Quantification of climate change and land cover/use transition impacts on runoff variations in the upper Hailar Basin, NE China

Quantification of runoff change is vital for water resources management, especially in arid or semiarid areas. This study used the Soil and Water Assessment Tool (SWAT) distributed hydrological model to simulate runoff in the upper reaches of the Hailar Basin (NE China) and to analyze quantitatively the impacts of climate change and land-use change on runoff by setting different scenarios. Two periods, i.e., the reference period (before 1988) and the interference period (after 1988), were identified based on long-term runoff datasets. In comparison with the reference period, the contribution rates of both climate change and land-use change to runoff change in the Hailar Basin during the interference period were 83.58% and 16.42%, respectively. The simulation analysis of climate change scenarios with differential precipitation and temperature changes suggested that runoff changes are correlated positively with precipitation change and that the impact of precipitation change on runoff is stronger than that of temperature. Under different economic development scenarios adopted, land use was predicted to have a considerable impact on runoff. The expansion of forests within the basin might induce decreased runoff owing to enhanced


INTRODUCTION
Water resources form the foundation of sustainable socioeconomic development (Barnett et al. ). However, owing to continued global extreme climatic events and the negative influence of human activities, China is facing water resource problems that are becoming increasingly severe and that represent an important constraining factor on China's future sustainable socioeconomic development (Barnett et al. ; Piao et al. ; Voeroesmarty et al. ). In recent decades, under the influence of global climate change, the climate in northern China has shown an obvious trend of warming and drying (Fang et al. ; Piao et al. ). The water cycle in many watersheds has been affected considerably, and runoff in major watersheds has shown a rapid decrease (Bao et al. ; Wang et al. a). Human activities affect the hydrological cycle and the formation process of water resources by changing the mode of land use to varying degrees (Wheater & Evans ; Sterling et al. ), which can result in a series of problems such as wetland shrinkage and groundwater funneling (Kaushal et al. ; Sterling et al. ; Li et al. ). Therefore, it is of considerable scientific and practical importance to investigate the impact of climate change and land-use change on the hydrological cycle to resolve water resources problems in a changing environment.
Runoff is a vital link in the hydrological cycle that is also important in relation to the allocation of water resources within a basin (Milly et al. ). Changes in runoff directly affect life and production activities in a basin (Piao et al. ). Therefore, it is of considerable importance to undertake quantitative research on the impact of climate and land-use changes on runoff. Methods commonly used for the quantitative analysis of the impact of environmental changes on runoff can be divided into three categories: comparative basin tests, statistical analysis methods, and hydrological model simulations (Mishra & Singh ).
The comparative watershed method is used for the manual evaluation of the human impact on runoff by changing the natural geographical conditions (one or more watershed characteristics) of the test watershed. However, this approach is usually considered best for studying the effect of climate change in small watersheds but it is difficult to find two similar medium-or large-sized watersheds, and even the same watershed might undergo notable changes at different stages (Lorup et al. ). Statistical analysis can be used to analyze the trends of the change of hydrometeorological data, but it cannot consider the spatial heterogeneity of watersheds or the mechanisms via which land-use change and climate change might affect runoff change in watersheds (Gampe et al. ; Ishida et al. ). Therefore, comprehensive physically based tools are needed to obtain as much information as possible from limited existing data (Li et al. ). Hydrological models provide a framework for conceptualizing and studying relationships (Wang & Xu ). By linking model parameters directly with physically observable surface features, hydrological models can establish relationships among climate, human activities, and runoff (Leavesley  Region to the southwest of the city of Hulunbeier (Figure 1).
It has a temperate continental monsoon climate, and it covers an area of 54,500 km 2 . The basin is located at the junction of the western slopes of the Daxing'anling mountains and the northeastern edge of the Inner Mongolia High Plain (Wang et al. b). It is a fan-shaped basin that has large topographic fluctuation (Han et al. a).
The general trend of elevation (range: 536-1,706 m) is from low in the west to high in the east. The upper reaches of the Hailar Basin constitute the main area of the basin (Han et al. b). There are two flood seasons annually: a spring flood season and a summer flood season. The spring flood season that usually occurs during March-May reaches its peak in May (Fang et al. b). Runoff in this period is derived primarily from snowmelt and precipitation.
The summer flood season usually occurs during June-October, which is the period with the most concentrated precipitation (Xue et al. b; A et al. b).

