Impacts of land use change and climate variations on annual inflow into the Miyun Reservoir , Beijing , China

The Miyun Reservoir, the only surface water source for Beijing city, has experienced water supply decline in recent decades. Previous studies suggest that both land use change and climate contribute to the changes of water supply in this critical watershed. However, the specific causes of the decline in the Miyun Reservoir are debatable under a non-stationary climate in the past 4 decades. The central objective of this study was to quantify the separate and collective contributions of land use change and climate variability to the decreasing inflow into the Miyun Reservoir during 1961–2008. Different from previous studies on this watershed, we used a comprehensive approach to quantify the timing of changes in hydrology and associated environmental variables using the long-term historical hydrometeorology and remote-sensing-based land use records. To effectively quantify the different impacts of the climate variation and land use change on streamflow during different subperiods, an annual water balance model (AWB), the climate elasticity model (CEM), and a rainfall–runoff model (RRM) were employed to conduct attribution analysis synthetically. We found a significant (p < 0.01) decrease in annual streamflow, a significant positive trend in annual potential evapotranspiration (p< 0.01), and an insignificant (p> 0.1) negative trend in annual precipitation during 1961–2008. We identified two streamflow breakpoints, 1983 and 1999, by the sequential Mann–Kendall test and double-mass curve. Climate variability alone did not explain the decrease in inflow to the Miyun Reservoir. Reduction of water yield was closely related to increase in actual evapotranspiration due to the expansion of forestland and reduction in cropland and grassland, and was likely exacerbated by increased water consumption for domestic and industrial uses in the basin. The contribution to the observed streamflow decline from land use change fell from 64–92 % during 1984–1999 to 36–58 % during 2000–2008, whereas the contribution from climate variation climbed from 8–36 % during the 1984–1999 to 42– 64 % during 2000–2008. Model uncertainty analysis further demonstrated that climate warming played a dominant role in streamflow reduction in the most recent decade (i.e., 2000s). We conclude that future climate change and variability will further challenge the water supply capacity of the Miyun Reservoir to meet water demand. A comprehensive watershed management strategy needs to consider the climate variations besides vegetation management in the study basin.


