Abstract
To study the impact of future climatic changes on hydrology in the Kaidu River Basin in the Tianshan Mountains, two sets of future climatic data were used to force a well-calibrated hydrologic model: one is bias-corrected regional climate model (RCM) outputs for RCP4.5 and RCP8.5 future emission scenarios, and the other is simple climate change (SCC) with absolute temperature change of −1~6°C and relative precipitation change of −20%~60%. Results show the following: (1) temperature is likely to increase by 2.2°C and 4.6°C by the end of the 21st century under RCP4.5 and RCP8.5, respectively, while precipitation will increase by 2%~24%, with a significant rise in the dry season and small change in the wet season; (2) flow will change by −1%~20%, while evapotranspiration will increase by 2%~24%; (3) flow increases almost linearly with precipitation, while its response to temperature depends on the magnitude of temperature change and flow decrease is significant when temperature increase is greater than 2°C; (4) similar results were obtained for simulations with RCM outputs and with SCC for mild climate change conditions, while results were significantly different for intense climate change conditions.
1. Introduction
The Tianshan Mountains, regarded as the “water tower of Central Asia” [1], are located in the innermost center of the Eurasian continent. The long distance to the surrounding oceans causes a dry climate, especially for the surrounding basins. Rivers starting in the mountainous regions provide agricultural and domestic water for the surrounding basins and oases. With their distinctive topographic and landscape features, the Tianshan Mountains show a unique energy balance and hydrological cycle and are expected to be sensitive to climate change [2, 3].
Many reports show a widespread climatic and hydrologic change in the Tianshan Mountains during the past few decades [4]. For example, temperature demonstrated a significant rising trend (significant level is smaller than 0.001) at a rate of 0.33~0.34°C/decade during 1960~2010, which is higher than China (0.25°C/decade) and the entire globe (0.13°C/decade) [5, 6]; precipitation increased substantially in most regions especially for the middle and high latitudes; glacier area decreased by 11.5% and the thickness of snowpack has also decreased [5, 7]. Pan evaporation and wind speed have also changed [3]. The annual runoff increased as well, for example, for the Urumqi River, the Kaidu River, and the Aksu River [8, 9].
Future changes in the streamflow and watershed hydrology have become increasingly important to water resource management in the Tianshan Mountains. However, only a limited number of studies currently focus on impact of future climate change on hydrology; for example, Sorg et al. [1] indicated that the total runoff is likely to remain stable or even increase slightly in the near future but it will decrease at the end of the 21st century for Central Asia. There are also researches demonstrating that the annual runoff will decrease to some extent in the first half of the 21st century [2, 10]. Previous studies seldom address implications of climate change on the hydrological cycle and hydrological components (e.g., ET, surface flow, and groundwater). To complement these studies, this paper aims at understanding the future hydrological system and assessing the responses of the hydrologic system to climate change.
In the present study, two sets of climatic data, that is, RCM outputs and SCC data, are used to force SWAT [11] and are applied to the Kaidu River Basin, a typical watershed on the south slope of the Tianshan Mountains, to assess future changes of the hydrologic cycle and the hydrological effects of changes in climate variables. Questions that are addressed include the followings. How will the future climate and hydrology change in this region? (2) What is the effect of climate change on the hydrologic cycling? (3) What is the difference between simulations with RCM outputs and SCC? Understanding these issues will enable assessing the future hydrological change and its unique hydrometeorological processes better. The remaining is constructed as follows: Section 2 introduces the study area and data; Section 3 describes the hydrological model, bias correction methods, and analysis procedures; Section 4 provides the results and discussion, followed by conclusions in Section 5.
