Changes in streamflow and sediment for a planned large reservoir in the middle Yellow River

Changes in streamflow and sediment runoffs would affect the reservoir's functional operation and the construction of soil and water conservation measures in China's Loess Plateau. In this study, the long‐term changes in streamflow and sediment were analyzed for a main stem section of the middle Yellow River where the to‐be‐built large Guxian Reservoir is to be located. Results showed that both streamflow and sediment had significant downward trends with the rates of −9.4 m3 s−1 yr−1 and −16.8 million t yr−1, respectively, during the period of 1961–2017. Using the range of variability approach, the change of streamflow regime in its postimpact period (1986–2017) was subjected to the moderate alteration, whereas the alteration of sediment regime was moderate and severe for the first (1980–1996) and second (1997–2017) postimpact periods. As an example, the attribution analyses of annual streamflow and sediment changes were conducted in a typical tributary catchment (Qingjian River) on the right bank of Guxian Reservoir. For the periods of 1980–2002 and 2003–2016, climate variability occupied the primary and secondary proportions to both streamflow and sediment reductions, respectively. Overall, human activities demonstrated the underlying contribution to the sharp declines of streamflow and sediment, accounting for 68% and 74%, respectively, during the period of 1980–2016. We suggest that, based on the operational life of warping dams (built on gully for mitigating gully erosion by raising the gully‐bed step‐by‐step), there are risks of flash flood and high sediment concentration events in the future because the streamflow/sediment‐reducing infrastructures may be damaged by extreme rainstorms and in turn become the flood and sediment amplifiers.

contributes nearly 90% of the total Yellow River sediment runoff, whereas most of the sediment runoff is produced in the coarse sandy hilly catchments of the middle Yellow River . Meanwhile, the middle Yellow River region is also one of the main sources of floods (Bai, Liu, Liang, & Liu, 2016;Li et al., 2018). Consequently, the reservoirs and large-scale soil and water conservation (SWC) projects built in the middle Yellow River region should be the key measures for streamflow and sediment regulation, such as the warping dam system. Warping dam is a dam built in a gully formed in a soil and water loss area for the purpose of creating newly arable land by silt deposition in front of the dam, decreasing gully slope, and mitigating gully erosion by raising the gully-bed step-by-step (Hu, Wu, Jayakumar, & Ajisawa, 2004). During the past 100 years, the streamflow and sediment transport in theYellow River basin have considerably decreased mainly due to the interception function of anthropogenic engineering projects (e.g., reservoirs and SWC measures; Liu, 2016). In order to maintain the low sediment concentration and control floods, new reservoirs and SWC projects are planned to build in the Loess Plateau of the middle Yellow River. Thus, it is of great significance to study the changes in streamflow and sediment in the middle reaches of Yellow River and further to quantitatively assess their attributions.
Many previous studies have conducted to study the hydrological alteration (streamflow and sediment changes) and its attribution using different methods in different catchments of the Yellow River, mostly focusing on the separation of the effects of climate variability and human activities (Cui et al., 2018;Li et al., 2014;Liang, Liu, Liu, & Song, 2013;Liu, Dai, Zhong, Li, & Wang, 2013;Wang et al., 2018;Zhang, Mu, Wen, Wang, & Gao, 2013). Gao, Mu, Wang, and Li (2011) detected the transition year of streamflow and sediment discharge series  in the 1980s and inferred that human activities occupied a dominant role in the streamflow and sediment discharge reduction with the percentages of 17.8% and 28%, respectively, and in a typical tributary (Wei River basin), the corresponding contribution rates were estimated as 82.8% and 95.6% (Gao, Geissen, Ritsema, Mu, & Wang, 2013). Wang (2014) conducted an overview of 15 recent representative studies in separating two types of effects effects of climate variability and human activity on stream discharge and suggested different methods for estimating absolute and relative magnitudes of these two type effects. Liang et al. (2015) stated that, in 14 catchments of the Loess Plateau, streamflow was more sensitive to changes in precipitation than that in potential evapotranspiration (PET), and ecological restoration contributing to streamflow reduction was much larger than that of climate change. Shi and Wang (2015) suggested that hydraulic structures may be an important cause for the sharp decrease in streamflow and sediment discharge in a catchment of middle Yellow River and further considered the impacts of the variation in rainfall patterns and the warping dam construction on hydrological regime changes. Gao et al. (2016) found that land use change and climate variability accounted for 70% and 30%, respectively, of the streamflow reduction in 17 catchments of Loess Plateau. Li et al. (2018) concluded a similar result (85.2-90.3% SWC measures vs. 9.7-14.8% climate variability) in the Wuding River basin located in the middle Yellow River. Although many case-studies have proved that anthropogenic activities accounted for more of streamflow decrease than climatic factors, there were also some exception, such as the Yan River and Beiluo River (Wu, Miao, Yang, Duan, & Zhang, 2018;Zhao et al., 2014). For the sediment change in the middle reaches of Yellow River, a study of Gao et al. (2017) showed that over 70% of sediment load reduction can be attributed to human-induced land use/cover change, whereas less than 30% was associated with climate variability. Wang et al. (2016) found that landscape engineering, terracing, and the construction of warping dams and reservoirs were the dominating factors to sediment load decrease of Yellow River and suggested that the Yellow River's sediment load would increase in the future after the storage of existing dams and reservoirs approaching their capacities. Fu et al. (2017) suggested that despite some erosion in China's Loess Plateau has been successfully controlled with the implementation of the Grain-for-Green Programme in 1999, the whole regional ecosystem remains very fragile.
Most of previous studies focused on the separation of climate variability and human activities (or land use/cover), and results supported the statement that human activities are the dominant influencing factor of the reduction of streamflow and sediment. In this study, a more detailed analysis of different postimpact periods and the discussion on possible future changes were provided. The Guxian Hydropower Station is a planned large reservoir in the middle reach of Yellow River, and currently, the development of Guxian Hydropower Station has been in the stage of feasibility study. Guxian Reservoir is the core project in the regulation system of the Yellow River water and sediment. It is a major strategic project in China and related to the long-term stability of the Huang-Huai-Hai Plain. Thus, the analysis of changes in streamflow and sediment runoffs at the Guxian River section is important. In this study, we aimed to analyze the changes in streamflow and sediment load at the Guxian Reservoir section, to calculate their alteration degrees, and further to conduct a case-study of distinguishing human and climate influences on streamflow and sediment changes in a typical tributary catchment on the right bank of Guxian Reservoir. The future changes of SWC measures' benefits were also discussed in this paper.