Introduction
Land use change and climate variations are two main factors directly affecting the watershed hydrological cycle.Land use change influences watershed water yield by changing canopy interception, soil properties, biophysical factors affecting evapotranspiration, and groundwater use, whilst climate variations alter precipitation, air temperature, humidity, plant growth, and consequently the hydrologic balances (Baker and Miller, 2013;Wang et al., 2013).Meanwhile, interactions of land use change and climate variations are complex, and understanding the individual effects on watershed water yield is of great importance for land use planning and water resource management (Zheng et al., 2013).To optimize watershed management, it is important to assess hydrological impacts of climate variations and land use change separately and collectively (Mango et al., 2011).Artificial neural networks and the Soil Conservation Service curve number were employed to evaluate the effect of land use change on daily streamflows in western Georgia, USA (Isik et al., 2013).The Soil and Water Assessment Tool (SWAT) was also applied to assess impacts of land use and climate change on hydrologic processes in a coastal Alabama watershed in USA (Wang et al., 2014) and the Hoeya River basin, South Korea (Kim et al., 2013).A clear understanding of the driving factors benefits both hydrological model development and hydrologic assessment of global change (Wang et al., 2013).Due to the nonlinearity of streamflow response in the synchronous evolution of driving forces, it is challenging to disentangle the integrative effects of climate forcing and basin characteristics (Risbey and Entekhabi, 1996;Beguería et al., 2003;Arabi et al., 2007;Morán-Tejeda et al., 2010).Many methods have been developed for isolating the effect of land use change from climate variations on regional hydrology.These methods include the paired catchment approach (Brown et al., 2005;Zégre et al., 2010), statistical methods (Costa et al., 2003;Sun et al., 2006;Petchprayoon et al., 2010), and a hydrological model (Haverkamp et al., 2005;Mao and Cherkauer, 2009;Baker and Miller, 2013).Raymond et al. (2008) suggested that land use change and management have been more important than climate variation to increase riverine water export from the Mississippi River over the past 50 years.However, other studies considered climate change as a dominant cause of annual water yield change (Aguado et al., 1992;Christensen et al., 2004;Barnett et al., 2005;Sun et al., 2013).Thus, both land use change and climatic variation should be considered to detect causes of hydrologic change at the same time.
The Miyun Reservoir provides 70 % of the total water supply for Beijing and is the only source of surface water supply for the severely water-stressed megacity with a population of 20 million (Tang et al., 2011).Over the past half-century, streamflow into the Miyun Reservoir has shrunk drastically.Mean annual inflow into the Miyun Reservoir declined from 88.2 m 3 s −1 in the 1950s to 15.8 m 3 s −1 in the 1980s (Gao et al., 2002).Meanwhile, population in Beijing increased from 2.8 million in 1953 to 20 million in the 2000s (Liu et al., 2003).The local water consumption in the catchment is believed to be the main driving factor in addition to climate (Ma et al., 2010).Now, 18 reservoirs have been built in the catchment with a total storing capacity of 0.214 billion m 3 (Li and Li, 2008).The contradiction between increasing water demand and water shortage constrains economic and social development of the region.Therefore, water resource assessment is extremely important to develop effective management strategies.
A few studies have tried to isolate hydrological impacts of land use change from climate change on streamflow in the Miyun Reservoir catchment (MYRC) (Wang et al., 2009(Wang et al., , 2013;;Xu et al., 2009;Ma et al., 2010;Zhan et al., 2011;Bao et al., 2012a).However, conclusions varied significantly.For example, Wang et al. (2009) and Ma et al. (2010) suggested that climate impact separately accounted for about 33 and 55 % of the decrease in reservoir inflow using the distributed time-variant gain model and geomorphology-based hydrological model, respectively.The discrepancies are mainly caused by assessment methodology due to parameter uncertainty (Shen et al., 2012), diversities of structural complexity (Velázquez et al., 2013), and inconsistencies of the evaluation period (López-Moreno et al., 2011).It remains a grand challenge in watershed hydrology (especially for large basins) to seperate the hydrological effect of land use and from climate change and variability.Hence, Wei et al. (2013) indicate that a combination of two or three methods would be a robust research strategy to assess hydrological effect within a certain range.
This study attempts to (1) detect the trend and break points of streamflow series for the period from 1961 to 2008, (2) explore an integrated approach to evaluate phased effects of climate and land use change on the inflow into the Miyun Reservoir, and (3) provide suggestions to watershed management for the studied watershed.
In this research, the relative contributions of land use change and climate variability to changes of the annual streamflow into the Miyun Reservoir were quantified using an annual water balance model based on Zhang et al. (2001), the climate elasticity model (CEM; Sankarasubramanian et al., 2001), and rainfall-runoff models (RRMs; Jones et al., 2006) for understanding water cycles and balance in the study area.Unlike previous studies that use one evaluation period, this study uses two evaluation periods to assess hydrological impact of land use change and climate variation over time.Meanwhile, three different hydrological models were combined to assess hydrological effect in each evaluation period.

Hydrometeorological data
Daily precipitation data recorded at 37 rainfall gauges and daily discharge data of 11 hydrological stations were obtained from "Hydrological Year Book" by the China Hydrological Bureau.Daily meteorological data for the period of 1961-2008 including precipitation, air temperature (maximum, minimum, and mean), wind speed, relative humidity, and sunshine hours of seven meteorological stations (Zhang-bei, Fengning, Weichang, Zhangjiakou, Huailai, Chengde, and Beijing) were obtained from the China Administration of Meteorology.Daily E p was calculated using the Hamon method (Hamon, 1963;Lu et al., 2005) as described in Sect.2.4.1.All the hydrometeorological data were processed in accordance with international standards.Abnormal data were replaced by the values obtained from kriging interpolation using nearby weather stations.Mean hydrometeorological data for the entire catchment were all obtained by the kriging interpolation method in ArcGIS 9.3.
Average monthly temperatures from November to February were below 0 • C. Minimum monthly temperature in January was lowest at −15 • , and maximum monthly temperature in July was highest at 29 • .Precipitation (P ) in summer (June, July, and August) accounted for 68 % of annual total precipitation.In comparison, summer potential evapotranspiration (E p ) accounted for 48 % of annual totals (Fig. 3).

