Projected Climate Change Impacts on Future Streamflow of the Yarlung 1 Tsangpo-Brahmaputra River 2

Abstract The Yarlung Tsangpo-Brahmaputra River (YBR) originating from the Tibetan Plateau (TP), is an important water source for many domestic and agricultural practices in countries including China, India, Bhutan and Bangladesh. To date, only a few studies have investigated the impacts of climate change on water resources in this river basin with dispersed results. In this study, we provide a comprehensive and updated assessment of the impacts of climate change on YBR streamflow by integrating a physically based hydrological model, regional climate integrations from CORDEX (Coordinated Regional Climate Downscaling Experiment), different bias correction methods, and Bayesian model averaging method. We find that (i) bias correction is able to reduce systematic biases in regional climate integrations and thus benefits hydrological projections over YBR Basin; (ii) Bayesian model averaging, which optimally combines individual hydrological simulations obtained from different bias correction methods, tends to provide hydrological time series superior over individual ones. We show that by the year 2035, the annual mean streamflow is projected to change respectively by 6.8%, −0.4%, and − 4.1% under RCP4.5 relative to the historical period (1980–2001) at the Bahadurabad in Bangladesh, the upper Brahmaputra outlet, and Nuxia in China. Under RCP8.5, these percentage changes will substantially increase to 12.9%, 13.1%, and 19.9%. Therefore, the change rate of streamflow shows strong spatial variability along the YBR from downstream to upstream. The increasing rate of streamflow shows an augmented trend from downstream to upstream under RCP8.5 compared to an attenuated pattern under RCP4.5.


Introduction
Water is a standout necessity amongst the most basic factors in human sustenance (Barnett et al., 2005).Global climate change has been found to intensify the global hydrological cycle, likely creating predominant impacts on regional water resources (Arnell, 1999;Gain et al., 2011).
Evaluation of the potential impacts of anthropogenic climate change on regional and local water resources relies largely on climate model projections (Li et al., 2014).The spatial resolution of typical global climate models (GCMs) (100-300 km) is insufficient to simulate regional events that are needed to capture different climate and weather phenomena at regional to local scales (e.g., the watershed scale) (Olsson et al., 2015).Climate simulations from GCMs can be dynamically downscaled with regional climate models (RCMs) to scales of 25-50 km.Despite that dynamical downscaling is computationally very demanding and that its accuracy depends to a large extend on that of its parent GCM, dynamical downscaling can provide more detailed information on finer temporal and spatial scales than GCMs (Hewitson and Crane, 1996).Such information is valuable for impact projections at regional to local scales that are more relevant to water resources management.
On the other hand, although the increased horizontal resolution can improve the simulation of regional and local climate features, RCMs still produce biases in the time series of climatic variables (Christensen et al., 2008;Rauscher et al., 2010).Bias correction is typically applied to the output of climate models.Most bias correction methods correct variables separately, with interactions among variables typically not considered (Christensen et al., 2008;Hessami et al., 2008;Ines and Hansen, 2006;Johnson and Sharma, 2012;Li et al., 2010;Piani et al., 2009;Piani et al., 2010).Separate-variable bias correction methods, for example, may result in physically unrealistic corrections (Thrasher et al., 2012) and do not correct errors in multivariate relationships (Dosio and Paruolo, 2011).Correspondingly, Li et al. (2014) introduced a joint bias correction (JBC) method and applied it to precipitation (P) and temperature (T) fields from the fifth phase of the Climate Model Intercomparison Project (CMIP5) model ensemble.
The Yarlung Tsangpo-Brahmaputra River (YBR) is an important river system originating from the Tibetan Plateau (TP), characterized by a dynamic fluvial regime with exceptional physiographic setting spread along the eastern Himalayan region (Goswami, 1985).Critical hydrological processes like snow and glacial melt are more important in this area compared to others.Hydrological processes of the YBR Basin are highly sensitive to changes in temperature and precipitation, which subsequently affect the melting characteristics of snowy and glaciered areas and thus affect streamflow.The YBR Basin is also one of the most under-investigated and underdeveloped basins around the world, with only few studies examined the impacts of climate change on the hydrology and water resources of this basin (Immerzeel et al., 2010;Lutz et al., 2014;Masood et al., 2015).Immerzeel et al. (2010) developed a snowmelt-runoff model in the upper YBR Basin using native output (without bias correction) from 5 GCMs under the A1B scenarios for 2046-2065 and found that its streamflow would decrease by 19.6% relative to 2000-2007.Subsequently, Lutz et al. (2014) implemented the SPHY (Spatial Processes in Hydrology) hydrological model in the upper YBR Basin using native simulations from 4 GCMs under RCP4.5 and RCP8.5 emissions scenarios for 2041-2050 and showed that the streamflow would increase by 4.5% and 5.2% relative to 1998-2007 under the examined two emissions scenarios.Masood et al. (2015) applied the H08 Hydrological model to the YBR Basin using bias corrected projections of 5 GCMs for near future   periods and found that relative to the period 1980-2001, the streamflow would increase by 6.7% and 16.2% for near and far future under RCP8.5, respectively.
Several factors could contribute to the discrepancy between these projections, such as the lack of high quality streamflow observations for hydrological model calibration, the choice of bias correction methods, simulations from global climate models, and future emissions scenarios, and a combination thereof.On the other hand, all the existing studies in the YBR basin rely on GCMs, which, as was discussed, cannot capture fine-scale climate and weather details that are required for a reliable regional impacts assessment.In the present study, we attempt to fill this gap by taking advantage of the recently compiled multi-model and multi-member high-resolution regional climate integrations from CORDEX (Coordinated Regional Climate Downscaling Experiment).We use different bias correction methods to alleviate the inherent biases in these regional climate integrations, and use a Bayesian model averaging technique to best combine different streamflow simulations obtained with different bias-corrected meteorological forcing data (e.g., precipitation and temperature).We synthesize our results and those in the existing studies with a hope to obtain a more comprehensive picture of changes in water resources in the YBR Basin in response to global climate warming.
We structured the paper into the following sections.Section 1 formulates the objectives of this study.Section 2 briefly introduces the YBR Basin, followed by the used materials and methods.
Our results and those in existing studies are compared in Section 3. Main conclusions along with a brief discussion of the future scope of this study are presented in Section 4. Tibetan Plateau (TP) is often referred as Asia's water tower, bordered by India and Pakistan in the west side and Bhutan and Nepal on the southern side, with a mean elevation of about 4000 m above sea level (Tong et al., 2014).The YBR is one of the largest rivers originating from the TP in Southwest China at an elevation of about 3100 m above sea level (Goswami, 1985;Xu et al., 2017).The total length of the river is about 2900 km (Masood et al., 2015), with a drainage area of the basin estimated to be around 530,000 km 2 .The YBR travels through China, Bhutan, and India before emptying into the Bay of Bengal in Bangladesh (Figure 1).The mean annual discharge is approximately 20,000m 3 /s (Immerzeel, 2008).The climate of the basin is monsoon-driven with an obvious wet season from June to September, which accounts for 60-70% of the annual rainfall.