2. Study Area and Data
The Kaidu River Basin, with a drainage area of 18,634 km2 above the Dashankou hydrological station, is considered to be a very typical watershed in the Tianshan Mountains (Figure 1). Its altitude ranges from 1,342 m to 4,796 m above sea level (asl) with an average elevation of 2,995 m. The climate here is temperate continental with alpine climate characteristics and obvious seasonal variation. This river provides water resources for agricultural activity and the ecological environment of the oasis, with an area of over 70,000 km2 and a population of over 1.15 million, which is mainly stressed by water scarcity [12]. Therefore, projecting the impact of future climate change on water resources is urgent for the sustainable development of this region and it also provides information on the implications of climate change on the water tower in Central Asia.
The daily observed meteorological data (including precipitation, maximum/minimum temperature, wind speed, and relative humidity of two meteorological stations Bayanbulak and Baluntai) from 1970 to 2005 are from the China Meteorological Data Sharing Service System (http://www.cma.gov.cn/2011qxfw/2011qsjgx/). The annual mean temperature at the Bayanbulak meteorological station amounts to −4.1°C and the mean annual precipitation is 278 mm (Figure 1). Generally, precipitation falls as rain from May to September each year and as snow from October to April.
The observed streamflow data at the Dashankou hydrologic station are from the Xinjiang Tarim River Basin Management Bureau. The average flow at the Dashankou hydrological station amounts to approximately 120 m3/s (equivalent to 202 mm runoff/year), ranging from 15 m3/s to 973 m3/s.
3. Methodology
3.1. Regional Climate Model and RCP Scenarios
The outputs of a regional climate model (RegCM4.0) [13] forced by a global climate model (Beijing Climate Center Climate System Model, BCC_CSM1.1) [14, 15] at a horizontal resolution of 50 km are used as future climate data. Firstly, the RCM model was validated with the observational data set over China for the period from 1976 to 2005 and then it was used to predict the future climate change under the new emission scenarios of the Representative Concentration Pathways, RCP4.5 (lower emission scenario) and RCP8.5 (higher emission scenario). RCP4.5 is a stabilization scenario with the total radiative forcing rising until 2070 and then remaining at a stable centration of 4.5 W/m2. In contrast, RCP8.5 is a continuously rising radiative forcing pathway (at a target of 8.5 W/m2 in 2100) with a further enhanced residual circulation and significant CH4 increases [16, 17]. The RCM validation shows reasonable simulations of temperature and precipitation were obtained over China and, compared to the BCC_CSM1.1 model, marked improvement of the RCM was achieved in reproducing present day precipitation and temperature (for more details refer to [18]).
3.2. Bias Correction Methods
Five precipitation and three temperature correction methods were selected to bias-correct the raw RCM outputs. Precipitation correction methods include linear scaling, local intensity scaling, power transformation, distribution mapping, and quantile mapping. Temperature correction methods include linear scaling, variance scaling, and distribution mapping. They are combined into 15 schemes to evaluate their performances in simulating streamflow. It turns out that the precipitation correction methods have more significant influence than the temperature correction methods on streamflow simulation, and the power transformation and quantile mapping perform best in terms of frequency based statistics. Thereafter, the quantile mapping method (for precipitation) and the distribution mapping (for temperature) are selected to correct the raw RCM outputs for the future period (for more details see [19]).
3.3. Hydrologic Model and Uncertainty Analysis Method
SWAT has been extensively used for the comprehensive modeling of the impact of management practices and climate change. SWAT simulates the hydrologic and sedimentary processes, plant growth, river routing, and in-stream water quality process, among which the surface runoff is calculated from daily rainfall and snowmelt with a modified Soil Conservation Service (SCS) curve number method [20]; water routing is simulated using variable storage or the Muskingum river routing method [11].
The SWAT model input includes the digital elevation model (DEM), soil textural and physicochemical properties, and land use data. The meteorological variables including daily precipitation, max./min. temperature, relative humidity, solar radiation, and wind speed were used to force the SWAT model. SWAT uses elevation bands to represent the topographic effects on precipitation and temperature. Within each elevation band, the precipitation and temperature are estimated based on their lapse rates. For more details, refer to the SWAT manual (http://www.brc.tamus.edu/).
