful when interpreting changes in the low flows as only 32% of the simulations have absolute PBIAS values within 20% for the control period. For the high flow, the extremely high flood, e.g., exceedance <0.1, will not have a significantly increasing trend, while the relatively small peaks with exceedance between 0.8 and 1.0 will increase, which may help ecological recovery in the lower reaches of the Kaidu River. Many previous studies (Ragettli et al. 2016; Zhang et al. 2016) concluded that the extreme flow has been increasing or will increase in many mountainous regions, e.g., the Aksu River Basin amd the Langtang River. This may be related to these rivers having a considerable part of the runoff fed by glacier melt water, while the contribution of glacier melt to runoff in the Kaidu River basin is approximately 10%, which cannot generate a severe flood under a warmer climate.

Figure 5

Projected changes in high and low flows under RCP4.5 (grey) and RCP8.5 (black) for 2080–2099 compared to the control period, whose values are shown as black stars on the right axis.

Figure 5

Projected changes in high and low flows under RCP4.5 (grey) and RCP8.5 (black) for 2080–2099 compared to the control period, whose values are shown as black stars on the right axis.

Changes in hydrological components

Figure 6 shows the projected changes in SM, surface streamflow (Rs), subsurface streamflow (Rg) and evapotranspiration (ET) for 2080–2099 under RCP8.5 (changes in the hydrological components under RCP4.5 (also shown) are similar but smaller and not discussed here). The changes exhibit an obvious seasonality, i.e., insignificant from October to March and significant from April to September during which SM and rainfall occur. SM increases by 16% and 18% in March to May and 49% and 79% in June to August under RCP4.5 and RCP8.5, respectively, for 2080–2099. The contribution of SM to streamflow will decrease from 0.22 for the control period to 0.19 and 0.17 for 2080–2099 under RCP4.5 and RCP8.5, respectively, which means that SM is decreasing in importance. The snow melting time will shift forward approximately 1–2 months. This shifting may be partially attributed to the increased temperature, which governs snow melt, as demonstrated by other studies (Barnett et al. 2005; Moore et al. 2007). Rs will shift forward with more water generated during March to May and less water during June to August. Changes in the annual Rg are insignificant (4%–5%), which indicates the groundwater flow is the most stable component. ET will increase throughout the 21st century with a median increment of 24%–42%.

Figure 6

Monthly average values of SM, surface streamflow (Rs), subsurface streamflow (Rg) and ET for the control period (black line) and 2080–2099 with 50% uncertainty bands.

Figure 6

Monthly average values of SM, surface streamflow (Rs), subsurface streamflow (Rg) and ET for the control period (black line) and 2080–2099 with 50% uncertainty bands.

Uncertainty decomposition

Table 4 lists uncertainty contributions of streamflow from GCMs, RCPs, BCpcp and BCtmp using the standard deviation method and the ANOVA approach. For both methods, GCMs is the most important uncertainty source in streamflow projection, which coincides with previous studies (Buytaert et al. 2010; Chen et al. 2011; Bosshard et al. 2013). Based on the standard deviation method, all contributions increase over these four periods slightly, e.g., contributions related to GCMs and RCPs increased from 0.215 and 0.093 during 2020–2049 to 0.345 and 0.124 during 2080–2099, respectively, indicating that the uncertainties from each source have increased (Wilby & Harris 2006; Exbrayat et al. 2014). Based on ANOVA, GCMs dominates the uncertainty with its contributions ranging from 0.907 to 0.967, while other sources can be ignored. The uncertainty proportion of GCM or RCP did not show an obvious increase with time based on ANOVA, which is different from Wilby & Harris (2006).