Forcing data sets
Due to the lack of adequate in-situ meteorological observations, the WATCH forcing data (WFD) (Weedon et al., 2014) were used as a reference for bias correction and hydrological model calibration (Table 1).This dataset provided a good representation of real meteorological events and climate trends (Weedon et al., 2011).In this study, we used daily rainfall, temperature and potential evapotranspiration (PET) data from 1980 to 2001.
The sources of other required non-meteorological variables for implementing the hydrological model are as follows.The 90-m resolution digital elevation model data were acquired from the Shuttle Radar Topography Mission (SRTM) (http://srtm.csi.cgiar.org/).The Leaf Area Index (LAI) and snow cover data from 2000 to 2016 were downloaded from the National Aeronautics and Space Administration (NASA) (https://reverb.echo.nasa.gov/reverb/).For the periods during which LAI and snow data did not cover, average values of LAI and snow were used as model (http://www.glcf.umd.edu/data/gimms/).The soil hydraulic parameters were derived from the soil classification data which were extracted from the global digital soil map with a spatial resolution of 10 km (http://www.fao.org/geonetwork/).

Hydrological data
The streamflow observations during 1980-2001 for hydrological model calibration were obtained at two hydrological stations, i.e., the Nuxia station located in upstream China (Gao et al. (2008)) and the Bahadurabad station located in downstream Bangladesh; see Figure 1 for their geographical locations.

