Application of SWAT model with CMADS data for hydrological simulation in western China

Precise simulations of hydrological processes under the influence of climate change and human activities have special significance in arid basins. During the past 60 years, the annual average temperature and precipitation at the northern foothills of the Tianshan Mountains have increased at the rates of 0.035 C/year and 0.881 mm/year, respectively. Rising temperatures will change the temporal and spatial distributions and forms of precipitation, accelerate glacier retreat, melt snow on high mountains, cause the degeneration of frozen soil, and change the runoff composition in the Tianshan area. In this work, the CMADS (China Meteorological Assimilation Driving Dataset for the SWAT model) was combined with the SWAT (Soil and Water Assessment Tool) model to simulate runoff in the upper reaches of the Jing River and Bo River Basins in the Tianshan area. The results were as follows. (1) On the monthly scale, the average Nash–Sutcliffe efficiency (NSE) coefficients of the calibration period in the Wenquan and Jinghe–Shankou hydrological stations were 0.79 and 0.87, respectively, and the NSE coefficients of validation period were 0.71 and 0.82, respectively. On the daily scale, the NSE coefficients of the two hydrological stations were between 0.69 and 0.77. The simulation results were considered to be ideal on the monthly and daily scales. (2) Under different climate scenarios and land-use patterns, the cultivated land in the basin leads to the reduction of runoff, and the grassland and woodland stabilise the river flood season. Lakes and wetlands, which can reduce the flow in the flood season and provide water for rivers in the dry season, are very important for runoff regulation. Compared with the traditional meteorological stations, CMADS demonstrates good representativeness and reliability in the Jinghe River and Bohe River Basins under different climate and land-use scenarios, greatly improving the runoff simulation ability.

Small changes in precipitation and temperature are usually not statistically significant; however, they have a significant impact on river runoff. The rise in global temperatures has led to the acceleration of regional water cycles and increased precipitation, with seasonal droughts, floods, and other disasters occurring more frequently (Zhao & Cheng ; Ye et al. ). In addition, human activities, such as building reservoirs and reclaiming farmland, have changed land use, affecting the spatial and temporal distribution of local water resources, as well as local hydrological processes (Guang et al. ; Yang et al. ).
The impact of human activities on water resources has radiated from many river basins to the atmosphere, soil circle, and other levels. Hydrological simulation research on climate change and land-use change in the Jing River and Bo River Basins has also been carried out. Climate change in the river basins during the years 1960-2015 has been analysed (Wang et al. , ), revealing that the annual average temperature in the region was 0.33 C, and the average temperature for the most recent 7 years has increased by 1 C compared with the average temperature over the past 50 years. The annual precipitation increased at a rate of 7.97 mm/year. Compared with the average precipitation over the past 50 years, the average precipitation for the past 7 years has increased by 32.45 mm/year. The annual potential evapotranspiration in Ebinur Lake Basin over the past 57 years is 2.51 mm/year, which shows a downward trend, whereas the average potential evapotranspiration for the past 7 years was 77.36 mm less than that for the preced- forests and grasslands, saline-alkali lands, dry lakebeds, and deserts in the basin increased in the research area, whereas water bodies and other land types prominently decreased by 34.8%. It was predicted that from 2014 to 2030, forests, grasslands, saline-alkali lands, and deserts in the Ebinur Lake wetland nature reserve would increase, whereas lakebeds, water bodies, and other land types would reduce. Meng et al. (b) fully verified the meteorological driving data by using CMADS meteorological reanalysis data to drive the SWAT model. According to its applicability in the watershed, the spatio-temporal evolution law of soil moisture and evapotranspiration was clarified.
The above-mentioned studies have revealed significant information regarding climate change, land-use change, and the adaptability of CMADS in the Jingbo River Basin.
Most of the research has focused on the following topics: (1) impact of climate change and human activities on water resources in the Jingbo River Basin; (2) ecological environment assessment of the Jingbo River Basin; (3) landscape pattern and desertification of the basin; and (4) ecological restoration benefits and simulation of the Ebinur Lake's tail region. However, the viewpoints of the studies are scattered, and the studies have not been carried out from the perspective of the entire basin. In particular, there is a lack of simulation studies on upstream hydrological processes. In fact, changes in the upstream water resources have an extremely important impact on the middle and lower reaches, especially under the influence of climate change and human activities, whereas the uncertainty of the output and assessment of hydrological simulation in this region is significantly enhanced. Therefore, to fully understand the essence of the Jingbo River Basin under climate change and hydrological responses to landuse change, particularly in the expected extreme scenarios of the future, this study uses the distributed hydrological model SWAT and data from CMADS to analyse different climate change and land-use change scenarios. In this manner, the northwest arid areas are given priority, with the ice and snow melt water supply mechanism from runoff assigned an important reference value.
To fully understand the hydrological response of the Jingbo River Basin under climate change and land-use changes, especially any changes in the hydrological elements under extreme future scenarios, we applied a distributed hydrological model and existing meteorological and reanalysis data as drivers to simulate the effect of different climates and land uses in this study. First, the SWAT model was driven by different meteorological data (measured meteorological data and CMADS reanalysis data) to obtain simulated results that can better reflect the basin's meteorological conditions. Second, we considered changes in land use to explore the processes of runoff response in different scenarios.