The SWAT model (forced by the observed meteorological data) was calibrated against the observed streamflow. The calibration period is from 1986 to 1989 and the validation period from 1990 to 2005 [21]. The calibrated optimal parameters are then kept fixed in the following simulations. The evaluation indices for the hydrological model include NS, PBIAS, and the determination coefficient . Considerwhere and are the th observed and simulated flows, is the mean of the observed data, and is the number of observations. Normally , , and are taken as the criteria for satisfactory modeling of the river discharge and the model performance can be evaluated as excellent if and [22].
GLUE (generalized likelihood uncertainty estimation) [23] is an uncertainty analysis technique, in which the parameter uncertainty accounts for all sources of uncertainty, such as input uncertainty, structure uncertainty, parameter uncertainty, and response uncertainty [24]. In GLUE, the parameter uncertainty is described as a set of discrete “behavioral” parameter sets with corresponding “likelihood weights.”
The procedure of a GLUE analysis consists of three steps. Firstly, after the definition of the “generalized likelihood measure,” , a large number of parameter sets are randomly sampled from the prior distribution and each parameter set is assessed as either “behavioral” or “nonbehavioral” by comparing its value of to the threshold value. Secondly, each behavioral parameter set is given a “likelihood weight” and we gave them equal weights in this study. Finally, prediction uncertainty is represented by 5% and 95% quantiles of the cumulative distribution of the behavioral parameter sets.
Two indices are used to quantify the quality of the uncertainty performance. Those indices are the percentage of measurements bracketed by the 95% prediction uncertainty band (-factor) and width of band (-factor, calculated by the average width of the band divided by the standard deviation of the corresponding measured variable).
3.4. SCC Data Description and Analysis Procedures
In the following section, temperature and precipitation are denoted as and and the absolute and relative changes are represented by Δ and . For example, refers to an absolute temperature change and a relative precipitation change. The hydrological processes analyzed in this study include streamflow, surface runoff, subsurface runoff, and evapotranspiration, which are denoted as , , , and ET, and their relative changes are described as , , , and , respectively.
The SCC was constructed to represent a wide range of changes in climatic variables and how these changes might translate in streamflow and other hydrological components and also to analyze the differences between simulations with RCM outputs and SCC. For SCC, perturbations of the corrected RCM simulated and from 1986~2005 (control period) are set; that is, for , an additive change () is used: , 0, 1, 2, 3, 4, 5, and 6°C. For , a relative change () is used: %, −10%, 0, 10%, 20%, 30%, 40%, 50%, and 60%. They are put into 81 SCC scenarios, with and being the climate for control period.
By investigating the transient evolution of climate change in the corrected RCM outputs on decadal scales, five periods (each spanning 20 years) are defined: 1986~2005 (control period) and 2020~2039, 2040~2059, 2060~2079, and 2080~2099. Due to the intra-annual characteristics of the hydrometeorology in the Kaidu River Basin (Figure 1), the wet season (from April to September) and dry season (from October to March next year) are defined based on the intra-annual distribution of and ; for example, and in the wet season account for 88% and 73% of their annual amounts. The climatic and hydrological changes are classified into three categories, that is, a significant change, small change, and insignificant change, to clearly demonstrate the changing magnitude according to the values of relative change for precipitation and hydrological components and absolute change of temperature. These categories are presented in Table 1.