Detecting the break points of streamflow time series
Both the double-mass curve (Searcy and Hardison, 1960) and the sequential version of the Mann-Kendall test (Mann, 1945;Sneyers, 1975) were applied to detect the break points.The double-mass curve represents two cumulative records.A break in the curve indicates a change in the relationship between the two records that may be caused by the processing of the data (Wigbout, 1973).A non-parametric test method, the sequential version of the Mann-Kendall test, is used to detect the change point of hydrological data series: 3 1960 1965 1970 1975 1980 1985 1990 1995 2000 where r i is as follows: (2) For each comparison, the number of cases x i > x j is counted and denoted by r i .It is assumed that the statistic sequential values are random and independent.Then statistic variance (UF k ) is defined as follows: Var where E(s k ) and Var(s k ) are mean and variance of s k , respectively.Statistic variance UF k is calculated as the forward data series (UF 1 = 0).The backward sequence UB k is calculated using the same equation but in the reverse data series.A null hypothesis is accepted if the critical value (u 0.05 ) lies within ±1.96 at a significance level (α = 0.05).The positive UF k denotes an upward trend, while the reverse series denotes a downward trend.When the value of UF k exceeds the critical value (u 0.05 ), it demonstrates an upward or downward trend significantly.If there are intersections of UF k and UB k lines in the range of critical value (u 0.05 ), the first cross point is the break point.

Hydrological models for attribution analysis
In this study, climate variations primarily refer to the changes of P and E p .Due to difficulty in quantitatively describing anthropogenic effects including water withdrawal and water consumption, land use change is used as the residuals affecting streamflow (Q) in addition to climate variations following Stohlgren et al. (1998) and Ma et al. (2010).Three models were built to provide a comprehensive evaluation on streamflow decreases in the MYRC.

Annual water balance model (AWB)
To detect the influence of land use change on Q, a model was developed based on the sensitivity of land use change to actual evapotranspiration (E a ) (Zhang et al., 2001).Formulas were described as follows: V d = 216.7Vs /(T + 273.3), ( 10) where δ (mm yr −1 ) is the water storage change of the watershed which can be neglected on long time averages (Donohue et al., 2010).On a meso-scale, the watershed annual Q (mm yr −1 ) can be estimated as the difference between the P (mm yr −1 ) input and the E a (mm yr −1 ) output (Sun et al., 2005).ω is the plant-available water coefficient that varies in soil water use for transpiration.For the MYRC, ω values of different land use, as a key indicator, were estimated by a trial-and-error approach with increments of 0.1 using a computer program.f i is the percentage of land use area, in which i represents diverse landscapes: forestland, grassland, cropland, water area, residential area, and bare area.E a(tot) is the sum of E a(i) .D is the day length (h).V d is saturated vapor density at the daily average temperature (g m −3 ); K is the correction factor.T is the daily average temperature ( • ).
V s is the saturated vapor under a certain temperature (mbar).

