1 Introduction

Hydrological cycle is a linkage between available water resources and agriculture, ecosystems and social activities. There are significant impacts of climate change on hydrological processes, and then on water resources, agriculture, and ecosystems as well as human societies at different spatial and temporal scales (Brown and Funk 2008; Ju et al. 2013; Piao et al. 2010; Sivakumar 2011). Therefore, it is essential to investigate the responses of hydrological cycle for better understanding the potential impacts of climate change on earth system and human societies. It is a frontier area to quantitatively estimate the magnitude and scope of impacts of climate change on hydrological processes.

The Tibetan Plateau (TP) is the highest plateau in the world with an average elevation over 4000 m above the see level. It is called the “Water Tower” of Asia and the “Third Pole” of the Earth (Immerzeel et al. 2010b; Qiu 2008). In its vast area of about 2.3 million km2 (almost one fourth of China’s area), the TP has sourced several most important river systems in Asia, including Yellow, Yangtze, Mekong, Yarlung Zangbo-Brahmaputra, Ganges, Indus, Salween, Tarim, etc., at the same time provided water resources for more than 1.4 billion people (over 20 % of the world population, Immerzeel et al. 2010b). Duo to the complex terrain, distinctive landscape, and high altitude, the TP exerts great impacts on the formulation of Asian monsoon, even the global climate. Meanwhile, it is also one of the most sensitive areas to the global climate change (Bolch et al. 2012; Li et al. 2013; Liu et al. 2009; Sun et al. 2013; Wang et al. 2013; Wu et al. 2006; Xu et al. 2009; You et al. 2007). During the last several decades, climate in the TP has experienced dramatic spatial and temporal changes and variations, with average temperature increasing at a rate of 0.447 °C per decade from 1961 to 2001, and most of meteorological stations have recorded rising trends precipitation at the same period (Xu et al. 2008). The reference evapotranspiration was reported to decrease at a rate of 0.691 mm a−1 from 1970 to 2009 (Xie and Zhu 2013). This change has significantly accelerated the hydrological dynamics and altered the land surface processes in terms of glacier retreat, permafrost degradation, and runoff variation (Barnett et al. 2005; Guo et al. 2012; Ji and Kang 2013; Kang et al. 2010; West 2008; Yao et al. 2010).

The Yarlung Zangbo River (YZR) is one of the most important international rivers in the TP. The land-use in the YZR basin did not change much during the last decades (Li et al. 2012). The climate change, however, is significant (You et al. 2007). Therefore, it offers an ideal site to investigate the climate change impacts without intensive intervention from human beings. In this study, the Lhasa River (LR) basin was selected as the case study area, which is the largest and most important tributary of the YZR (Guan et al. 1984). The LR basin is the political, economic and cultural center of Tibet, rendering its irreplaceable role in Tibet. But, the hydrological regimes and the climate change impacts in this region are still not fully reported (Li et al. 2014; Prasch et al. 2011; Qiu et al. 2014), making it difficult to predict future available water resources. Therefore, it is significant to simulate the hydrological cycle based on distributed hydrological model and explore the plausible impacts of climate change for better water resources management and sustainable social development in Tibet.

In this study, the Variable Infiltration Capacity (VIC) model was coupled with the Shuffled Complex Evolution developed at the University of Arizona (SCE-UA) to simulate the hydrological processes at a resolution of 10 × 10 km. The future climate change conditions were formed by downscaling from two selected General Circulation Models (GCMs) under the A1B, A2, B1 scenarios of Special Report of Emission Scenarios (SRES) for the period of 2046–2065. Here, we mainly focused on five major variables: evaporation, soil moisture, spring snow cover, runoff and streamflow, which are also the primary water resources for agriculture, ecosystem, and domestic and industrial uses. By simulating the watershed hydrological processes and estimating the range of precipitation and temperature trends, the objective of this study is to analyze the impacts of climate change on those key hydrological elements, and further to analyze the implication of change on agriculture, ecosystem and human society in the middle of twenty-first century. The simulation results are expected to be served as reference for climate change mitigation and adaptation, and water resources management in this vulnerable region.

2 Simulation framework

The overall simulation framework is presented in Fig. 1. The Automatic Statistical Downscaling (ASD) model is used to downscale the GCMs outputs. Land-use, soil, and climate data are inputted into the VIC model to simulate hydrological processes. The SCE-UA will calibrate the model parameters automatically based on an objective function (OBJ).