RCM data
The simulations of daily precipitation and temperature during the historical period of 1980-2001 and the projections under two examined emissions scenarios (RCP4.5 and RCP8.5) during the future period of 2020-2035 from the CORDEX experiment for the East Asia domain (which covers the whole YBR Basin) were downloaded from http://www.cordex.org/.The CORDEX program, which was coordinated by the World Climate Research Program, provides a unique opportunity for generating high-resolution regional climate projections and for assessing the impacts of future climate change at regional scales (Piani et al., 2009).As shown in Table 1, climate data from 5 CORDEX models were chosen.These models include HadGEM3-RA (denoted by RCM1), RegCM4 (RCM2), SNU-MM5 (RCM3), SNU-WRF (RCM4) and YSU-RSM (RCM5).To keep consistent with the WATCH forcing data, climate model integrations were interpolated to the grid of the WFD using the bilinear interpolation method.The adopted hydrological model, as will be introduced later, also requires PET as a forcing variable.We used the method proposed by Leander andBuishand (2007) andS. C. van Pelt (2009) to calculate PET with daily temperature T as follows: where T 0 is the observed mean temperature (•C) and PET 0 is the observed mean PET0 (mm/day) during the historical period.Daily PET0 data were acquired directly from the WFD dataset and were used to compute PET 0 .The proportionality constant α0 was determined for each calendar month by regressing the observed PET at each grid cell onto the observed daily temperature.

Hydrological model: THREW
We adopted the Tsinghua Representative Elementary Watershed (THREW) model (Tian, 2006;Tian et al., 2006) to simulate streamflow of the YBR Basin.The model consists of a set of balance equations for mass, momentum, energy and entropy, including associated constitutive relationships for various exchange fluxes, at the scale of a well-defined spatial domain.Details of the model can be found in Tian et al. (2006).The THREW model has been successfully applied to quite a few watersheds across China and United States (Li et al., 2012;Mou et al., 2008;Sun et al., 2014;Tian et al., 2006;Tian et al., 2012;Xu et al., 2015;Yang et al., 2014).For the simulation of snow and glacier melting processes which is important for the YBR Basin, we modify the original THREW model by incorporating the temperature-index method introduced in Hock ( 2003).The index-temperature method has been shown to exhibit an overall good performance in mountain areas in China (He et al., 2015).

Bias correction methods
Quantile mapping (QM) with reference observations has been routinely applied to correct biases in RCM simulations (Maraun, 2013).Using WFD as reference observations and following the principle of QM, first we estimated cumulative distribution functions (CDFs) for the observed and native RCM-simulated time series of daily precipitation or temperature during the historical/calibration period (which is 1980-2001 in this study); then for a given RCM-simulated data value from an application period (which may be historical 1980-2001 period or future 2020-2035 period), we evaluated the CDF of the native RCM simulations at the given data value, followed by evaluation of the inverse of the CDF of the observations at the thus obtained CDF value; the resulting value is the bias-corrected simulation (see Figure 2 for an schematic illustration of this procedure).
Independent bias correction for multiple meteorological variables can produce non-physical corrections.To alleviate the deficits of independent bias correction, Li et al. (2014) introduced a joint bias correction (JBC) method, which takes the interactions between precipitation and temperature into account.This approach is based on a general bivariate distribution of P-T and essentially can be seen as a bivariate extension of the commonly used univariate QM method.
Depending on the sequence of correction, there are two versions of JBC including JBCp, which corrects precipitation first and then temperature, and JBCt, which corrects temperature first and then precipitation.For more details of the QM and JBC methods, readers can refer to Wlicke et al. (2013) and Li et al. (2014), respectively.

Bayesian model averaging method
Bayesian model averaging (BMA) is a statistical technique designed to infer a prediction by weighted averaging predictions from different models/simulations.We refer readers to Dong et hydrology with meaningful results (Bhat et al., 2011;Duan et al., 2007;Wang and Robertson, 2011;Yang et al., 2011).

Bias correction of meteorological variables during the historical period
We applied the three bias correction methods (i.e., QM, JBCp and JBCt) to the CORDEX simulations of daily precipitation and temperature.We found that without bias correction, the native RCM1 and RCM2 simulations (see Table 1 for the full names of different RCMs) overestimate precipitation for all months during the 1980-2001 baseline period (Figure 3a-3b), while native simulations by the other models tend to overestimate precipitation of the dry-season (November to May of next year) and underestimate precipitation of other months.After bias correction, the above mentioned overestimation and underestimation reduces considerably.For temperature, we found that all the examined climate models consistently exhibit cold biases across all the months, and that such biases are largely eliminated after applying bias correction (Figure 4).In general, the three bias correction methods exhibit similar skills in reducing temperature biases (Table 2), with JBCt and QM showing somewhat better performance than JBCp.As expected, PET calculated from bias-corrected temperature simulations was quite close to WFD observations.
In summary, we found that almost all the bias correction methods can substantially reduce biases for all the three variables across the months, though with sizeable variations between bias correction methods and across variables and seasons, consistent with existing studies on the comparison of different bias correction methods (Maraun, 2013;Prasanna, 2016).