The climate elasticity model
To quantitatively evaluate the influence of climate variation on streamflow, CEM was built.The CEM defines the proportional change of streamflow divided by the proportional change in a climate variable such as precipitation (Ma et al., 2010).The model was first developed by Schaake and Waggoner (1990) to evaluate the sensitivity of streamflow to climate changes and then employed widely to assess the climate variability impact (Sankarasubramanian et al., 2001;Jones et al., 2006;Fu et al., 2007;Bao et al., 2012b): where ε 1 and ε 2 are elasticity coefficients for P (mm yr −1 ) and E o (mm yr −1 ), respectively, which are estimated by least-square estimation with Matlab7.0.Q 0 (mm yr −1 ), P (mm yr −1 ), and E p (mm yr −1 ) refer to the mean annual Q, P , and E p in the reference period, respectively.P i and E p(i) are the change of annual P and E p compared to P and E p , respectively.Annual Q (mm yr −1 ) for the period of 1984-1999 and 2000-2008 can be derived from Eq. ( 12) and calculated into mean value (Q e ).dQ clim is the average change in Q caused by climate impact.dQ land is the average change in Q caused by land use change, and dQ tot is the average change in Q between the reference period and evaluation period.O e and Q e are the average annual Q observed and simulated during the evaluation periods, respectively.

Rainfall-runoff model
In addition to the CEM method discussed in SEct.2.4.2, the impact of climate variability on streamflow was also estimated using the following empirical rainfall-runoff models (Jones et al., 2006;Li et al., 2007).
Here, Q i (mm yr −1 ) and P i (mm yr −1 ) are the annual observed streamflow and precipitation, respectively.σ 2 i is the variance of the monthly precipitation; a, b, and c are constants determined by hydrometeorological data in the reference period.Q e (mm yr −1 ) and Q r (mm yr −1 ) are the simulated mean annual streamflow during the evaluation period and reference period, respectively.3 Results

Evolution and break points of annual streamflow series
As described in Fig. 4, a significant decreasing trend at the rate of 0.96 mm yr −1 was observed for annual streamflow during 1961-2008 (p < 0.01).Simultaneously, E p increased by 1.25 mm yr −1 significantly (p < 0.01), and precipitation decreased by 0.45 mm yr −1 insignificantly (p > 0.1) (Fig. 4).
In the Chao River basin and Bai River basin, break points occurred in different years according to different methods.
Using the ordered clustering analysis method (Xie et al., 2005), one break point at 1979 was detected in the runoff record in the river basins (Wang et al., 2009).Yang and Tian (2009) found that abrupt changes in runoff occurred in 1983 and 1980 for the Chao River basin and Bai River basin, respectively, based on the sequential Mann-Kendall test.Owing to significantly increasing direct water abstraction from upstream of the reservoir since 1984, two subperiods, one from 1956 to 1983 and the other from 1984 to 2005, were detected for the Chao and Bai River basins (Ma et al., 2010).Tang et al. (2011) noted that soil conservation practice positively affected the intensified reduction of streamflow after 1999.In this study, the year 1984, as intersection point of the UF k and UB k curves inside the dotted lines, was the break point.In addition, changes in streamflow from 2000 to 2008 were more significant because points of the curves fall outside the dotted lines (Fig. 5).Further- more, the double-mass curve was also used to divide annual streamflow series into three phases (Fig. 6).Combining sequential Mann-Kendall test analysis with the double-mass curve test, we determined the reference period , the evaluation period I (1984-1999), and the evaluation period II (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008) in the MYRC.

AWB model
A total of 41 sub-catchments with different land use composition were used to build the model.According to plantavailable water coefficient w of different land use in the AWB model, the catchments were composed of forestland, grassland/cropland, water area, and residential/bare area.Forestland accounts for more than 50 % of the whole area in the DG, DY, XH, YZSK, SDY, XB, and ZJF catchments; more than 80 % of the total landmass in the BCZ, HMC, and HH watersheds; and 100 % of total area in the TM catchment (Fig. 7).The model was calibrated with the data prior to 2001 and was validated with the data after 2001 (Fig. 8).The range of w values was determined to be [0, 3] for forestland, [0, 2] for grassland/cropland, and [0, 1] for residential area/bare area.The E a of water area was assumed to be smaller between P and E p .Based on the method of trial and error, w values of grassland/farmland, forestland, and residential area/bare area were identified as 1.5, 2.8, and 0 during the calibrated period, respectively.When the average annual water balance residual E a = P − Q was compared with that estimated using Eqs.( 7) and ( 8), the determination coefficients were 0.803 and 0.783 during calibration period and validation period, respectively (Fig. 8).
Compared to the reference period , annual observed streamflow for 1984-1999 and 2000-2008 reduced by 18.1 and 39.7 mm, respectively.Using the land use data in 1988, the model was applied to evaluation periods.The difference of observed value and simulated value represented the impacts of land use change on inflow declines.As shown in Table 1, dQ land was −11.5 and −19.6 mm, which contributed 64 and 49 % of dQ tot for evaluation period I and II, respectively.