| Study area
This paper focused on the changes of streamflow and sediment runoffs at the Guxian River section in the middle reach of Yellow River, where a large hydropower station will be built. This river section controls 490,000 km 2 drainage area (about 65% of the whole Yellow River basin), among which more than 60,000 km 2 area is the coarse sandy hilly underlying surface of the Loess Plateau. We defined the study area as the interbasin between the upstream (Wubu) and downstream (Longmen) hydrological stations of the Guxian Reservoir site (Figure 1). The Wubu-Longmen drainage area is a main coarse sandy hilly region in the Loess Plateau and covers an area of 64,038 km 2 , whereas more than 88% area is controlled by the to-be-built Guxian Reservoir. The drainage areas upstream and downstream and the hydrological station could be defined as 'gaged area' and 'ungaged area' , respectively. In the drainage area between Guxian and Longmen stations (defined as 'Guxian-Longmen drainage area' hereafter), the Xinshihe, Dacun, and Jixian stations control the area of 1,662, 2,141, and 436 km 2 , respectively. The ungaged area between Guxian and Longmen stations is 3,409 km 2 .
The Wubu-Longmen drainage area is a typical semiarid region under the temperate continental monsoon climate. The average annual precipitation is 486 mm , decreasing from southeast to northwest. Precipitation distribution is uneven within the year, and precipitation in the flood season accounts for about 65% of its total yearly amount. The corresponding average annual streamflow is 2.718 billion m 3 , and average annual runoff depth is 42.4 mm. This region has a typical loess hilly and gully underlying surface with widely distributed loose loess, ravines, broken, and undulating ground. In the rainy season, rainstorm easily produces the high-content sediment floods. The average sediment concentration in the Wubu-Longmen drainage area reaches 126 kg m −3 , which is 39% of that at Longmen station. Thus, the Wubu-Longmen drainage area is a main sediment yielding area in the middle Yellow River region. In addition, a typical tributary on the right bank, Qingjian River basin, was selected to demonstrate a case-study on streamflow and sediment change attribution. This tributary basin covers an area of 4,078 km 2 , whereas its downstream hydrological station (Yanchuan) controls 85% of the total basin area. Its underlying surface is loess hilly and gully with the potential soil erosion area percentage >90%.
The harnessing of the Loess Plateau in the middle Yellow River basin has experienced different stages ( Figure 2). Large-scale SWC measures (including terrace and warping dam construction, afforestation, and pasture reestablishment) were implemented in the Loess Plateau and the Yellow River basin since the 1950s. In addition, China central government launched a large ecological restoration campaign in the 1990s that increases the forest and pasture lands.
Under these stages, different SWC project construction intensities would have different effects on streamflow and sediment. In the Qingjian River basin, daily precipitation from 16 rain gages were collected and processed to areal mean data series   The land use data of the Qingjian River basin as at 1978, 1996, and 2015 was determined with Landsat MSS and TM remote sensing images. The random forest supervised classification method was used in the remote sensing image interpretation. Six land use types were classified, that is, cropland, forestland, grassland, waterbody, construction land, and barren land.