Hydrological model setup and simulation
To setup the THREW model, the whole basin was discretized into 237 representative elementary watersheds (REWs).There are in total 16 parameters involved in THREW, as listed in Table 3.
The first 6 parameters were determined for each REW a prior from the data described in the section 'Materials and methodology'.The remaining parameters were subjected to calibration and assumed to be uniform across the 237 REWs.Automatic calibration was implemented by the -NSGAII optimization algorithm developed by Reed et al. (2003).We chose the commonly used Nash Sutcliffe efficiency coefficient (NSE) (Nash and Sutcliffe, 1970) as the single objective function for model calibration.
We divided the whole period 1980-2001 into two sub-periods, which were used respectively for model calibration (1980-1990) and validation (1991-2001).Simulated daily streamflow time series at Bahadurabad were compared against the corresponding observations to compute the NSE objective function.To warm up the model, we dropped the first year of the calibration period (i.e., 1980).Observed and simulated daily streamflow of remaining years were used to compute NSE as follows: where N denotes the total number of days in the calibration period (which is 1981-1990  model, NSE for the 1991-2001 validation period can be likewise computed so as to assess the calibrated model performance in simulating streamflow that is not seen in the calibration period.During the calibration period the daily and monthly NSE values are 0.84 and 0.92, respectively, and during the validation period the daily and monthly NSE values are 0.78 and 0.84, respectively.We also compared the observed and simulated monthly discharges at the Nuxia station, which is not involved in model calibration.The monthly NSE values of calibration and validation periods were 0.66 and 0.73, respectively.In summary, these results suggest that the THREW model does a good job in simulating the hydrological processes in the YBR Basin during this historical period.We assume that the calibrated THREW model is applicable to the future period.This assumption is necessary in this study and has been widely adopted in previous climate impacts studies. Figure 6 compares the seasonal streamflow simulated by the THREW model with observed streamflow data at Bahadurabad.It is observed that the streamflow generated by native RCM simulations tends to either over-or underestimate the observations, and that all the adopted bias correction methods can alleviate, to varying degrees, these biases.We found that in general bias correction is more effective in improving the simulation of dry season streamflow (from November to April in the next year) than that of wet season (May to October).Table 4 shows the annual mean observed streamflow at Bahadurabad as well as the simulated streamflow with the WFD data and with the native and bias-corrected RCM integrations.We can see that at annual scale, streamflow simulated with native RCMs is on average higher (e.g., RCM1, RCM2) or and simulations.These indices are relative error (RE) and root mean squared error (RMSE), both evaluated at daily scale, as defined in the following: where N denotes the total number of days during the considered period;    and    represent respectively the observed and simulated streamflow of time n.As seen from Table 6, based on the above indices, after applying BMA we obtain considerably better results than almost all those generated by different bias-corrected climate simulations from different climate models with different bias correction methods.Figure 7 shows the mean prediction (red line) and 90% uncertainty interval of BMA during the historical period at Bahadurabad.The uncertainty interval of BMA can cover almost all observations, which further indicated the sound performance of BMA.consistently decrease in all the studied RCM models.Therefore, the general pattern of "wet getting wetter, dry getting drier" (Chou et al., 2013) associate with climate change exists in YBR as well.Also, as expected, precipitation under RCP8.5 is on average higher than that under We found that temperature is projected to increase by all RCM simulations in both dry seasons and wet seasons (Figure 9).It is surprising to see that there is no significant difference in temperature between RCP8.5 and RCP4.5 scenarios except for RCM3 and RCM4.In fact, this is not inconsistent with the IPCC AR5 ( 2013), which shows that the projected future global mean temperature does not significantly diverge under different RCP scenarios until 2030 (our future period is 2020-2035).Similar to precipitation, there are obvious variations in the projected changes among different climate models and different bias correction methods.Using BMA weight coefficient calculated in Section 3.2, weighted temperature in historical period, RCP4.5 and RCP8.5 is 8.7, 9.8 and 10.0℃, respectively.