Fig. 1
figure 1

Flowchart of the simulation framework

Full size image

2.1 Methodology description

2.1.1 The ASD model

The ASD model was developed by Hessami et al. (2008) under the Matlab environment. It is a station-based statistical downscaling technique, which builds the relationship between observed climate variables and climate predictors at each station. In the ASD model, predictor selection is based on backward stepwise regression and partial correlation coefficients. The simulation of precipitation is conditional, in which two steps are involved: precipitation occurrence (PO) and precipitation amount (PA):

$$ PO_{j} = \alpha_{0j} + \sum \alpha_{ij} p_{ij} $$
(1)
$$ PA_{j}^{0.25} = \beta_{0j} + \sum \beta_{ij} p_{ij} + e_{j} $$
(2)

where PO j is the precipitation occurrence at station j; PA j is the precipitation amount at station j; p ij is the selected predictor at station j; α ij and β ij are model parameters for predictor p ij ; and e j is model error.

The simulation of temperature (T) is unconditional. One step is needed:

$$ T_{j} = \gamma_{0j} + \sum \gamma_{ij} p_{ij} + e_{j} $$
(3)

where T j is the temperature at station T j and γ ij is model parameter for predictor p ij .

2.1.2 The VIC model

The VIC model is a large-scale semi-distributed land surface model (Liang et al. 1994; Liang and Xie 2001). It simulates the soil–vegetation–atmosphere interactive dynamics with the function of climate, vegetation, and soil properties based on experimental algorithms. A wide range of physical processes are included in the model: precipitation interception, evaporation, infiltration, snow accumulation and melt, and runoff generation, etc. For runoff generation, the VIC model is able to consider two different runoff mechanisms: saturation excess runoff and infiltration excess runoff (Xie et al. 2003). It employs a spatially varying infiltration capacity strategy to represent sub-grid scale heterogeneity of soil properties used in Xinanjiang model (Zhao 1992), and can consider the spatial variability of precipitation and land-use (Liang et al. 1996a). More details on the VIC model can be found in Liang et al. (1994) and Liang and Xie (2001). Additionally, the land surface processes are modeled at each grid separately, making it easy to integrate with the GCMs output (Liang et al. 1996b). Since its development, the VIC model has been applied widely to simulate the hydrological processes and investigate the impacts of climate change at the regional or global scales (Hostetler et al. 2000; Liu et al. 2009; Sridhar et al. 2012).

2.1.3 Calibration and assessment criteria

Generally, seven parameters (see Fig. 1) related to soil characteristics in VIC are sensitive to local study area which should be calibrated carefully. However, no automatic calibration algorithms are integrated with the VIC model. In this study, a Matlab-based program was developed to search the optimal parameters by combining the VIC executable file and SCE-UA algorithm. The SCE-UA algorithm was first developed by Duan et al. (1993), and then widely used for parameters optimization (Duan et al. 1994). The optimal parameters were obtained by minimizing OBJ (Viney et al. 2009), which was calculated from simulated daily streamflow against observation. The OBJ is expressed as:

$$ {\text{OBJ}} = \left( {1 - E_{ns} } \right) + 5|\ln (1 + E_{r} )|^{2.5} $$
(4)
$$ E_{ns} = \frac{{\sum (Q_{i,o} - \bar{Q}_{o} )^{2} - \sum (Q_{i,s} - Q_{i,o} )^{2} }}{{\sum (Q_{i,o} - \bar{Q}_{o} )^{2} }} $$
(5)
$$ E_{r} = \frac{{\sum Q_{i,s} - \sum Q_{i,o} }}{{\sum Q_{i,o} }} \times 100\,\% $$
(6)

in which E ns is the Nash–Sutcliffe coefficient of efficiency; E r is the relative error; Q i,o and Q i,s are observed and simulated streamflow at the ith day; \( \bar{Q}_{o} \) is the average of observed streamflow. This objective function is more flexible since it considers both the Nash–Sutcliffe coefficient of efficiency and the relative error.

Coefficient of determination (R2) and root mean square deviation (RMSD) were used to assess the ASD model. R2 and RMSD are calculated as:

$$ R^{2} = \frac{{(\sum (Q_{i,o} - \bar{Q}_{o} )(Q_{i,s} - \bar{Q}_{s} ))^{2} }}{{\sum (Q_{i,o} - \bar{Q}_{o} )^{2} \sum (Q_{i,s} - \bar{Q}_{s} )^{2} }} $$
(7)
$$ {\text{RMSD}} = \sqrt {\frac{{\sum (C_{o} - C_{s} )^{2} }}{n}} $$
(8)

in which \( \bar{Q}_{s} \) is the average of simulated streamflow, C o and C s are the observed and simulated climate value, and n is the number of observations. At the same time, four criteria were used to evaluate the VIC model performance: E ns and E r , coefficient of correlation (r), and R 2.