4. Results and Discussion
4.1. Validation of the Hydrological Model and the Bias Correction Methods
Performance of the hydrological model forced by observed meteorological data and the 95% prediction uncertainty bands are shown in Figure 2. The simulated streamflow agrees quite well with the observation for both calibration period (1986~1989) and validation period (1990~2002). For the uncertainty analysis, NS is used as and 0.70 as threshold value with 10,000 initial parameter sets; 288 sets were selected as behavioral points. The results show that most of the observations are bracketed by the 95% prediction uncertainty band (-factor being 87% and 80% for calibration and validation periods and -factor being 1.18 and 1.19, resp.). The lower -factor for the validation period can be partly attributed to operation of hydropower station since 1991 (Figure 2) which leads to great fluctuation in winter streamflow. Statistics of model efficiency (Table 2) indicate excellent performances for both calibration and validation periods, with “NS”s and “”s over 0.80, which is highly acceptable according to Moriasi et al. [22]. Concerning the monthly streamflow, the “NS” is 0.89 during 1986~2005 and it indicates that the SWAT model captured the natural monthly streamflow variability adequately.
(a)
(b)
(c)
The performances of bias-corrected RCM outputs (compared to observed meteorological data) are listed in Table 2. The “NS”s are −0.57 (0.57) and 0.77 (0.95) for daily (monthly) precipitation and temperature for 1990~2005, respectively. And the statistics of the streamflow simulated with the bias-corrected RCM outputs shows acceptable results with “NS”s equal to 0.46 and 0.62 and PBIAS within 10% for daily and monthly streamflows.
4.2. RCM Projected Hydrometeorologic Changes
4.2.1. Changes in Temperature and Precipitation
Temperature is highly likely to increase in the future, with a basin warming of 1.0~2.2°C and 1.6~4.6°C under RCP4.5 and RCP8.5 in the 21st century (Table 3). Temperature increases continuously under both scenarios but the magnitude is larger under RCP8.5 (Figure 3).
Precipitation shows an overall increasing trend in the 21st century with an annual increase of 2%~16% and 7%~24% under RCP4.5 and RCP8.5, which confirms the previous arguments of Sorg et al. [1]. However, precipitation change varies substantially among seasons (Figure 3). Normally, a small increase in the wet season (−2%~16%) and a significant increase during dry season (18%~78%) are projected. Note that the relative increase (not the absolute increment) of precipitation for the dry season is much bigger than for the wet season, which is in line with the climate changes in other regions, for example, the semiarid Colorado River Basin [25] and the wet Ganges-Brahmaputra-Meghna basin [26].
4.2.2. Changes in the Hydrological Cycle
The changes in precipitation and temperature cause changes in potential streamflow. The average annual streamflow rises by −1%~18% and 4~20% under the RCP4.5 and RCP8.5 in the 21st century, based on the average annual streamflow of 194 mm for the control period (1986~2005) (Table 3). Note that the streamflow stopped increasing in 2080~2099 (end of 21st century) under RCP8.5 despite the rise in precipitation, which confirms the finding of Sorg et al. [1] and may aggravate water scarcity in this region.
Figure 3 also shows the projected changes in surface runoff (), subsurface runoff (), and evapotranspiration (ET) under RCP4.5 and RCP8.5. Overall, changes of hydrologic components are bigger for RCP8.5 than for RCP4.5. The annual change of is insignificant (<5%) but with obvious seasonal variability; for example, changes of range from −22% to 2% for the wet season and in 4%~78% for the dry season under RCP8.5. The annual changes by −0.7%~17% and 4%~18% for RCP4.5 and RCP8.5, which is consistent with the changes of . ET increases continuously in the 21st century with average increases of 2%~10% and 7%~24% under RCP4.5 and RCP8.5.
4.3. Response of Hydrological Cycle to Climate Change
The response of the hydrological cycle to climate change is estimated by running the hydrological model forced by SCC. The responses of , , , and ET to climate change are demonstrated with response surfaces in Figure 4. is positively related to and negatively related to . The relationship of and is almost linear with the streamflow elasticity () being about 1.0 when °C; that is, a 1% change in the mean annual precipitation results in a 1% change in the mean annual streamflow. is lower than that for other arid regions, for example, 2.0~3.5 for Australia [27]. The possible reasons are mentioned as follows: the Kaidu River Basin, located in the south slope of the Tianshan Mountains with a high average altitude (2,995 m), is characterized by a cold climate (average annual temperature is −4.1°C for the Bayanbulak station) and accordingly there is a low amount of energy available for ET, which results in a relatively high runoff coefficient () and consequently a low streamflow elasticity; (2) the streamflow is also influenced by temperature dominated snowmelt (snowfall accounts for about 17% of watershed precipitation), which reduces the dependence of streamflow on precipitation and therefore results in a low streamflow elasticity [27].