Projections of future streamflow and comparison with previous studies
Figure 10 shows the mean prediction and 90% uncertainty interval of streamflow simulated by BMA method during (a) RCP4.5,(b) RCP8.5 scenarios at Bahadurabad.Uncertainty interval of RCP4.5 is similar with that of RCP8.5.All the following discussions in this subsection is based on BMA weighted streamflow.
For the sake of comparison between Immerzeel et al. (2010), Lutz et al. (2014), Masood et al. (2015) and our results, we also examined an upstream outlet location (the red dot in Figure 1), which was studied in the referred studies.To be noted, the observed streamflow data at this upstream outlet are unavailable.
Table 7 shows a summary of the referred existing studies about climate impact on future streamflow in the YBR Basin.and far future (2075-2099) and also applied bias correction method.The streamflow increased by 6.7% and 16.2% in the near future and far future, respectively, when compared with the observed data .
The comparisons among the streamflow projection of YBR during different periods in different studies are shown in Figure 11.In our study, the projected streamflow is 1466 mm/a during 2020-2035 under RCP8.5 at Bahadurabad, which is substantially higher than the findings of Masood et al. (2015) at the same location, which is 1244 mm per year during 2015-2039 under RCP8.5.The projected streamflow is 692 mm per year during 2020-2035 under RCP8.5 at the upper YBR outlet.This result is quite close to the findings of Lutz et al. (2014), which is 727 mm per year during 2041-2050 under RCP8.5.To be noted, our study adopted RCMs integrations, BMA method by incorporating different bias correction methods, and a physically based hydrological model accounting for snow and glacier melting processes, which could explain the differences from the existing studies.
Table 8 shows the relative changes of projected runoff and its driving factors under different emission scenarios compared to the historical period at different locations of the YBR.At the basin-wide scale represented by Bahadurabad station, future streamflow shows an evidently increasing trend under both RCP4.5 and RCP8.5 scenarios.The increasing rate under RCP8.5 (12.9%) is not-surprisingly higher than RCP4.5 (6.8%).Also, the trends of streamflow exhibit strong spatial variability along the YBR.Under RCP4.5, upstream locations are more likely to experience an increasing trend at a much less rate.For example, the change rate of streamflow is projected to decrease at 0.4% and 4.1% at the YBR outlet and Nuxia, respectively.Under RCP8.5, however, upstream locations would more likely witness an augmented increasing rate of streamflow change, e.g., 13.1% and 19.9% at the YBR outlet and Nuxia, respectively.

Conclusions
In this study, we conducted a comprehensive evaluation of future streamflow in the YBR Basin.
We adopted RCMs integrations, BMA method by incorporating different bias correction methods, and a physically based hydrological model accounting for snow and glacier melting processes to implement the evaluation.The major findings are summarized as follows.
(1) The three bias correction methods implemented in this study can all substantially reduce biases in the three variables (precipitation, temperature and potential evapotranspiration).
Specifically for precipitation, when native RCMs show overestimations, all bias correction methods perform reasonably well.While, none of them can provide satisfying corrections when native RCMs exhibit strong underestimations.This finding is consistent with existing studies (Maraun, 2013;Prasanna, 2016) and requires further in-deep studies.For temperature and potential evapotranspiration, all of the three bias correction methods performed well, especially QM and JBCt.
(2) The basin-wide discharge is projected to increase substantially during the future period (2020-2035) under the two examined emissions scenarios of RCP4.5 and RCP8.5.The projected annual mean streamflow at Bahadurabad is 1386.7 mm per year under RCP4.5 with an increasing rate of 6.9%, and the number becomes higher as 1466.4mm per year under RCP8.5 with an increasing rate of 12.9%.Increasing mean annual streamflow indicates more flood events that would occur in this already vulnerable region, which calls for more close collaborations among upstream and downstream riparian countries.
(3) Projected streamflow exhibits different spatial patterns under different scenarios in the YBR basin.Under RCP4.5, the annual mean streamflow is projected to change by 6.8%, -0.4%, and -4.1% in the future period (2020-2035) compared to the historical period  at three locations from downstream to upstream along the YBR, i.e., Bahadurabad, the upper YBR outlet, and Nuxia.Therefore, the increasing rate of streamflow exhibits an attenuated trend from downstream to upstream.Under RCP8.5, however, the increasing rate of streamflow (12.9%, 13.1%, and 19.9% at the three locations) exhibits an augmented trend from downstream to upstream.The different trends are likely associated with varying spatial patterns of projected future precipitation, but more detailed investigations are needed.