CEM model
Based on Eq. ( 12) and data in the period of 1961-1983, ε 1 and ε 2 were separately set as 2.12 and −2.2000-2008, respectively.The contribution of climate variation to the decrease of inflow during these two periods is about 1.5 mm (8 %) and 16.5 mm (42 %), respectively.Correspondingly, land use change contributed 16.6 mm (92 %) and 23.2 mm (58 %) to the decrease of inflow (Table 1).

RRM model
Using annual P and the variance of the monthly P from 1961 to 1983, the values of a, b, and c were obtained as 0.85, 0.0004, and 0.74 from Eq. ( 16), respectively.Then annual inflow into the reservoir was simulated as 56.4 and 33.8 mm for evaluation period I and II, respectively.Derived from Eq. ( 17), climate variation constituted 2.7 mm (15 %) and 25.3 mm (64 %) of total Q decrease for these two periods (Table 1).Compared to estimations from the CEM model, the contribution of climate variations to the decrease of inflow was about 7 % higher during the period of 1984-1999, but 22 % lower during the period of 2000-2008.
Table 1.Estimations on the contribution of land use change and climate variability to streamflow decreasing.The numbers directly following the ± signs are the standard deviation.
The numbers in brackets represent the contribution percentage (mm yr −1 ). Period

Data limitation and likely impact of other human factors on streamflow
This study spans multiple years and uses multiple data sources for land use, meteorology, and hydrology.Data bias often exists in field measurements, inventory, aggregation, and spatial analysis of long-series spatiotemporal data (Kavetski et al., 2006;Verburg et al., 2011).In the process of building the annual water balance model, 30 land use scenarios were utilized to calibrate the model and 11 land use scenarios were employed to verify it.To some extent, land use images were not comparable because the data were interpreted from different days of a year.Meanwhile, interpretation of remote-sensing imagery also increases the possibility of the errors.Only 37 rainfall gauges and 7 meteorological stations were available to clarify spatial change of precipitation and air temperature for a mountainous catchment with a drainage area of 15 380 km 2 ; thus interpolation errors may exist Since the 1980s, water uses in the MYRC have been intensified due to the increased water demand by people (Bao et al., 2012a).On one hand, due to the growth of population (Fig. 2) and development of industry and agriculture, the annual direct water withdrawal from the MYRC increased from 2.2 mm yr −1 in 1956-1983 to 13.4 mm yr −1 in 1984-2005 (Ma et al., 2010).At the same time, daily water consumption per capita accrued from 0.03 m 3 in 1959 to more than 0.20 m 3 in 2000 (Gao et al., 2002).Population growth aggravates water scarcity because it reduces per capita availability even with unchanged water resources (Schewe et al., 2014).Meanwhile, soil and water conservation projects have been implemented considerably with slopes transformed into terraces, and the construction of silt retention dams and reservoirs in the 1970s and 1980s (Chaobai River Management Bureau of Beijing, 2004;Chang et al., 2015).For example, the Yunzhou Reservoir (113.7 million m 3 ) and Baihebao Reservoir (90.6 million m 3 ) were built in 1970 and 1983, respectively (Hydrological yearbook of the people's Republic of China, 1991).In addition to water consumption, these water control projects enhanced evaporation and leakage losses from the catchment (Gao et al., 2013).Consequently, total water loss from the catchment had increased since the 1980s.In recent years, the Paddy to Dry Land Project and programs of closing water-based industries have been carried out to reduce water consumption that might have compensated the streamflow decline trend and have improved water quality (Wang, 2010).