Study area
The Jing River and Bo River Basins are located in the hinter-

Digital elevation model
In this study, we used the SRTM 90 m digital elevation model (DEM) data (http://srtm.csi.cgiar.org/selection/ inputCoord.asp). We considered the surface of the study area and height of the river basin in the region 160-4,600 m above sea level, which accounts for a total area of 11,275 km 2 . The west, south, and north of the basin have continuous mountains, whereas the east has a relatively flat plain.

Soil data
The SWAT model for calculating the field water-holding capacity and hydraulic conductivity coefficients needs soil attribute parameters (Fischer et al. ; Meng et al. d).
The soil data used in our study are based on the World Soil Data Set, which has a resolution of 1 km, provided by the Cold and Arid Regions Scientific Data Center (http:// westdc.westgis.ac.cn) and the Chinese soil data set of HWSD (v1.1). In our model, we reclassified the soil types with the same soil properties, utilising the soil water characteristics from the software.

Land-use data
The LUCC land-use data were obtained from remote-sensing observations of the Ebinur Lake Basin. Glaciers are indicators of climate change and important water resources in northwest China. Thus, research on hydrological response to climate change is necessary. In this research, changes in glaciers need to be considered. Therefore, the second glacier inventory data set was superimposed on the original land-use data (Meng et al. b).

Meteorological driving data
The CMADS used in the SWAT model form part of the public data set developed by Prof. Xianyong Meng at the China Agriculture University (http://www.cmads.org).
CMADS was established using the CMA Land Data Assimilation System, which combines data loop nesting, model estimation, and other technical means. The data set includes seven types of meteorological elements: daily accumulated precipitation, maximum temperature, minimum temperature, daily average relative humidity, daily average wind speed, daily average solar radiation, and daily average air pressure. The amount of precipitation according to the background field from the CPC morphing technique and the observational data of China precipitation automatic station were also integrated. Temperature, pressure, relative humidity, and wind speed were based on data from the National

Centers for Environmental Prediction-Global Forecast
System, which used the Space and Time Mesoscale Analysis System algorithm to fuse the observational data. Radiation fluxes were based on the International Satellite Cloud Climatology Project data as background data, and the fourth and fifth series of FengYun-2 nominal map data were obtained using the Discrete Ordinate Radiative Transfer atmospheric model. Inversion was carried out to calculate the total radiation irradiance of the ground-incident solar radiation at each grid point.
CMADS provides a variety of data utilisation formats, and watersheds can be selected that are directly used as input to the SWAT model. CMADS has a high spatial and temporal resolution, comprehensive meteorological elements, and strong reliability. It is widely used in research for hydrological simulations of many watersheds in China, such as the Hei River, Hun River, Manas River, Jingbo River, and Juntang Lake River (Wang et al. ; Meng et al. a, b, c). The time range of the CMADS data is 2008-2016. Meanwhile, the SWAT website recommends using CMADS as the weather-driven data set for the model.

Climate change scenarios
The climate change process is relatively slow and has longterm characteristics in terms of water circulation in the basin. The process of river runoff in cold and dry regions has typical characteristics. Precipitation is the primary source of water in the basin. However, it participates in the water cycle of the basin in various forms. Air temperature is the main factor controlling the form of precipitation and its spatial and temporal distributions. An increase in temperature changes rain to snow, accelerates the melting of glaciers and snow, and increases runoff in the watershed in the short term. greatly from model to model and is mainly concentrated between À10 and 10% ( Jiang & Fu ). However, in the central and eastern parts of Xinjiang, the increase in rainfall could be as much as 15% (Zhou et al. ). Accordingly, we set up a scenario in which the temperature increases by 2 C, and the rainfall increases by 10% to simulate the scenario in which human activities are not changed and the scenario in which human activities are reduced by half (1 C, 5%). Overall speaking, to discuss the runoff changes under different temperature and precipitation scenarios in the basin, we set up eight climate change scenarios (Table 1): temperature changes of À2 , 2, and 4 C (assuming that precipitation does not change), precipitation changes of À10, 10, and 15% (assuming that the temperature does not change), and situations in which the temperature and precipitation both change simultaneously.

Land-use change scenarios
The expansion of urban construction land and changes in cultivated land area and human activities, such as damming rivers, directly cause changes in the quantity of water in the river basin, which affect water circulation in the river basin, incurring both short-term and far-reaching consequences.
Increases in the basin's population, improvements in the level of urbanisation, and the development of industrial and commercial sectors have led to an increase in water consumption. Moreover, surface changes will affect the processes of water interception, infiltration, and depression filling. Indeed, water conservation projects, such as reservoirs and fabricated ditches, directly change the runoff process.
In this study, three soil types were considered in the land-use scenarios: (1) on the basis of a consistent water area, with other lands including construction land, woodlands, grasslands, and shrubbery reduced by one-third and converted into arable land; (2) on the basis of consistent land-use, but with the total water area reduced by onethird and converted into grassland; (3) on the basis of consistent cultivated land and water areas, but with other lands including grasslands, woodlands, and shrubbery reduced by one-fifth to be used for urban construction (Table 2).