List of Tables
, where

Figure 5
Figure 5 shows the observed (black line) and simulated (red line) discharges at Bahadurabad at (a) daily, (b) monthly, and (c-d) seasonal time scales for both the calibration and validation periods.It can be seen that the THREW model performs well in the YBR Basin at all time scales.
Hydrol.Earth Syst.Sci.Discuss., https://doi.org/10.5194/hess-2018-251Manuscript under review for journal Hydrol.Earth Syst.Sci. Discussion started: 14 May 2018 c Author(s) 2018.CC BY 4.0 License.lower(e.g., RCM3, RCM4 and RCM5) than the observations; while streamflow simulated with bias-corrected RCMs is much more consistent with the observations.Table5presents the NSE values for the daily and monthly streamflow over the calibration and validation periods simulated by the THREW model with the WFD data and with native and bias-corrected RCM simulations at Bahadurabad.We found that QM and JBCp can improve NSE for almost all the RCMs except RCM5, while JBCt can improve NSE for three of the five climate models (RCM1, RCM3, and RCM4).We also found that none of the 3 bias correction methods is compelling better than others, suggesting the necessity of combining different streamflow simulations generated with different bias-corrected climate simulations.Moreover, it is seen that most of the NSEs values are higher than 0.55 with a few exceptions, indicating reasonably well simulations of daily and monthly streamflow for both calibration and validation periods on average across the entire basin, and thus enhancing our confidence in applying the calibrated THREW model and the bias-corrected CORDEX simulations to projecting future hydrological conditions in the YBR Basin.Given the fact that none of the bias correction methods and none of the RCM models are compellingly superior over others, as we have found, we therefore integrate streamflow simulations generated by different bias-corrected climate simulations from different climate models with different bias correction methods in terms of BMA.Our attempt is to take advantages of individual streamflow simulations.Daily streamflow simulations and observations during the THREW model calibration period(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990) were used to calibrate the BMA weights, and those during the validation period are used to evaluate the calibrated BMA weights.In addition to NSE, two other indices were used to measure the closeness between observations Hydrol.Earth Syst.Sci.Discuss., https://doi.org/10.5194/hess-2018-251Manuscript under review for journal Hydrol.Earth Syst.Sci. Discussion started: 14 May 2018 c Author(s) 2018.CC BY 4.0 License.

Figures 8 -
Figures 8-9 show changes in seasonal precipitation and temperature during the near future period 2020-2035 relative to the historical 1980-2001 period based on bias-corrected RCM simulations under RCP4.5 and RCP8.5 emissions scenarios.It is found that precipitation in wet seasons will increase under both emissions scenarios and in all bias-corrected RCM simulations with one exception of RCM3 under RCP4.5.In contrast, precipitation in dry seasons is projected to Hydrol.Earth Syst.Sci.Discuss., https://doi.org/10.5194/hess-2018-251Manuscript under review for journal Hydrol.Earth Syst.Sci. Discussion started: 14 May 2018 c Author(s) 2018.CC BY 4.0 License.RCP4.5, especially for RCM3 and RCM4 in the wet season.We also found obvious variations in the projected changes among climate models and bias correction methods.This suggests the importance of exploring multi-models and multi-methods to obtain a more comprehensive picture about the uncertainty of the impacts of climate change on local hydrology.Using BMA weight coefficient calculated in Section 3.2, weighted precipitation in historical period, RCP4.5 and RCP8.5 is 1425.3,1529.8 and 1608.0 mm per year, respectively.

,
u-zone, u s is the saturation degree of u-zone, u y is the soil depth of u-zone, d is the diffusion index ( 1 When S is the topographic slope, y s is the depth of s-zone, Z is the total soil depth 1Syst.Sci.Discuss., https://doi.org/10.5194/hess-2018-251Manuscript under review for journal Hydrol.Earth Syst.Sci. Discussion started: 14 May 2018 c Author(s) 2018.CC BY 4.0 License.

Table 6 .
Evaluation merits of streamflow simulations for individual RCM and BMA scenarios.30

Table 7 .
Summary of existing studies on projected streamflow under climate change in the YBR

Table 3 .
Principal parameters of THREW model.