Table 4

Period average of uncertainty decomposition based on standard deviation method and ANOVA

  Standard deviation method
 
ANOVA
 
GCM RCP BCpcp BCtmp GCM RCP BCpcp BCtmp 
2020–2039 0.215 0.093 0.061 0.036 0.947 0.033 0.017 0.003 
2040–2059 0.300 0.094 0.082 0.038 0.967 0.009 0.023 0.001 
2060–2079 0.303 0.110 0.098 0.042 0.907 0.053 0.035 0.005 
2080–2099 0.345 0.124 0.115 0.049 0.943 0.014 0.039 0.004 
Average 0.291 0.105 0.089 0.041 0.941 0.027 0.028 0.003 
  Standard deviation method
 
ANOVA
 
GCM RCP BCpcp BCtmp GCM RCP BCpcp BCtmp 
2020–2039 0.215 0.093 0.061 0.036 0.947 0.033 0.017 0.003 
2040–2059 0.300 0.094 0.082 0.038 0.967 0.009 0.023 0.001 
2060–2079 0.303 0.110 0.098 0.042 0.907 0.053 0.035 0.005 
2080–2099 0.345 0.124 0.115 0.049 0.943 0.014 0.039 0.004 
Average 0.291 0.105 0.089 0.041 0.941 0.027 0.028 0.003 

The uncertainty from GCMs based on the standard deviation method is the most important, which accounts for 55.3% of the total uncertainty, which is much lower than that based on the ANOVA approach (over 90% uncertainty caused by the GCMs). The reason may be that the ANOVA uses the square index to quantify the uncertainty contribution, which tends to favour high uncertainty sources.

The uncertainty result demonstrates the high contribution of climate models in uncertainty estimation of streamflow and suggests that the most effective way to reduce projection uncertainty is to reduce uncertainties in climatic predictions, as shown in Zhang et al. (2015).

As uncertainty in hydrological modelling is inevitable, it is important to determine the credibility and robustness of the projected streamflow change. Here, we used signal-to-noise ratio (S/N) (Zhou & Yu 2006; Addor et al. 2014) to represent this credibility. The median S/N for 2080–2099 is only 0.214 and 0.246 (less than 1) for RCP4.5 and RCP8.5 (Figure 7), respectively. Higher credibility was found in winter months (January, February, November and December) and SM season (April) under RCP8.5, indicating that the projection has higher credibility in these months compared to other months.

Figure 7

The signal-to-noise ratio (S/N) (top) and consistencies (bottom) based on the projected streamflows in 2080–2099 under RCP4.5 and RCP8.5.

Figure 7

The signal-to-noise ratio (S/N) (top) and consistencies (bottom) based on the projected streamflows in 2080–2099 under RCP4.5 and RCP8.5.

Furthermore, the consistency is estimated based on 504 simulations. Approximately 63.4% and 66.9% of the simulations demonstrate a negative streamflow change in 2080–2099 under RCP4.5 and RCP8.5 compared to 1986–2005. Most simulations (70.0% and 74.4%) show decreasing trends during June to February in the next year while only 30.9% and 29.4% of the simulations show a decreasing trend in April for these two scenarios. The S/N and consistency results both suggest that the streamflows are likely to increase in April and decrease in winter months, and for other months the changes are with high uncertainty, thus caution needs to be taken when used for decision-making.

CONCLUSIONS

In this study, we analysed the climate change impacts on hydrological processes for an important headwater of the Tarim River Basin with uncertainty analysis. A well-calibrated SWAT model forced by the downscaled and bias-corrected outputs of 21 GCMs was applied to investigate the effects of climate change on hydrological processes and the impact of GCM structural uncertainty on the hydrological processes.

While all the state-of-the-art GCMs predicted an increased temperature, the predicted precipitation has both decreasing and increasing trends of different magnitudes. Precipitation will increase by 3.1%–18% and 7.0%–22.5% while temperature will increase by 2.0 °C–3.3 °C and 4.2 °C–5.5 °C (represented by their 25% and 75% quantiles), respectively, for 2080–2099 under RCP4.5 and RCP8.5.

For the 21st century, streamflow is likely to increase until the 2060s and then decrease thereafter. Streamflow will change by −26% to 3.4% under RCP4.5 and by −38% to −7% under RCP8.5, respectively, for 2080–2099. Seasonally, streamflow will decrease by −27% and −15% for the summer months (June, July and August), while it will increase by 3.0% and 3.7% in spring (March, April and May) under RCP4.5 and RCP8.5, which may result in a potential water shortage during the critical water-demand summer. The seasonal shift of streamflow may be related to the spring freshet because SM will shift forward for approximately 1–2 months.