2.2 Data description

2.2.1 Input data in the ASD model

The ERA-40 re-analysis of meteorological observation data (Uppala et al. 2005) were used to build the ASD model for selecting the most relevant predictors and deriving the regression parameters at each station. Table 1 lists the predictors used in this study. More than 25 GCMs are commonly used to investigate the climate change. However, the performance of each model varies at a given region due to the differences in physical processes, parameterization, and grid resolution. How to choose a reliable model for local study should be carefully addressed (Casado and Pastor 2011; Fu et al. 2013; Trenberth 1997). Our previous study suggested that five models: namely Echam5, Miroc3.2_Hires, Miroc3.2_Medres (Medres hereafter), Echam4 and GFDL:CM21 are relatively more suitable in the YZR basin (Liu et al. 2013). Considering the data availability and the completeness of climate scenarios, Echam5 and Medres were selected in this study. The same predictors as ERA-40 from Echam5 and Medres were used for downscaling. Three representative CO2 emission scenarios (A1B, A2 and B1 of SRES) were used to represent a wide range of emission levels, in which A2 predicts high rate of greenhouse gas (GHG) emissions of 850 parts per million by volume (ppmv) by 2100, while A1B as medium rate of 720 ppmv and B1 as low rate of 550 ppmv (IPCC 2000). The combination of two GCMs and three emission scenarios forms six scenarios: Echam5 A1B, Echam5 A2, Echam5, B1, Medres A1B, Medres A2 and Medres B1.

Table 1 Predictors used for developing the ASD model
Full size table

2.2.2 Input data in the VIC model

Inputs for the VIC model mainly include climate, soil and land-use data. Climate data (daily precipitation, daily maximum and minimum temperature) were obtained from six meteorological stations in or around the LR basin (Fig. 2), which were provided by China Meteorological Data Sharing Service System. This dataset covers a period of 1974–2000, with the first 2-years data being used for model warming-up. The station-based data were then interpolated to each grid by applying the Inverse Distance Weighted (IDW) method. In order to reduce the influence of complicated terrain of TP on the interpolation, a declination of 0.65° per 100 m was adopted to the interpolation of maximum and minimum temperature (Li et al. 2003). The soil parameters were derived from the global 5′ soil date offered by the NOAA hydrology office using the method described in Su and Xie (2003). Land-use data are 1 km global land cover provided by the University of Maryland. Both soil data and land cover data were resampled to the resolution of 10 × 10 km.

Fig. 2
figure 2

The study area and locations of the meteorological stations

Full size image

2.2.3 Streamflow data

Daily streamflow dataset in Lhasa hydrological station provided by Tibet Bureau of Hydrology and Water Resources was used for calibrating model parameters and validating model performance. This dataset covers a period of 1976–2000, in which data from 1976 to 1990 were used for model calibration, and the rest for model validation.

3 The VIC model performance

The model performance results are documented in Table 2. When the meteorological observed data were used as model inputs, the relative error is low (2 %) and the estimated flow process matches well with the observed process in terms of E ns , r and R 2, which are 0.94, 0.97 and 0.94, respectively. Additionally, the calibrated VIC model reproduced the observed flow satisfactorily when the independent climate data were used in the validation period. The E r falls in the range of 10 %, and the values for E ns , r and R 2 are 0.92, 0.96 and 0.93, respectively. Figure 3 presents the comparison between the monthly simulated and observed flow for both calibration and validation periods. Results show that the observed and simulated streamflow is comparable, indicating the suitability of the VIC model in representing the flow process in the LR basin.