The response of to depends on the magnitude of . decreases slightly when °C while it decreases dramatically when °C for both wet and dry seasons. For example, when °C, a 40% precipitation increase results in an average value of being 240 mm (23% increase compared to the average streamflow of 194 mm) but when °C, the same precipitation increase only generates an average of 180 mm (about 7% decrease) (Figure 4).
The responses of , , and ET to climate change are also demonstrated in Figure 4. For , the responses of to are quite different for the wet and dry seasons: the higher , the lower for the wet season but the higher for the dry season. Since in the dry season only accounts for 13% of the annual , the response of annual is consistent with that of the wet season. For , the responses of to and are similar to the responses of due to the dominant role of groundwater recharge in water yield in the Kaidu River Basin. For ET, it is mainly influenced by with temperature sensitivity () being 7.3%/°C. To verify this result, we firstly investigated basin-scale energy and water budget using the Budyko method [28, 29]. It is shown that ET is mainly energy limited rather than water limited (average ET/ and PET/). Secondly, the high determinate coefficient (significant level is smaller than 0.01) between the mean annual and ET also indicates that ET has a strong correlation with . This is consistent with previous studies, which have shown that a significant variation in ET is expected to follow changes in air temperature [30, 31].
In addition, simulations with RCM outputs are shown in Figure 4 to analyze the differences between simulations of these two data sets. Two typical periods of RCM simulations are selected, that is, 2020~2039 under RCP4.5 and 2080~2099 under RCP8.5, to represent mild and intense climate change scenarios (shown as blue and red stars in Figure 4). It is indicated that the simulations of hydrological components with RCM outputs for 2020~2039 under RCP4.5 (mild climate change) are close to these of the nearby contour lines (simulations with SCC), which suggests that similar results of , , , and ET are obtained for RCM outputs and for SCC under mild climate change scenarios. However, for 2080~2099 under RCP8.5 with and °C, the simulated values of , , , and ET deviate from the simulations of SCC. There are two possible reasons: changes of other meteorological inputs, that is, solar radiation, wind speed, and humidity, are slightly smaller for 2020~2039 under RCP4.5 than those for 2080~2099 under RCP8.5 (−0.8%, 2.6%, and 0.9% compared to −2.2%, 4.1%, and 1.4%); (2) for 2080~2099 under RCP8.5, precipitation increases by 24% with great seasonal variation, which may alter the hydrological regime; for example, precipitation increases by 139% for March, April, and May while it decreases by −0.1% for June, July, and August. Since changes of solar radiation, wind speed, and humidity are within , the second reason, that is, the shift of the precipitation temporal distribution, contributes a lot to the deviation of simulations with RCM outputs from simulations with SCC.
Furthermore, the exceedance probability curves of the annual runoff in response to climate change are demonstrated in Figure 5. The exceedance probability curves are almost parallel when ranges in 0~6°C. However, the responses of to are not the same for each exceedance probability: high sensitivity of with probabilities less than 0.1 and low sensitivity of with probabilities larger than 0.9. A comparison of the simulations with RCM outputs (four future periods under RCP4.5 and RCP8.5) and with SCC (the same changes in or with the corresponding RCM outputs) indicates that differences between simulation with RCM outputs and SCC are becoming greater as climate change gets more intense; for example, the simulation with RCM for 2080~2099 under RCP8.5 overestimates the corresponding simulations with SCC (Figure 5(h)), which collaborates the conclusion that under intense climate change scenarios the simulated hydrology with RCM deviates from that simulated with SCC.
