Hydrological simulation of the Jialing River Basin using the MIKE SHE model in changing climate

Climate change and human activities have an important impact on the changing environment, leading to significant changes in the basin water cycle process. The Jialing River Basin, the largest tributary of the upper Yangtze River, is selected as the study area. Three different rainfall datasets, the China Meteorological Assimilation Driving (CMAD) dataset, the Tropical Rainfall Measuring Mission data, and gauged observation data, were used as inputs for the MIKE System Hydrological European (MIKE SHE) model. By comparing the simulation results driven by various meteorological data, the applicability of the MIKE SHE model at four stations is evaluated, and the sensitivity and uncertainty of model parameters are analyzed. Meanwhile, the impact of large hydropower stations on the runoff of the Jialing River Basin is assessed, and the influence of human activities on the runoff change is determined. The future climate change of the watershed was analyzed by using the typical representative concentration pathway (RCP) 4.5 and RCP8.5 climate scenarios. Based on the MIKE SHE model, the runoff of the Jialing River Basin in the future climate scenario is predicted, and the corresponding response of the Jialing River Basin is analyzed quantitatively. The results show that the CMAD data-driven model has better Nash–Sutcliffe efficiency and correlation coefficient for each period. By analyzing the influence of the hydropower station on the runoff process at the outlet of the basin, it is found that the hydropower station has a certain regulating effect on the runoff process at the outlet of the basin. In addition, the RCP4.5 scenario is more consistent with the future scenario, indicating that the Jialing River Basin will become colder and drier.


INTRODUCTION
Climate change triggers global temperature rises, changes in rainfall patterns, and affects the regional water cycle, as well as the hydrological situations of basins (Labat ). These hydrological changes affect the ecological environment and biological community of wetlands (van der Valk ). Therefore, it is of great significance to study the potential hydrological impact of climate change, which is becoming an increasingly popular topic of interest in the fields of hydrology and water resources (Fung et al. ; de Moura et al. ).
As a tool to study complex hydrological phenomena, the hydrological model has always been a focus of hydrology research (Dooge ). Hence, a number of distributed hydrological models have emerged, including Topography Based Hydrological Model (TOPMODEL) (Beven & Kirkby ), variable infiltration capacity (Lohmann et  In distributed hydrological models, precipitation input data are particularly important for accurate simulation results. Precipitation input data sources mainly include gauged meteorological stations and reanalysis data derived from data assimilation technology (Michaelides et al. ). The precipitation data of gauged meteorological stations are often used as the measured data, as the accuracy of the data is very high. However, there are still some problems such as uneven distribution of stations and discontinuous precipitation data. Nevertheless, the reanalysis data can compensate for these shortcomings. Therefore, it has become a common phenomenon that the reanalysis data are used as precipitation input data sources in hydrological models. At present, the commonly used reanalysis data come from the Climate Fore- The results of this study will provide reasonable and effective theoretical support for relevant policies and plans and a strong guarantee of economic development.

Study area
The Jialing River Basin (29 18 0 -34 30 0 N and 102 33 0 -109 00 0 E) originates from the southern side of the Qinling Mountains and has a total length of nearly 1,120 km. Its drainage area is approximately 160,000 km 2 , spreading across four provinces and cities, including Sichuan, Chongqing, Gansu, and Shanxi. The basin primarily includes three major water systems: the Jialing, Fujiang, and Qujiang rivers (Figure 1).
Among these, the Fujiang and Qujiang rivers merge into the mainstream of the Jialing River in Hechuan city, which is 100 km away from the urban Chongqing area.
The topography of the Jialing River Basin is complex and diverse. The northwest has high terrain with high mountains and plateaus more than 4 km, while the north is slightly lower with middle-low mountains. The center of the basin has the lowest terrain, with mainly basins and hills, and the southeast is parallel to ridges and valleys. The spatial distribution of precipitation in the basin decreases from southeast to northwest. Precipitation mainly occurs from June to September, which accounts for 66% of the total annual precipitation. The annual average precipitation is 935.2 mm, and the annual maximum and minimum precipitation are 1,283 and 643 mm, respectively.
The annual average water production of the Jialing River Basin is 6.99 × 10 10 m 3 , which accounts for approximately 17.5% of the total production of the Yangtze River Basin.  The data from 2009 to 2015 of three different data sources were selected as the meteorological data input of the model. The spatial distribution of precipitation was selected based on the station location of different data, and the Thiessen polygon was generated using ArcGIS to determine the control range of each station. Because there is no reference evapotranspiration data in the CMAD atmospheric assimilation data or the TRMM data, the reference evapotranspiration data were unified using ground observation data to facilitate a comparison.

Model calibration
The auto-calibration tool in MIKE SHE was used for a parameter sensitivity analysis and parameter calibration. In this study, 2009-2012 was selected as the calibration period, and 2013-2015 was the validation period. A sensitivity analysis of six model parameters in the hydrological model of the Jialing River Basin was conducted using the auto-calibration tool. A brief description and the parameter sensitivity test rankings are summarized in Table 2. Note that the objective function is the root-mean-square error.