GCMs-related uncertainty was the most important based on the standard deviation method and ANOVA approach, while uncertainties linked to RCPs and bias corrections for precipitation and temperature are less important. The standard deviation method generated more mediocre results compared to the ANOVA approach.

Although the impacts of climate change on hydrological processes have been investigated in many previous studies, this study presents a complete study on future hydrological changes, highlighting the uncertainties caused by climate models. This study provides useful information on predicting uncertainty and credibility for water resource management and agricultural planning.

ACKNOWLEDGEMENTS

The research was supported by the ‘Thousand Youth Talents’ Plan (Xinjiang Project Y371051), the National Natural Science Foundation of China (41630859) and the CAS ‘Light of West China’ Program (2016-QNXZ-B-12).

REFERENCES

REFERENCES
Addor
,
N.
,
Rössler
,
O.
,
Köplin
,
N.
,
Huss
,
M.
,
Weingartner
,
R.
&
Seibert
,
J.
2014
Robust changes and sources of uncertainty in the projected hydrological regimes of Swiss catchments
.
Water Resources Research
50
,
7541
7562
.
Arnold
,
J. G.
,
Srinivasan
,
R.
,
Muttiah
,
R. S.
&
Williams
,
J.
1998
Large area hydrologic modeling and assessment part I: model development
.
JAWRA Journal of the American Water Resources Association
34
(
1
),
73
89
.
Barnett
,
T. P.
,
Adam
,
J. C.
&
Lettenmaier
,
D. P.
2005
Potential impacts of a warming climate on water availability in snow-dominated regions
.
Nature
438
(
7066
),
303
309
.
Block
,
P. J.
,
Souza Filho
,
F. A.
,
Sun
,
L.
&
Kwon
,
H. H.
2009
A streamflow forecasting framework using multiple climate and hydrological models
.
JAWRA Journal of the American Water Resources Association
45
,
828
843
.
Bosshard
,
T.
,
Carambia
,
M.
,
Goergen
,
K.
,
Kotlarski
,
S.
,
Krahe
,
P.
,
Zappa
,
M.
&
Schär
,
C.
2013
Quantifying uncertainty sources in an ensemble of hydrological climate-impact projections
.
Water Resources Research
49
(
3
),
1523
1536
.
Buytaert
,
W.
,
Vuille
,
M.
,
Dewulf
,
A.
,
Urrutia
,
R.
,
Karmalkar
,
A.
&
Celleri
,
R.
2010
Uncertainties in climate change projections and regional downscaling in the tropical Andes: implications for water resources management
.
Hydrology and Earth System Sciences
14
(
7
),
1247
1258
.
Chen
,
Y.
2014
Water Resources Research in Northwest China
.
Springer
,
Dordrecht
,
The Netherlands
.
Chen
,
J.
,
Brissette
,
F. P.
,
Poulin
,
A.
&
Leconte
,
R.
2011
Overall uncertainty study of the hydrological impacts of climate change for a Canadian watershed
.
Water Resources Research
47
(
12
),
W12509
.
Déqué
,
M.
,
Somot
,
S.
,
Sanchez-Gomez
,
E.
,
Goodess
,
C.
,
Jacob
,
D.
,
Lenderink
,
G.
&
Christensen
,
O.
2012
The spread amongst ENSEMBLES regional scenarios: regional climate models, driving general circulation models and interannual variability
.
Climate Dynamics
38
,
951
964
.
Dobler
,
C.
,
Hagemann
,
S.
,