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
The contributions of hydrologic components to water yield are displayed by the De Finetti diagram in Figure 6. For the control period, the averages of , , and ET are 0.22, 0.28, and 0.50. For SCC, as increases from 0 to 6°C, the contribution of ET increases rapidly from 0.49 to 0.73 and the contributions of and decrease from 0.22 to 0.11 and from 0.29 to 0.16. has a more significant influence on the proportion change than . As changes from −20% to 60%, ET decreases from 0.71 to 0.58 and and increase from 0.13 to 0.15 and from 0.16 to 0.27. For simulations with RCM outputs, proportions of hydrological components do not change significantly under RCP4.5, while the proportion of ET shows a significant increase under RCP8.5.
4.4. Sources of Uncertainty and Other Considerations
There are uncertainties in estimating climate change impact on hydrology. As indicated by previous studies [32], the sources of uncertainty may rise from climate models, emission scenarios, downscaling, and the hydrological model.
For hydrological modeling itself, the effect of future climate in any specific catchment is difficult to project due to the possibility that the hydrological system may not be stationary with complex feedbacks [33]. For example, the same land cover and soil data were used for both control period and future climate change period, which may not well represent the land surface under the future climate changes. Effects of land cover change on streamflows and other components of the hydrological cycle are not considered.
Though uncertainty in hydrologic modeling was quantified with the GLUE method, it only accounts for part of the total uncertainty in climate change impact studies [34]. Uncertainties associated with the climate model and downscaling were not considered here although two emission scenarios were included. Any uncertainty associated with them may cause the results to deviate from reality. However, we are dedicated to pursuing a thorough investigation of the response of the hydrological cycle to future climate change for this region and we believe this study is an important first step in achieving this goal.
5. Conclusions
This study assessed the implications of climate change on hydrology in a typical watershed in the Tianshan Mountains with two sets of climatic data, that is, RCM outputs and SCC, loosely coupled to a hydrological model (SWAT). Major conclusions can be summarized as follows:(1)The hydrological model shows excellent performance with “NS”s over 0.8 for the daily streamflow for both calibration and validation periods. And the selected bias correction methods were effective in downscaling RCM outputs, with “NS”s being 0.57 and 0.95 regarding monthly precipitation and temperature.(2) increases by 1.0°C~2.2°C and by 1.6°C~4.6°C under RCP4.5 and RCP8.5 in the 21st century. For , it shows an overall increasing trend (2%~24%) with significant increase for the dry season (18%~78%) and relatively small change for the wet season (−2%~16%). The projected shows an overall increasing trend (−1%~18% and 4~20% for RCP4.5 and RCP8.5) in the 21st century.(3) increases almost linearly with while the response of to depends on the magnitude of and decreases significantly when is greater than 2°C.(4)Similar responses of , , , and ET to and are obtained for the RCM outputs and for SCC under mild climate change scenarios. However, for intense climate change scenarios, simulations of , , , and ET with RCM outputs (e.g., for 2080~2099 under RCP8.5) deviate from simulations with SCC.(5) has more significant influence on the proportion change of each hydrologic component than does. As increases from 0 to 6°C, the contribution of ET increases rapidly from 0.49 to 0.73 and and decrease by 0.11 and 0.13. As changes from −20% to 60%, ET, , and change by −0.13, 0.02, and 0.11 as a result.It is valuable to quantify the future responses of hydrology to climate change in the Tianshan Mountains. This study will provide useful information for water resource management and will serve as a basis for further climate change impact studies.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgments
Research was supported by the “Thousand Youth Talents Plan” (Xinjiang Project), the National Natural Science Foundation of China (41471030), and the Foundation of State Key Laboratory of Desert and Oasis Ecology (Y371163). The authors wish to thank Professor Xuejie Gao at the National Climate Center (China) for providing the outputs of the regional climate model used in this paper.