Table 2 Performance of the VIC model based on observation and GCMs outputs
Full size table
Fig. 3
figure 3

Comparison between observed and simulated streamflow for the calibration and validation periods (Obs observed streamflow, Mete simulated streamflow based on meteorological climate data, Echam5 simulated streamflow based on Echam5 outputs, Mderes simulated streamflow based on Medres outputs)

Full size image

In the previous studies, the future climate data were directly inputted into hydrological models, which were generally calibrated with observed climate data, without justifying model performance by using the historical GCMs data (Liu et al. 2009). This implicitly assumes that the GCMs data have the comparable abilities in representing the hydrological processes as the measured meteorological data. The fact is, however, that the GCMs data do not always have such ability. Therefore, it is important to verify the historical GCMs outputs. In this study, the model performance was also investigated by using historical Echam5 and Medres outputs. Results indicate that the VIC model successfully represents the observed streamflow, with E r in ±10 %, E ns larger than 0.63, and r exceeding 0.80 for both GCMs (Table 2; Fig. 3). The good performance of VIC using historical GCMs data as inputs further implies the fitness of Echam5 and Medres and success of the downscaling framework in the study area. On the basis of the reasonability of the historical downscaled data, it is expected that the future climate series are also suitable for investigating the impacts of climate change in the LR basin.

Figure 4 spatially describes the averaged values of five factors for the period of 1981–2000. Precipitation shows a dramatic spatial variation with a decreasing trend from 700.5 mm in the upstream (northeast part) to 357.1 mm in the downstream (southwest part). Evaporation does not change much with a narrow range of 172.1–269.9 mm for the whole basin. Soil shows the similar spatial distribution with precipitation, varying from 56.3 to 306.0 mm. Spring snow cover is found at the upstream and some grids around the basin boundary. In contrast, grids in the middle- and down-stream do not show much spring snow cover. Runoff (including surface and subsurface runoff) shows a similar spatial pattern as soil moisture, but with a greater spatial variation of 505.1 mm. Furthermore, the evaporation ratio (ratio of evaporation to precipitation) and runoff ratio (ratio of runoff to precipitation) were also explored in the whole basin (Fig. 5). Evaporation ratio shows higher value of 0.72 in the downstream which implies more precipitation was used for evaporation in this region, but lower value of 0.11 in the upstream and around the boundary where more precipitation was used to generate runoff. Greater values are found in the high-elevation grids.

Fig. 4
figure 4

Average of hydrological variables represented by the VIC model for the period of 1981–2000 (mm)

Full size image
Fig. 5
figure 5

Simulated evaporation ratio and runoff ratio for the period of 1981–2000

Full size image

4 Climate downscaling

Table 3 documents the selected predictors and simulation results of the ASD model by using the ERA-40 re-analysis data at each station. The representation of temperature regimes is better than precipitation in terms of R 2, with the former greater than 0.90, and latter between 0.13 and 0.22. This is consistent with Hessami et al. (2008), where R 2 for precipitation is around the same range. Considering the low values of RMSD (smaller than 0.25 for precipitation and 0.015 for temperature), the performance of the ASD model in the LR basin is satisfactory. More details about the downscaling framework can be found in Liu et al. (2014).

Table 3 Simulation results of the ASD model at each station
Full size table

After the regression function was formulated, it was then applied to the GCMs climate scenarios to generate the station-based climate variables. Figure 6 presents the magnitudes of change for precipitation and average temperature in term of differences between the future scenarios and the historical records for the years of 1981–2000. Monthly precipitation shows dramatic change for all scenarios. In July and August, precipitation may increase significantly with the largest value of 60 % (Fig. 6a). However, from October to May the following year, precipitation will decrease largely, especially in November, February and March. At the annual scale, precipitation shows an increasing trend, varying from 5.89 to 20.09 % (Fig. 6; Table 4). A question arises as to how precipitation increases in the annual scale when most of months show declining trend. The reason seems to be that precipitation mainly occurs in summer (60–70 % of the annual amount), therefore, large increase in summer results in the increase of the annual precipitation. Additionally, the spatial change for annual precipitation is given in Fig. 7. The distribution presents a clear spatial variation. For Echam5, the change shows the similar spatial pattern with the historical precipitation, with a decreasing trend from the northeast part to the southwest part. Most of grids in the east region present an increase, but a decreasing trend is found in the west region. However, a distribution of increasing in north and decreasing in south is estimated by Medres.

Fig. 6
figure 6

Future climate change at the monthly and annual scales between 2046–2065 and 1981–2000

Full size image
Table 4 Scenarios of future climate for the period of 2046–2065
Full size table
Fig. 7
figure 7

Percentage change of annual precipitation between 2046–2065 and 1981–2000 (%)

Full size image

Daily average temperature will increase largely for all scenarios, especially between April and October (Fig. 6b). This change implies the warming tendency in the LR basin. Daily maximum temperature will experience the largest warming trend with average value greater than 2 °C, whereas the magnitude of change for daily minimum temperature is not significant with an average about 0.6 °C.