GLUE method
The GLUE uncertainty evaluation method is a commonly

Future scenario analysis
The scenario analysis method assumes that a certain phenomenon or trend will continue to predict the changes that may be caused by the research object.   and P bias were determined using the following equations:   the R 2 and NSE values driven by the gauged meteorological data were slightly higher than those driven by the CMAD data, whereas for the Xiaoheba and Luoduxi stations, the performance of the gauged meteorological data was slightly worse than that of the CMAD data. The TRMM data performed slightly better than the gauged meteorological data at the Xiaoheba station, while its performance at the other stations was not as good as those driven by the other data.
During the validation period, except for the R 2 and NSE driven by the CMAD data at the Wusheng station being slightly lower than the gauged meteorological data, the performance of the CMAD data was generally better than both the gauged meteorological and TRMM data.
At the Xiaoheba and Luoduxi stations, the R 2 value driven by CMAD data was greater than 0.8, and the NSE value was above 0.6. In terms of P bias , the deviation of the model driven by the gauged meteorological site data was slightly smaller than that of the model driven by the CMAD data. Thus, the performance of the gauged meteorological data at the Beibei and Wusheng stations during the calibration and validation period outperformed the CMAD data. However, the CMAD data-driven model had NSE values above 0.5 and R 2 values greater than 0.75 at the four stations during the calibration and validation periods.
Furthermore, its performance in the validation period was significantly better than the model simulation performance driven by the gauged meteorological data. The overall performance of the TRMM data in both periods was not as good as that of the other data sources. Therefore, in terms of the model evaluation indices, the model simulation performance driven by CMAD data was generally more stable.
In addition, Figure 2 Therefore, the simulation results based on the TRMM data were the worst, which may be because the TRMM

Uncertainty analysis of MIKE SHE
In this study, the Monte Carlo sampling method was used to simulate 30,000 groups of uniform distribution parameters that were randomly selected from the prior distribution range of six parameters. The NSE was selected as the objective likelihood function, with a threshold set to greater than 0.5. When the NSE was used as the objective function, 7,243 sets of parameters met the threshold condition. These parameters were called behavioral parameters. Figure Figure 6. We found that four hydropower stations had a certain effect called 'peak shaving and dry supplement' on the runoff process at the basin outlet. The influence of the Dongxiguan hydropower station was the most distinct, which may be because it is closer to the outlet, thereby strengthening its impact.
The impact of the hydropower stations on the outlet runoff of the basin was analyzed using the non-uniform  coefficient of runoff annual distribution C vy , the distribution adjustment coefficient C r , and the relative change range C m .
These parameters were calculated as follows: where r i is the monthly runoff during the year, r is the average monthly runoff during the year, Q max and Q min are the maximum and minimum monthly values of average runoff, respectively, and C vy and C r reflect the annual distribution of non-uniformity.
The calculation results are listed in Table  period and decreased in the second. This shows that the Baozhusi hydropower station has a certain influence on the runoff process, but not for the whole period.

Climate models and prediction methods
Herein, we used the climate data of the BNU-ESM model for prediction modeling, wherein the data of the base period were compared with the data from ground observation stations.
The average values of 13 observation points in the Jialing River Basin were obtained according to the Thiessen polygon method, and the data of 311 selected grid points were averaged, also in accordance with the Tyson polygon method.  annual precipitation, and maximum and minimum temperatures of the two data were tested for variance, exhibiting a variance greater than 0.05. Thus, there is no significant difference between the two kinds of data. Therefore, in general, the climate data calculated using the BNU-ESM data meet the prediction and credibility requirements.

Forecast of future climate change
Herein, we selected precipitation and temperature data under two climate scenarios (RCP4.5 and RCP8.5). The changes relative to the base period under each scenario were calculated, as summarized in       Figure 9 shows the correlation between annual rainfall and annual average flow for the base period and both scenarios. We found that the rainfall flow correlation in the base period was weak, but the correlation under the two climate scenarios was strong. This may be because the simulation calculation of the MIKE SHE model is discrete to the grid of watershed division, and the average rainfall of the whole basin obtained using the Tyson polygon method makes it difficult to reflect the spatial difference of rainfall during the model calculation. Figure 10 shows the changing trend of the annual average runoff and annual rainfall at the outlet of the basin, which basically conforms to the rule that runoff increases with increasing rainfall.

CONCLUSIONS
From the simulation results driven by three different meteorological data, we found that the performance of the CMAD data at the four stations was relatively stable and the performance of TRMM data was the worst.
By analyzing the influence of several large hydropower stations in the Jialing River Basin on the outlet runoff  Jiangkou hydropower station had a moderating effect on the runoff process but had little influence on the runoff variation. Meanwhile, the Baozhusi hydropower station had a certain influence on the runoff process trend, but its impact was not continuous. One shortcoming of this study is that our influence analysis only involved the 2-year runoff processes of four hydropower stations. If the data conditions permit, an analysis of more hydropower stations should be carried out over a longer period.

DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.