Wilby
,
R. L.
&
Stoetter
,
J.
2012
Quantifying different sources of uncertainty in hydrological projections in an Alpine watershed
.
Hydrology and Earth System Sciences
16
(
11
),
4343
4360
.
Exbrayat
,
J. F.
,
Buytaert
,
W.
,
Timbe
,
E.
,
Windhorst
,
D.
&
Breuer
,
L.
2014
Addressing sources of uncertainty in runoff projections for a data scarce catchment in the Ecuadorian Andes
.
Climatic Change
125
,
221
235
.
Fang
,
G.
,
Yang
,
J.
,
Chen
,
Y.
,
Zhang
,
S.
,
Deng
,
H.
,
Liu
,
H.
&
De Maeyer
,
P.
2015a
Climate change impact on the hydrology of a typical watershed in the Tianshan Mountains
.
Advances in Meteorology
2015
,
1
10
.
Fang
,
G. H.
,
Yang
,
J.
,
Chen
,
Y. N.
&
Zammit
,
C.
2015b
Comparing bias correction methods in downscaling meteorological variables for a hydrologic impact study in an arid area in China
.
Hydrology and Earth System Sciences
19
(
6
),
2547
2559
.
Fang
,
G.
,
Yang
,
J.
,
Chen
,
Y.
,
Xu
,
C.
&
De Maeyer
,
P.
2015c
Contribution of meteorological input in calibrating a distributed hydrologic model in a watershed in the Tianshan Mountains, China
.
Environmental Earth Sciences
74
(
3
),
2413
2424
.
Graham
,
L. P.
,
Hagemann
,
S.
,
Jaun
,
S.
&
Beniston
,
M.
2007
On interpreting hydrological change from regional climate models
.
Climatic Change
81
,
97
122
.
Hagemann
,
S.
,
Chen
,
C.
,
Clark
,
D. B.
,
Folwell
,
S.
,
Gosling
,
S. N.
,
Haddeland
,
I.
,
Hanasaki
,
N.
,
Heinke
,
J.
,
Ludwig
,
F.
,
Voss
,
F.
&
Wiltshire
,
A. J.
2013
Climate change impact on available water resources obtained using multiple global climate and hydrology models
.
Earth System Dynamics
4
,
129
144
.
Horton
,
P.
,
Schaefli
,
B.
,
Mezghani
,
A.
,
Hingray
,
B.
&
Musy
,
A.
2006
Assessment of climate-change impacts on alpine discharge regimes with climate model uncertainty
.
Hydrological Processes
20
,
2091
2109
.
Huss
,
M.
,
Zemp
,
M.
,
Joerg
,
P. C.
&
Salzmann
,
N.
2014
High uncertainty in 21st century runoff projections from glacierized basins
.
Journal of Hydrology
510
,
35
48
.
IPCC
2013
The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change
.
Cambridge
,
UK and New York
,
USA
,
1552 pp
.
IPCC
2014
Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change
.
Cambridge University Press
,
Cambridge
,
UK
and
New York, USA, 1132 pp
.
Jayakrishnan
,
R.
,
Srinivasan
,
R.
,
Santhi
,
C.
&
Arnold
,
J.
2005
Advances in the application of the SWAT model for water resources management
.
Hydrological Processes
19
(
3
),
749
762
.
Jiang
,
T.
,
Chen
,
Y. D.
,
Xu
,
C.
,
Chen
,
X.
,
Chen
,
X.
&
Singh
,
V. P.
2007
Comparison of hydrological impacts of climate change simulated by six hydrological models in the Dongjiang Basin, South China
.
Journal of Hydrology
336
,
316
333
.
Lenderink
,
G.
,
Buishand
,
A.
&
Deursen
,
W.
2007
Estimates of future discharges of the river Rhine using two scenario methodologies: direct versus delta approach
.
Hydrology and Earth System Sciences
11
,
1145
1159
.
Li
,
P.
,
Qian
,
H.
,
Howard
,
K. W. F.