It is clear from the simulated results that uncertainties between different scenarios are significant. In order to obtain a reasonable and reliable range, it is necessary to include more GCMs under different scenarios to give a reliable range.

5 Impacts of climate change on hydrological components

5.1 Evaporation

Monthly evaporation shows dramatic temporal variation. A large drop of more than 25 % is found from November to March the following year, while July and August will witness dramatic increase over 40 % (Fig. 8a). However, no significant change of evaporation is found in other months. This kind of change pattern may be good for the agriculture due to the decrease of evaporation in the spring when crops need a large amount of water from the soil. At the annual scale, evaporation will increase with a median value of 29 % (Fig. 8a).

Fig. 8
figure 8

Projected change of hydrological factors at the monthly and annual scales between 2046–2065 and 1981–2000

Full size image

Spatially, the increasing trend of annual evaporation is more dominant than decreasing, especially for the scenarios of Echam5 A1B, Medres A1B and Medres A2 in the northeastern part of the basin, where the magnitudes of change exceed 50 % compared with the baseline (Fig. 9). Annual evaporation will change modestly in the middle- and lower-stream regions within the range of ±25 % for all scenarios.

Fig. 9
figure 9

Percentage change of annual evaporation between 2046–2065 and 198–2000 (%)

Full size image

5.2 Soil moisture

All months show a decreasing trend for soil moisture in the future except August (Fig. 8b). Although the change is minor, it may still post challenge to agriculture and ecosystem duo to the decreases during the months between April and July when crops and vegetation need more water from the soil. Spatially, soil moisture only presents slight change with the similar spatial pattern as precipitation (Fig. 10). Regarding Echam5, three scenarios give a small rising tendency with a range of below 5 % in the east part. An insignificant decrease appears in the west part. For Medres, slight increase is found in the northern part under three scenarios, whereas the decrease dominates the southern part.

Fig. 10
figure 10

Percentage change of annual soil moisture between 2046–2065 and 1981–2000 (%)

Full size image

5.3 Spring snow cover

Snowmelt contributes much water to the runoff for rivers in the TP (Immerzeel and Bierkens 2010a; Siderius et al. 2013; Zhang et al. 2013). Spring snowmelt plays an important role for the irrigation during the period of crop growth. The projection of spring snow cover in the LR basin shows a negative impacts since most of grids will experience dramatic decrease (Fig. 11). Spring snow cover may disappear in a large number of grids for all scenarios in the middle twenty-first century.

Fig. 11
figure 11

Percentage change of spring snow cover between 2046–2065 and 1981–2000 (%)

Full size image

5.4 Runoff

Runoff in months from September to February the following year will change with a slight decrease within 10 % (Fig. 8c). However, the magnitudes of decrease in spring are significant ranging from 20 to 40 %. In summer, most of scenarios predict an upward tendency except that by Medres in June. The annual runoff will increase with low uncertainty.

Spatially, runoff shows a pronounced variation in the LR basin for all scenarios with the similar pattern as precipitation (Fig. 12). The change of runoff shows an increasing trend in the eastern part of the watershed with the largest magnitude of more than 50 %. In contrast, significant decline can be observed in the western part, especially for the scenario Echam5 A1B. Regarding Medres, about 60 % grid cells located in the northern part show an upward trend, whereas the southern part gives a downward trend up to 50 %.

Fig. 12
figure 12

Percentage change of annual runoff between 2046–2065 and 1981–2000 (%)

Full size image

5.5 Streamflow

River channel flow is the combined result of precipitation, evaporation, soil moisture, and snowmelt, etc. Change in these factors will result in change of the flow regime. Figure 8d demonstrates a significant decrease of river flow in the Lhasa hydrological station from November to May the following year. In summer and autumn, significant increases will be expected, especially in August. Streamflow at the annual scale shows a slight increase with low uncertainty.

6 Discussion

6.1 Comparison with other studies

Several similar studies were carried out to investigate the impacts of climate change on hydrological processes in the headwater catchments of large river systems in the TP. Liu et al. (2009) projected the climate change impacts on evaporation and runoff in headwater catchment of the Tarim River basin by using the VIC model. A decreasing trend was detected for winter runoff, which may be resulted from the decreasing winter air temperature, as well as an increasing trend of spring runoff. Xu et al. (2009) applied the SWAT model to explore the impacts of climate change on streamflow in headwater catchment of the Yellow River. An overall decreasing trend of annual streamflow was found. The climate change impacts on hydrological processes in the upper reaches of the Yangtze River were also assessed by using EasyDHM with the climate scenarios provided by HadCM3 and Echam4 (Sun et al. 2013). It was found that the runoff may decline with the increasing temperature and decreasing annual rainfall. Compared with these studies, climate change may bring more negative impacts on hydrological processes in the LR basin than other river systems in the TP, with not only water scarcity but also floods. This may cause serious problems to the development of social-economics in the local region or even the whole Tibet.