Data sources
The data used in this study were divided into two parts. Part 1 data were used to analyze the water resources situation of the upper reaches of the Hailar Basin and to determine the relationship between runoff and climate factors. Part 2 were the input data required by the SWAT model. The input data of the SWAT model also included two main parts: spatial data and attribute data. Spatial data mainly included digital elevation model data (90 × 90 m), land use/cover data (1,000 × 1,000 m), soil distribution data, and spatial distribution data of the hydrological stations and meteorological stations. Attribute data mainly included soil type data, meteorological data, and hydrological data. Index tables of land-use type and soil type were established in the modeling process to ensure the required data in the model corresponded to each database.
Digital elevation model data: first, geometric correction was performed, and then clipping and projection transformation were undertaken using ArcGIS. Land use/cover data: first, the original data were downloaded from the Resources and Environment Science Data Center of the Chinese Academy of Sciences, and reclassification and projection transformation performed using ArcGIS. The data were divided into six categories: forest land, grassland, water area, urban land, unused land, and cultivated land, as shown in Figure 2(a). Soil type data and soil attribute data: these data were obtained from the Harmonized World Soil Database, which contains a large number of soil parameters.
The data are presented in a gridded format using the WGS1984 coordinate system. The soil classification system adopted is mainly FAO-90. As the data provided in the database are international standards, the SPAW software was required to convert the data from international standards to soil parameters of the United States Geological Survey standard. The soil distribution data in the Hailar Basin are shown in Figure 2(b). This study used daily meteorological and runoff data from 1980 to 2012. Meteorological data included daily precipitation, maximum and minimum temperature, humidity, and wind speed.

METHODOLOGY Mutation analysis
Mann-Kendall mutation test x i is greater than x j (1 j i), we have Under the assumption that the time series is random and independent, we have the following: When the elements x 1 , x 2 , . . . , x n are independent of each other and continuously and uniformly distributed: UF i is a standard normal distribution. Given significance level α, if |UF i | > U, there is an obvious trend change in the sequence, and the critical value of UF and UB is ±1.96.
Arranging time series X in the reverse order and then performing the calculation according to the above equation, we have By analyzing the statistical sequences UF k and UB k , the trend change of sequence X can be analyzed further, which allows the mutation time to be defined and the mutation region to be identified. If the value of UF k is >0, it indicates that the sequence shows an upward trend and vice versa.
When the value exceeds the critical line, it indicates that the upward or downward trend is significant. If the UF k and UB k curves have intersection points and the intersection points are between the critical straight lines, then the time corresponding to the intersection points is the time when the abrupt change is considered to start (Hamed ).

Sliding t-test technique
For a time series, the principle of the sliding t-test is to extract two subsequences of the main time series and then to test whether there is a significant difference between the average values of those two subsequences (Machiwal & Jha ). If there is a significant difference, the sequence is considered to have mutation (Zhao et al. ). Assuming Then, by taking a certain time in the series as a reference point and taking . , x n } and subsequence . , x n } forward and backward, respectively, based on the reference point, statistic t can be obtained based on the two subsequences as follows: In the above equation, Then, statistic t obeys the t distribution with n 1 þ n 2 À 2 degrees of freedom.
The specific steps of the sliding t-test for mutation are described in the following.
1. On the basis of the determination of the reference point, the lengths of subsequences n 1 and n 2 are determined. Normally, the lengths of the two subsequences are taken as the same, i.e., n 1 ¼ n 2 . Then, subsequences n 1 and n 2 are taken forward and backward, respectively.
2. By sliding the reference point backward in turn, taking out the corresponding subsequences, and calculating the corresponding statistics, the n À (n 1 þ n 2 ) þ 1 statistics can be obtained.
3. The obtained statistics are arranged in sequence to obtain the statistics sequence, the significance level @ is selected, and the corresponding standard statistics t @ are obtained from a lookup where SW 0 is the initial soil water content on the ith day, R day is the precipitation on the ith day, Q surf is the surface runoff on the ith day, E a is the soil evaporation and plant transpiration on the ith day, W seep is the seepage flow on the ith day, and Q gw is the amount of groundwater on the ith day.