&
Wu
,
J.
2015
Building a new and sustainable ‘Silk Road economic belt’
.
Environmental Earth Sciences
74
,
7267
7270
.
Liu
,
Z.
,
Xu
,
Z.
,
Huang
,
J.
,
Charles
,
S. P.
&
Fu
,
G.
2010
Impacts of climate change on hydrological processes in the headwater catchment of the Tarim River basin, China
.
Hydrological Processes
24
(
2
),
196
208
.
Liu
,
T.
,
Willems
,
P.
,
Pan
,
X. L.
,
Bao
,
A. M.
,
Chen
,
X.
,
Veroustraete
,
F.
&
Dong
,
Q. H.
2011
Climate change impact on water resource extremes in a headwater region of the Tarim basin in China
.
Hydrology and Earth System Sciences
15
(
11
),
3511
3527
.
Lung
,
T.
,
Dosio
,
A.
,
Becker
,
W.
,
Lavalle
,
C.
&
Bouwer
,
L. M.
2013
Assessing the influence of climate model uncertainty on EU-wide climate change impact indicators
.
Climatic Change
120
,
211
227
.
Lutz
,
A. F.
,
Immerzeel
,
W. W.
,
Gobiet
,
A.
,
Pellicciotti
,
F.
&
Bierkens
,
M. F. P.
2013
Comparison of climate change signals in CMIP3 and CMIP5 multi-model ensembles and implications for Central Asian glaciers
.
Hydrology and Earth System Sciences
17
,
3661
3677
.
Moore
,
J. N.
,
Harper
,
J. T.
&
Greenwood
,
M. C.
2007
Significance of trends toward earlier snowmelt runoff, Columbia and Missouri Basin headwaters, western United States
.
Geophysical Research Letters
34
(
16
),
L16402
.
Piani
,
C.
,
Weedon
,
G. P.
,
Best
,
M.
,
Gomes
,
S. M.
,
Viterbo
,
P.
,
Hagemann
,
S.
&
Haerter
,
J. O.
2010
Statistical bias correction of global simulated daily precipitation and temperature for the application of hydrological models
.
Journal of Hydrology
395
,
199
215
.
Piao
,
S.
,
Ciais
,
P.
,
Huang
,
Y.
,
Shen
,
Z.
,
Peng
,
S.
,
Li
,
J.
,
Zhou
,
L.
,
Liu
,
H.
,
Ma
,
Y.
&
Ding
,
Y.
2010
The impacts of climate change on water resources and agriculture in China
.
Nature
467
(
7311
),
43
51
.
Ragettli
,
S.
,
Pellicciotti
,
F.
,
Bordoy
,
R.
&
Immerzeel
,
W. W.
2013
Sources of uncertainty in modeling the glaciohydrological response of a Karakoram watershed to climate change
.
Water Resources Research
49
,
6048
6066
.
Ragettli
,
S.
,
Immerzeel
,
W. W.
&
Pellicciotti
,
F.
2016
Contrasting climate change impact on river flows from high-altitude catchments in the Himalayan and Andes Mountains
.
Proceedings of the National Academy of Sciences
113
,
9222
9227
.
Rumbaur
,
C.
,
Thevs
,
N.
,
Disse
,
M.
,
Ahlheim
,
M.
,
Brieden
,
A.
,
Cyffka
,
B.
,
Duethmann
,
D.
,
Feike
,
T.
,
Frör
,
O.
,
Gärtner
,
P.
,
Halik
,
Ü.
,
Hill
,
J.
,
Hinnenthal
,
M.
,
Keilholz
,
P.
,
Kleinschmit
,
B.
,
Krysanova
,
V.
,
Kuba
,
M.
,
Mader
,
S.
,
Menz
,
C.
,
Othmanli
,
H.
,
Pelz
,
S.
,
Schroeder
,
M.
,
Siew
,
T. F.
,
Stender
,
V.
,
Stahr
,
K.
,
Thomas
,
F. M.
,
Welp
,
M.
,
Wortmann
,
M.
,
Zhao
,
X.
,
Chen
,
X.
,
Jiang
,
T.
,
Luo
,
J.
,
Yimit
,
H.
,
Yu
,
R.
,
Zhang
,
X.
&
Zhao
,
C.
2015
Sustainable management of river oases along the Tarim River (SuMaRiO) in Northwest China under conditions of climate change
.
Earth System Dynamics
6
(
1
),
83
107
.
Schmidli
,
J.
,
Frei