Model uncertainties
Three different approaches were used to isolate hydrological impacts of land use change from those of climate change.The AWB offered a direct approach to evaluate hydrolog-ical impacts of land use change (Zhang and Wang, 2007).E a , as the predominant part of the water cycle, is the key to build this model.It is attributed primarily to land use and also affected by several other factors such as soil types and topographic slope (Moiwo et al., 2010).The daily E a (mm day −1 ) might be improved by the Surface Energy Balance Algorithm for Land (SEBAL), remote-sensing-based models validated by the Penman-Monteith approach, and the SWAT model (Gao and Long, 2008;Gao et al., 2008).The Penman-Monteith method is commonly considered as the best way to estimate the value of E p .However, the application was difficult due to insufficient climate data, especially variables about solar radiation.Therefore, the Hamon method (Hamon, 1963) recommended by the Food and Agriculture Organization of the United Nations (FAO) was used to calculate E p .The model parameter (ω) had been derived from numerous catchments (Zhang et al., 2001).Then a simple two-parameter model based on these coefficients was applied to many other catchments (Sun et al., 2005;Ma et al., 2008;Zhang et al., 2008).Our research specified an analytical expression to determine the value of 2.8 and 1.5, respectively, for forestland and grassland/cropland with a correlation coefficient of 0.78 and 0.80 during calibration and validation phases, respectively.The data errors combined with uncertainty of model structure increased uncertainty to attribution of land use change.Meanwhile, to detect the potential streamflow response of land use change in the MYRC, the model adopted the land use data in 1988 to estimate streamflow since 1984, which may cause errors due to variation of land use from 1984 to 1988.Besides, spatial and temporal variations of land use also affected streamflow change (Donohue et al., 2011;Roderick and Farquhar, 2011).In the model, recharge to groundwater and change of soil water storage might be ignored for water balance in a meso-scale catchment (Sun et al., 2005).Moreover, uncertainty of the model would be exaggerated when applied to small catchments, such as the BCZ catchment (65.2 km 2 ) and the TM catchment (3.4 km 2 ).
In the CEM, P and E p were employed to assess hydrological impacts of climate variation.Annual P in the evaluation period I was 9 mm yr −1 more than that in the reference period.Simultaneously, E p in the evaluation period I was 25 mm yr −1 more than that in the reference period.dQ clim was only −1.5 mm yr −1 , which indicated that the Q increment as the result of the P increment was slightly less than Q reduction as the result of the E p increment.
As a quantitative assessment on hydrological impacts of climate change, without spatial input, especially for the catchment area of 15 380 km 2 with altitude range from 50 to 2292 m (Fig. 1), the climate elasticity model lacks physical mechanisms and ignores the spatial details of the impact of climate variation (Yang et al., 2014a).The relative error increases with a median of 3.0 % and a maximum of 20 % with a 10 % precipitation alteration in moutains in China (Yang et al., 2014b).
The RRM accounts for rainfall as the only climatic indicator to estimate the impact of climate change.This simplification might be the main reason resulting in the differences from other two approaches.P for 1984-1999 was 9 mm yr −1 greater than that for 1961-1983, while dQ clim was 2.7 mm yr −1 smaller correspondingly (Table 1), which illustrated that the variance of the monthly precipitation played an important role in modeling streamflow besides annual P .Moreover, the watershed in the Miyun Reservior was characterized by thin soils (< 30 cm) in a rocky mountain environment (He et al., 2010).Therefore, the watershed is rather responsive to rainfall events.