SWAT model
The SWAT model is based on the Simulator for Water Resources in Rural Basins model, a semi-distributed  values. The calculation formula is as follows: where Q is the runoff value variable; Q m and Q s represent the observed and model values, respectively; and Q m indicates the mean observed runoff. The NSE is in the range of À∞ to 1. When the NSE calculation result is 1, the observed value and modal value can be regarded as consistent; when the value is between 0.5 and 1, the model results are acceptable; and when the NSE is less than 0, the model results can be considered as poor. The deterministic coefficient R 2 determines the degree of correlation between the variables, and it was calculated using the following equation: where Q m and Q s are the same as in Equation (1) (Table 4).

Runoff simulation
At the monthly scale, the NSE coefficients of OBS þ SWAT hydrological station in the lower reaches of the river also shows this characteristic apparently. When the temperature drops by 2 C, the runoff decreases by 21%. When the air temperature increases by 2 C, the annual average runoff increases from 12.9 to 21.3 m 3 /s, and the monsoon significantly advances, in which the spring runoff increases to 5.7 m 3 /s.

Precipitation changes
The simulated results for different precipitation conditions are shown in Figure    the lower reaches of the basin and has a larger catchment area. Simulated increases of 10 and 15%, respectively, representing the amount of runoff in summer, were found with a slight increase in runoff in spring as well.

Comprehensive changes
For scenarios where the temperature and precipitation changed simultaneously (i.e., scenarios G and H), the simulation results are shown in Figure 6. In these cases, the runoff significantly increased. On the one hand, the water produced by melting ice and snow increased considerably and the mountains dropped increased amounts of water, all of which directly led to a substantial increase in the runoff and the advancement of the flood season. At the Wenquan hydrological station, the temperature increased by 1 C, the precipitation increased by 5%, and the annual runoff increased from 9.8 to 25.3 m 3 /s. When the temperature rose by 2 C, the precipitation increased by 10%, and the annual runoff increased to 32.4 m 3 /s. Runoff at the Jinghe hydrological station was similar to that of the hot spring station but less severe. When the temperature  increased by 1 C, the precipitation increased by 5%, and the annual average runoff increased by 13.4% to 27.3 m 3 /s. For an increase in temperature of 2 C, the precipitation increased by 10%, and the annual runoff increased to 40.3 m 3 /s.

Hydrological response under different land-use scenarios
The simulated results for the three chosen land-use scenarios are shown in Figure 7. The annual average runoff at the Jinghe station for scenarios S1, S2, and S3 were 10.35, 11.6, and 12.75 m 3 /s, respectively, which corresponded to relative increases of 21, 12, and 11.6%. The annual average runoff at the Wenquan station for these three scenarios was 10.3, 9.1, and 11.2 m 3 /s, respectively. Under the scenario of increased cultivated land area (S1), irrigation water consumption increased, which also led to evapotranspiration.
The amount of runoff also increased, resulting in a significant decrease in the runoff, especially in summer. Water areas have certain replenishment and regulation functions for runoff: in the S2 scenario, the reduction of the water area has little effect on runoff. Reduced summer runoff was seen at the Jinghe station and Wenquan station when the water area was reduced. Finally, in the S3 scenario where the amount of grass in the basin decreased, the area of construction land increased significantly. As woodlands, grasslands, and shrub forests have a certain degree of water-holding capacity, when large-scale degradation of these lands occur or the land is developed into cities, the basin's ability to retain water correspondingly changes.
Indeed, the simulated water-holding capacity of the basin decreased, resulting in more concentrated runoff during the summer flood season. During July and August, runoff at the Wenquan station increased significantly due to grassland degradation.

CONCLUSIONS
In this study, the SWAT model was driven by data from CMADS, and the representativeness and feasibility of these data to different climate and land-use change scenarios in the Jing River and Bo River Basins were analysed.
The simulation results were analysed to characterise the hydrological processes of the basin. The main conclusions of this study are as follows: (1) CMADS has good representativeness and reliability in the Jing River and Bo River Basins, which greatly improves the runoff simulation ability. However, the simulation ability of the SWAT model driven only by the measured meteorological data is poor, with an NSE coefficient of less than 0.5.
(2) According to the simulation of different climate change scenarios, we found that an increase in temperature will lead to the advance of the snow melting period, mainly the increase in runoff during spring. An increase in rainfall will significantly increase the annual runoff.
(3) Meanwhile, the increase in cultivated land area in the Jing River and Bo River Basins is the key factor to reduce runoff. Grassland and forest land mainly provide the river flood season with more stability. Lakes and wetlands are very important for runoff regulation, which can reduce the flow in flood conditions and provide water for rivers in the dry season.