| Trend test and abrupt change analysis
Annual time series of hydrometeorological variables were tested for the analysis of streamflow and sediment load of the study area. The trendfree prewhitening (TFPW) and Mann-Kendall (MK) coupled trend test framework was used to remove the autocorrelation and further analyze the trend of a time-series (Kendall, 1975;Mann, 1945;Yue, Pilon, Phinney, & Cavadias, 2002). A Z-statistic could be obtained from the TFPW-MK test to reflect the trend of a time-series, whereas a positive value of Z represents an upward (increasing) trend, and vice versa. The Z-statistic also indicates the significance of trend in a time series. The change rate could be estimated via the Sen's slope (Sen, 1968).
The climatic and human factors would affect the streamflow and sediment runoffs and produce the abrupt change points in streamflow or sediment time series. The order clustering (OC) analysis method (Li et al., 2014;Xie, Chen, Li, & Zhu, 2005) was employed to detect the change points in the streamflow and sediment time series and further was verified by the Pettitt test (Pettitt, 1979). The OC method constructs a time-variant statistic, V = {v 1 , v 2 , … , v n }, as follows: where X = {x 1 , x 2 , … , x n } is an n-sized hydrometeorological time series to be tested, the subscripts i and j are the sequence numbers (1 ≤ i ≤ n and 1 < j < n), and x j and x n− j represent the average values of two subseries of X 1 = {x 1 , x 2 , … , x j } and X 2 = {x j + 1 , x j + 2 , … , x n }, respectively.
Furthermore, the change point j 0 would be detected through the objective function: Because there may be more than one abrupt change points in the streamflow and sediment time series, especially for small-size catchments. In this study, Equation (2)  ters were calculated to characterize interannual variation of the streamflow and sediment data series. In an RVA analysis, the hydrological alteration degree of each RVA parameter is defined as follows: ranging between 0-33%, 33-67%, and 67-100% represent the low, moderate, and high alterations, respectively, for the ith indicator. Furthermore, the overall degree of hydrologic alteration could be calculated as follows (Shiau & Wu, 2004): where n is the number of RVA indicators. In this study, all RVA parameters except for the parameters of 'number of zero days' and 'base-flow index' in group number 2 were used to analyze the alteration degree of both streamflow and sediment regimes, and thus, n = 31. is the drainage area. Furthermore, streamflow or sediment yield of the Guxian-Longmen drainage area could be resolved as the constituents from its gaged and ungaged drainage areas.
For annual streamflow, it could be estimated by assuming that runoff coefficient is uniform in the gaged and ungaged areas.
where W Guxian − Longmen is annual total streamflow of the Guxian-Longmen drainage area (m 3 s −1 ), and W Jixian , W Xinshihe , and W Dacun are annual streamflow at the Jixian, Xinshihe, and Dacun stations, respectively (m 3 s −1 ). α 1 is annual runoff coefficient of Zhouchuan River basin (calculated at Jixian station), and P 1 is annual areal mean precipitation (mm) in the ungaged area of the left bank of Yellow River (A u1 = 1, 477 km 2 ). α 2 represents the average annual runoff coefficient of the total area controlled by the Xinshihe and Dacun stations, and P 2 is annual precipitation (mm) in the ungaged area of the right bank of Yellow River (A u2 = 1, 932 km 2 ). U = 1/31536 is a unit conversion variable.
Accordingly, annual sediment yield in the Guxian-Longmen drainage area, S Guxian -Longmen , could be estimated through the assumption that sediment runoff modulus uniform in the gaged and ungaged areas.
where S Jixian , S Xinshihe , and S Dacun are annual sediment load at the Jixian, Xinshihe, and Dacun stations, respectively (t), and M 1 and M 2 are annual sediment runoff modulus (t km −2 ) in the gaged area of the left and right banks of Yellow River, respectively.