Implications to water resources management
In the Miyun Reservoir catchment, forestlands accounted for more than half of the total area.Compared to 1978, forestland area increased by 5. 0 % in 1988, 16.3 % in 1998, and 18.2 % in 2008, whereas cropland decreased by 6.6, 8.7, and 10.8 %, respectively.Meanwhile, grassland area increased from 16.5 % in 197816.5 % in to 18.5 % in 198816.5 % in , and then reduced to 10.4 % in 199816.5 % in , and 9.8 % in 2008 (Fig. 9 (Fig. 9).Compared to the reference period, land use change resulted in streamflow decline for the 1984-1999 and 2000-2008 periods.It appears that the land use change effect was most pronounced during 1984-1999.Since January 1982, implementation of the household contract responsibility system has brought a huge impact on cropland and forestland.Reforestation has been widely implemented to develop forest industry and tourism especially along with implementation of Grain for Green and the Beijing-Tianjin Sandstorm Cource Control Project since the later 1990s (Zheng et al., 2010).
This study shows that the study region has experienced global warming, and climate change may increase the uncertainty of the estimated land use impact (Lauri et al., 2012).Climate change should be considered as a critical factor to optimize future water management (Gosling et al., 2011).Furthermore, anthropogenic effects, including water withdrawal and water restriction policy, could have both negative and positive effects on water supply to the Miyun Reservoir.Monitoring and objectively evaluating spatial and temporal variation of water resources are the prerequisites for water resource planning.Land use could also offset the negative effects of climate variation.For example, paddy-to-dry-land conversion in the study basin is considered as an effective means to increase inflow into the Miyun Reservoir.Moreover, artificial forest plantations widely implemented during the last 30 years are considered to aggravate water stress in this semi-arid region (Wang et al., 2012).More local vegetation rather than man-made forests with exotic tree species should be established to achieve the desired hydrological functioning of the MYRC.At the same time, proper allocation of water resources, such as water demand mangement,  1978, 1988, 1998, and 2008.can play an important role in solving water crises.In summary, comprehensive measures including vegetation restoration and water allocation are necessary to deal with water shortages facing the MYRC.

Conclusions
The comprehensive modeling approach developed by this study offers insights into the hydrological changes in the Miyun Reserior, which has experienced a significant decreasing trend of streamflow in the past 3 decades due to a combination of changes in land use and climate.The dramatic change of land use in the 1980s and 1990s due to expansion of forestland and reduction of cropland has exacerbated streamflow decline by boosting catchment evapotransiration.Climate change during the 1990s-2000s has resulted in an increase in air temperature and thus poential evapotransiration, resulting in an increase in total water loss from the study basin.Land use change dominated the streamflow decline in the 1980s-1990s, but climate change contributed most to the water supply decline in the 2000s.We conclude that future climate change must be considered in designing watershed management strategies including reforesation efforts to protect water quality and to reduce soil erosion in the Miyun Reservoir to meet the increasing water supply demand of the megacity of Beijing.Active land use management such as converting marginal croplands to natural grasslands, planting local species rather than exotic species, and water resources management such as irrigation or industry water uses should be optimized to adapt to future climate changes to sustain the water supply functions of the Miyun Reservior.Future studies should focus on sceanario analysis to examine the trade-offs of water management options in terms of hydrologic impacts under future climate change conditions.

Figure 2 .
Figure 2. Change in the population of four main counties located in Hebei Province from 1961 to 2007.

Figure 4 .
Figure 4. Evolution of streamflow (Q), precipitation (P ), and potential evapotranspiration (E p ) of the Miyun Reservoir catchment over 1961-2008.The dashed lines are the fitted linear trend for variables.

Figure 5 .
Figure 5.The sequential Mann-Kendall test for annual streamflow in the Miyun Reservoir catchment with forward-trend UF k (solid line) and backward-trend UB k (dotted line).Dashed bold horizontal lines represent critical values at the 95 % confidence level.

Figure 6 .Figure 7 .
Figure 6.The double-mass curve showing the relations between cumulative streamflow and cumulative precipitation for the Miyun Reservoir catchment (1961-2008).

Figure 8 .
Figure 8. Scatterplots of calculated evapotranspiration using Eqs.(7) and (8) against E a = P − Q during the calibration phase (a) and validation period (b).The thin line is the 1 : 1 line, and the bold line is the line of best fit provided by the equation.
et al.: Impacts of land use change and climate variations on annual inflow into the Miyun Reservoir 1569