Model calibration and validation
In this study, the simulation of daily runoff was performed In the above equations, Q obs i is the measured runoff value, Q sim i is the simulated runoff value, Q obs ave is the measured average runoff value, and Q sim ave is the simulated average runoff value. The R 2 values (range: 0-1) represent the fitting degree between the simulated and measured values. The

Quantification of climate and land-use contributions to runoff change
Based on the SWAT model, this study used scenario analysis to separate the influences of various factors on runoff, i.e., assuming that climatic factors or human activity factors remain constant and that another factor changes accordingly, the contribution rate of this factor to runoff can be analyzed quantitatively.
This study used the M-K mutation test and sliding t-test In this study, Q 1 , Q 2 , Q 3 , and Q 4 were used to represent the average annual runoff simulated under scenarios 1, 2, 3, and 4, respectively. Thus, Q 2 À Q 1 represents the impact of climate change on runoff, Q 3 À Q 1 represents the impact of land use/cover change (land-use change) on runoff, and Q 4 À Q 1 represents the total change of runoff within the basin. In this paper, we define the following: The CA-Markov model describes land-use change from one period to another, which allows predictions of the future trend of land-use change. The following formula can be used to predict land use: where S t and S tþi are the states of the land-use structure at time t and t þ i, respectively and P ij is the state-transition matrix.
A cellular automaton represents a gridded dynamics model with strong spatial computing capability. The CA-Markov model can be expressed as follows: where S is a finite discrete state set element, N is a cellular

Runoff trend analysis
The  1980, 1982, 1986, 1988, and 1992. Therefore, the sliding t-test method was used to determine the mutation year in this study.
In this study, the sliding t-test method was also used to analyze the Bahou Station runoff time series. To avoid variable point drift caused by different subsequence lengths, the sliding length was changed repeatedly. Finally, it was determined that the length of the two sliding sequences should be 4, i.e., n 1 ¼ n 2 ¼ 4, and 33 statistics relating to the Bahou Station formed the corresponding statistical sequence. If the significance level @ ¼ 0:05 were selected, the statistic t-t(6) and the critical value were found to be t 0.05 (6) ¼ China currently shows an overall downward trend. Therefore, the identified sudden change is probably attributable to climate change because temperature rise leads to increased evapotranspiration. Based on these two methods, the mutation point was set to 1988, and the study period was divided into the reference period (before 1988) and the interference period (after 1988).

Model calibration and validation
Parameter sensitivity analysis and model calibration are closely linked, and both are essential processes for runoff simulation using hydrological models. The SWAT-CUP software can be used to perform sensitivity analysis and calibration on relevant parameters for a runoff simulation.
Combined with the actual situation of the basin, the LH-OAT method provided in SWAT-CUP was used in this study to select parameters related to runoff simulation in the northern region for sensitivity analysis and calibration, as shown in Table 2.
Through the sensitivity analysis of the SWAT model, 11 parameters with high sensitivity were selected to calibrate and verify the model (

Contribution of land use and climate change to runoff variation
According to the abrupt change years, the study period was divided into the reference period (before 1988) and    Scenario simulation of climate change and land use/ cover change