| Attribution analysis methods for streamflow and sediment changes
The  (7) to separate the contributions of climatic and anthropogenic factors to the changes in streamflow or sediment load.
where ΔY climate and ΔY human are the changes in streamflow (or sediment load) caused by climate variability and human activities, respectively; Y 1, obs is the average value of observed streamflow (or sediment load) for the preimpact period, whereasY 2, obs and Y 2, cal are the observed and calculated streamflow (or sediment load) for the postimpact period, respectively. It could be expected that the uncertainty of streamflow/sediment simulation models would also introduce errors to the calculation of Equation (7). We assume the streamflow/sediment simulation models used in this study produce systematic underestimated or overestimated results for both preimpact and postimpact periods. It is on this premise that such systematic errors could be reduced as much as possible through the revision of Equation (7).
where Y 1, cal is the average value of calculated streamflow (or sediment load) in the preimpact period. This revised calculation method was first introduced by Dai (2002b) and proved to be applicable to produce more accurate results than Equation (7) in the arid/semiarid Wuding River catchment in the middle Yellow River.
In this study, monthly streamflow and sediment runoffs were calculated using the model proposed by Dai (2002a).

Base flow. A linear statistical relation between base flow and
precipitation on annual scale could be established as follows: where W b is annual base flow, and P t is annual precipitation; A reflects the storage and recharge capacity of groundwater aquifer, and B represents the correlation between base flow and precipitation.
Furthermore, monthly base flow W b, i could be estimated using the 2. Surface flow. Monthly surface flow W s could be calculated from the saturation excess and infiltration excess runoff components by establishing the exponential relations to monthly precipitation P: where D 1 is a precipitation threshold parameter, k 1 and k 2 are the model parameters related to the underlying surface condition, and m 1 and m 2 are the exponential coefficients.
Thus, the total monthly streamflow is estimated as W = W s + W b, i .
3. Hillslope sediment yield. Sediment yield on the hillslope depends on rainstorm intensity, erosion resistibility, and sand carrying capacity of surface flow. The similar exponential function forms were established for estimating monthly hillslope sediment yield: where S h is monthly hillslope sediment yield; D 2 is the threshold parameter of surface flow; and k 3 , k 4 , m 3 , and m 4 are parameters.
4. River sand scouring by base flow. Clear water of base flow could carry and transport deposition in the riverbed. Similarly, the river sand scouring by base flow could be estimated in the exponential forms: where S b is monthly river sand scouring by base flow; D 3 is the threshold parameter of base flow; and k 5 , k 6 , m 5 , and m 6 are parameters.
5. Channel sand carrying capacity. The empirical exponential relation of channel sand carrying capacity S max for a specified streamflow flux W could be as follows: where k 7 and m 7 are parameters.
6. Sediment transport. The actual monthly sediment transport S could be estimated as follows:  cause the elevation of streambed to decrease, and thus the discharge converted from measured water level in the downstream sections would be less than its 'real' value. The 'real' discharge of the downstream would certainly greater than that in the upstream sections. Figure 4b shows annual sediment load data series at the Guxian hydropower station and its upstream Wubu station for the study period. It was found that more than 60% of the sediment load at Guxian station (multiyear average of 5.41 × 10 8 t) was from the drainage basin above the Wubu station (multiyear average of 3.33 × 10 8 t). Furthermore, a coherent downward trend with 99% confidence level could be found at these two mainstream stations.