,
C.
&
Vidale
,
P. L.
2006
Downscaling from GC precipitation: a benchmark for dynamical and statistical downscaling methods
.
International Journal of Climatology
26
,
679
689
.
Shi
,
Y.
,
Shen
,
Y.
,
Kang
,
E.
,
Li
,
D.
,
Ding
,
Y.
,
Zhang
,
G.
&
Hu
,
R.
2007
Recent and future climate change in northwest China
.
Climatic Change
80
(
3–4
),
379
393
.
Shrestha
,
B.
,
Cochrane
,
T. A.
,
Caruso
,
B. S.
,
Arias
,
M. E.
&
Piman
,
T.
2016
Uncertainty in flow and sediment projections due to future climate scenarios for the 3S Rivers in the Mekong Basin
.
Journal of Hydrology
540
,
1088
1104
.
Singh
,
H. V.
,
Kalin
,
L.
,
Morrison
,
A.
,
Srivastava
,
P.
,
Lockaby
,
G.
&
Pan
,
S.
2015
Post-validation of SWAT model in a coastal watershed for predicting land use/cover change impacts
.
Hydrology Research
46
,
837
853
.
Sorg
,
A.
,
Bolch
,
T.
,
Stoffel
,
M.
,
Solomina
,
O.
&
Beniston
,
M.
2012
Climate change impacts on glaciers and runoff in Tien Shan (Central Asia)
.
Nature Climate Change
2
(
10
),
725
731
.
Tamm
,
O.
,
Luhamaa
,
A.
&
Tamm
,
T.
2016
Modeling future changes in the North-Estonian hydropower production by using SWAT
.
Hydrology Research
47
,
835
846
.
Terink
,
W.
,
Hurkmans
,
R.
,
Torfs
,
P.
&
Uijlenhoet
,
R.
2010
Evaluation of a bias correction method applied to downscaled precipitation and temperature reanalysis data for the Rhine basin
.
Hydrology & Earth System Sciences
14
,
687
703
.
Van Vuuren
,
D. P.
,
Edmonds
,
J.
,
Kainuma
,
M.
,
Riahi
,
K.
,
Thomson
,
A.
,
Hibbard
,
K.
,
Hurtt
,
G. C.
,
Kram
,
T.
,
Krey
,
V.
&
Lamarque
,
J.-F.
2011
The representative concentration pathways: an overview
.
Climatic Change
109
,
5
31
.
Vidal
,
J. P.
,
Hingray
,
B.
,
Magand
,
C.
,
Sauquet
,
E.
&
Ducharne
,
A.
2016
Hierarchy of climate and hydrological uncertainties in transient low-flow projections
.
Hydrology and Earth System Sciences
20
,
3651
3672
.
Xu
,
C.
&
Xu
,
Y.
2012
The projection of temperature and precipitation over China under RCP scenarios using a CMIP5 multi-model ensemble
.
Atmospheric and Oceanic Science Letters
5
(
6
),
527
533
.
Xu
,
C.
,
Zhao
,
J.
,
Deng
,
H.
,
Fang
,
G.
,
Tan
,
J.
,
He
,
D.
,
Chen
,
Y.
,
Chen
,
Y.
&
Fu
,
A.
2016
Scenario-based runoff prediction for the Kaidu River basin of the Tianshan Mountains, Northwest China
.
Environmental Earth Sciences
75
,
1126
.
Yip
,
S.
,
Ferro
,
C. A.
,
Stephenson
,
D. B.
&
Hawkins
,
E.
2011
A simple, coherent framework for partitioning uncertainty in climate predictions
.
Journal of Climate
24
,
4634
4643
.
Zhang
,
Y.
,
Su
,
F.
,
Hao
,
Z.
,
Xu
,
C.
,
Yu
,
Z.
,
Wang
,
L.
&
Tong
,
K.
2015
Impact of projected climate change on the hydrology in the headwaters of the Yellow River basin
.
Hydrological Processes
29
,
4379
4397
.
Zhang
,
Q.
,
Gu
,
X. H.
,
Singh
,
V. P.
,
Sun
,
P.
,
Chen
,
X. H.
&
Kong
,
D. D.
2016
Magnitude, frequency and timing of floods in the Tarim River basin, China: changes, causes and implications
.
Global and Planetary Change
139
,
44
55
.