6.2 Implication of this study

The simulation results imply that precipitation plays a central role in the formulation of hydrological processes in the LR basin from the temporal and spatial perspectives. This is consistent with the results reported by Zhang et al. (2013). They found that precipitation contributes 65.4 % of annual runoff in the Brahmaputra River basin. There is a good temporal agreement between change patterns of precipitation and evaporation as well as runoff. Also, the spatial distribution of soil moisture and runoff matches well with precipitation. Therefore, a reliable prediction of future precipitation is essential for the purpose of obtaining a better estimation of climate change impacts.

The decrease of runoff in spring (from March to May) is much greater than that from October to February the following year, although the decreases of precipitation in these two periods are similar. This may be caused by the large decrease of spring snow cover (Lin et al. 2008), which is used for supplementing the runoff in the baseline period. The reduction of spring runoff driven by disappearing snow cover may bring negative impacts on agriculture and ecosystem, since crops and plants need a large amount of soil water (green water) and irrigation water (blue water) for growth. Spatially, the considerable decrease of runoff in the western (from Echam5) or southern part (from Medres) might affect the agricultural water consumption and crop growth. Meanwhile, large reduction of river channel flow from November to May the following year may cause difficulties to withdraw water for irrigation and domestic water uses, whereas more critical floods will be expected during the summer, especially in August.

6.3 Limitation of this study

In this study, outputs of two GCMs from IPCC AR4 were adopted to project the impacts of climate change in the LR basin. So far, however, simulation results from IPCC AR5 are available from the Internet. It may be helpful to use the new climate scenarios from Representative Concentration Pathways (RCPs) (Meinshausen et al. 2011) rather than the SRES for improved knowledge and skills in estimating the future climate conditions. But, it is still important to report the impacts of climate change under the SRES framework in order to take a comparison with the simulation results derived from the IPCC AR5 outputs, considering the huge uncertainties involved in the projection of future climate and the scarcity of studies in this area.

The YZR is an important international river and plays a crucial role in providing water resources and maintaining ecosystem health not only for China but also for downstream countries, such as India and Bangladesh. It is important to investigate the climate change impacts on hydrological cycle, water resources and ecosystem with a perspective of the whole basin. The study is the first step of this work. In the next step, we plan to expand the study area to the whole YZR basin.

7 Conclusion

Climate change is projected to exert significant impacts on hydrological cycle and water availability, and consequently on agriculture and ecosystem systems in the TP. The VIC model coupled with SCE-UA was applied to quantify the impacts of climate change from the spatial and temporal perspectives in the middle of twenty-first century, taking the LR basin as a case study. Future climate datasets were generated by downscaling two GCMs under SRES A1B, A2, and B1 scenarios.

The VIC model satisfactorily reproduced the main characteristics of observed streamflow in terms of low relative errors and good agreement between observed and simulated flow series. The simulation results describe the spatio-temporal distribution with great variations for major hydrological variables. Temperature in the future may increase significantly, especially for daily maximum temperature with a value of more than 2 °C. Annual precipitation shows a slight increase, but monthly precipitation is likely to vary significantly, with a sharp increase in summer while dramatic decreases in other three seasons.

Spatially, spring snow cover will decrease tremendously. At the same time, the change of evaporation and runoff is conspicuous and the variations are substantial within the whole basin. On the temporal dimension, evaporation will rise with the magnitudes ranging from 14.89 to 29.71 % at the annual scale. The change of annual runoff and flow is comparable with the range being 8.26–14.23 and 6.04–13.92 %, respectively. However, the monthly change patterns are inconsistent, with rising trend in summer but reducing in other three seasons.

The significant spatio-temporal change of hydrological processes caused by climate change may greatly influence the potential water resources availability and agricultural production as well as ecosystem functions in the LR basin. Therefore, it’s important to take measures to mitigate the adverse impacts and adapt to the changing environment. The results obtained in this study will be expected to serve as a reference for future water resources management and climate change mitigation and adaptation.