| Streamflow and sediment change analysis
The slope rates were −1.19 × 10 7 t yr −1 and −1.68 × 10 7 t yr −1 at Wubu and Guxian stations, respectively. The Wubu-Guxian area is only 11% of the cover area of the whole Guxian drainage basin but accounted for about 40% of its total sediment load. It suggests that this interval basin has very strong soil erosion characteristic and is one of the main sediment yielding areas of the Yellow River basin.
Despite the streamflow/sediment differences between Wubu and  respectively, which both were subject to the moderate alteration.

| Model setup
The streamflow and sediment runoff models described in Section 2.5 were established and tested with the data in the preimpact period with 1954-1969 and 1970-1979 for model calibration and validation, respectively. Table 1

| Attribution of streamflow and sediment decline
In the postimpact periods, annual precipitation slightly decreased by 5-8% when compared with the preimpact period (Table 2). Annual streamflow slightly decreased by 9% in the first postimpact period and considerably decreased by 42% in the second postimpact period.
All years experienced the decrease of sediment yield except for 5 years in the first postimpact period (Figure 10; Table 4 Among the human activities affecting the natural hydrological processes, the SWC measures including terra and warping dam construction, afforestation, and pasture reestablishment have contributed considerably to the reduction of streamflow and sediment in different ages. Figure 11 shows the land use maps of the Qingjian River basin in 1978, 1996, and 2015. The cropland of the basin was reduced,  suggested that the total silted capacity basically accounts for 77% of the storage capacity (Liu, 2016;Liu, Gao, & Wang, 2017). This ratio (defined as sedimentation ratio hereafter) is higher than that from the design specification. Certainly, this ratio would be slightly different in a specific catchment within the middle Yellow River region. For instance, the average sedimentation ratio for main warping dams in the Wuding River basin is 0.795 (Liu, 2016). This phenomenon is related to the structure and utilization of warping dams. The storage position corresponding to sedimentation ratio = 77% is basically the spillway floor position of the current backbone dams. As for small and medium-size warping dams, YRCC also found the same phenomenon as backbone warping dams, but with a higher sedimentation ratio of 88% based on the data of 1,640 small and medium-size warping dams in Shaanxi Province (Liu, 2016;Liu et al., 2017). This sedimentation ratio is larger than that of backbone warping dams because most of the small and medium warping dams do not have spillways.
Using the data of completion time, total storage capacity and accumulated storage capacity of warping dams from YRCC, we fitted the empirical statistical equations of sedimentation ratio for both backbone and small/medium warping dams in Figure 12. It could be used to estimate the operation life of the basin warping dam system and further analyze the future changes of benefits of the dam system.
Data of 274 backbone warping dams as of 2009 in the Qingjian River basin (as shown in Figure 1) were used to estimate the number of

| SUMMARY AND CONCLUSION
In the past half-century or more, the streamflow and sediment load of Yellow River has been greatly reduced mainly due to the water con-

FIGURE 12
Fitted relationship between average sedimentation ratio (SR) and operation life (T) of warping dams in the middle Yellow River region. Data were collected from YRCC with the backbone warping dams located in Hekouzhen-Longmen region and Jing River basin and the small/medium warping dams located in Shaanxi Province (Liu, 2016) precipitation with a slope of −0.30 mm yr −1 and a remarkable warming trend of average annual air temperature with a slope of 0.02°C yr −1 for the period of 1961-2017. Meanwhile, the Thornthwaite-based PET series showed the increasing trend at a slope of 0.45 mm yr −1 .
The annual streamflow and sediment load data were estimated from its upstream (Wubu) and downstream (Longmen)  whereas two abrupt change points (1979 and 1996) were obtained for their annual sediment series. Using the abrupt points, the entire study period could be divided into the preimpact and postimpact periods.