Climate change scenarios
As shown in Table 5, under the condition of constant basin temperature, surface runoff will increase with the increase of precipitation. Under the condition of unchanged precipitation, surface runoff will change with the change of temperature. In all temperature-drop scenarios, surface runoff shows an increasing trend, whereas in all temperature-rise scenarios, surface runoff shows a decreasing trend.
As shown in Table 6, values when ΔT ¼ 0 represent the response of runoff to changed precipitation at constant temperature. It can be seen that the increase of precipitation increases runoff, and the more precipitation increases, the greater the increase in runoff. Values when ΔP ¼ 0 indicate the response of runoff to temperature change when precipitation is constant. It can be seen that runoff decreases as temperature increases, and the more the temperature increases, the more runoff decreases. Values when neither ΔT nor ΔP are zero indicate the runoff response effect when both temperature and precipitation change. It can be seen that when temperature decreases and precipitation increases, the increase in runoff is greatest. When temperature increases and precipitation decreases, a decrease in runoff is greatest. When precipitation increases by more than 10%, runoff increases markedly, indicating that precipitation has a more significant impact on runoff.

Land-use change scenarios
According     It can be seen from Table 7 that future areas of various land-use types will increase or decrease to varying degrees.
In the case of natural growth, from 2030 to 2050, the proportions of cultivated land, forest land, water areas, and construction land will increase, while the proportions of grassland and unused land will decrease. In the case of ecological protection, from 2030 to 2050, the proportions of forest land, water areas, and construction land will increase, while the proportions of cultivated land, grassland, and unused land will decrease. In the context of economic development, from 2030 to 2050, the proportions of cultivated land, water areas, and construction land will increase, while the proportions of forest land, grassland, and unused land will decrease.
The runoff simulation results under the different landuse scenarios are shown in Table 8. Under the natural growth scenario, ecological protection scenario, and economic development scenario, runoff shows a decreasing trend, but the degree of runoff reduction differs between the scenarios. Under the condition of natural growth, the proportion of the forest area will increase, the proportion of the grassland area will decrease, and runoff will decrease   showed that runoff in the basin will increase with the increase of precipitation. For every 10% increase in precipitation, the runoff will increase by 13.34% on average. Runoff in the basin will decrease with an increase in air temperature. For every 1 C increase in air temperature, runoff will decrease by 6.5% on average. Overall, the influence on runoff of precipitation change is greater than that of temperature change. The main reason might be that runoff in this basin is replenished primarily by precipitation (A et al.  (1) Annual runoff at the Bahou Station generally showed a downward trend that exhibited diminishing volatility.
The long-term runoff sequence changed abruptly in 1988. This study divided the runoff sequence into the reference period (before 1988) and the interference period (after 1988).
(2) The SWAT model demonstrated satisfactory applicability to simulating runoff at the Bahou Station. The R 2 , Ens, and R E values of the monthly runoff simulated in the calibration period were 0.75, 0.71, and 20.8%, respectively, while the corresponding values in the validation period were 0.74, 0.70, and 20.3%, respectively, proving that the model was suitable both for the calculation of the runoff contribution rate and for the simulation of runoff under different scenarios.
(3) Compared with land-use change, climate change was found to have a more significant impact on runoff change in the river basin. The contribution rate of climate change (land-use change) to runoff change in the river basin was 83.58% (16.42%).
(4) According to different combinations of temperature and rainfall change, temperature decrease will lead to evaporation and snowmelt runoff decrease, which will result in increased surface runoff in the watershed. A rise in temperature will increase both snowmelt runoff and evaporation; however, the amount of evaporation will be greater than the amount of runoff derived from snowmelt, which will lead to an overall reduction in runoff. A significant positive correlation was found between the change in precipitation and runoff change. Moreover, it was found that when precipitation increases (decreases) and temperature decreases (increases), an increase (decrease) in runoff is greatest.