Streamflow into Beijing and Its Response to Climate Change and Human Activities over the Period 1956–2016

Abstract

Streamflow is likely affected by climate change and human activities. In this study, hydro-meteorological data from six rivers upstream of Beijing, namely, the Yongdinghe, Baihe, Heihe, Chaohe, Juhe, and Jumahe Rivers, were analyzed to quantify the spatial and temporal variability of streamflow and their responses to climate change and human activities over the period of 1956–2016. The Mann–Kendall test and moving t-test were used to detect trends and changing points of the annual streamflow. Results showed that the streamflow into Beijing experienced a statistically significant downward trend (p < 0.05), abruptly changing after the early 1980s, owing to climate and human effects. The climate elasticities of the streamflow showed that a 10% decrease in precipitation would result in a 24.5% decrease in total streamflow, whereas a 10% decrease in potential evapotranspiration would induce a 37.7% increase in total streamflow. Human activities accounted for 87% of the reduction in total streamflow, whereas 13% was attributed to climate change. Lastly, recommendations are provided for adaptive management of water resources at different spatial scales.

1. Introduction

Streamflow is influenced by global climate change and human activities. Changes in streamflow, especially drastic increases and decreases, have increased the stress on ecological and socio-economic systems [1,2]. In recent decades, hydrological research has been focused on identifying the driving factors behind changes in streamflow [1,3,4,5,6,7]. The findings can support sustainable water resource use and are of practical significance. Climate change, such as changes in precipitation and potential evapotranspiration, results in streamflow changes [8,9]. Human activities can alter streamflow through land use/land cover change, soil and water conservation, reservoir operation, urbanization, water withdrawal, etc. [10,11,12,13]. However, the impacts of these activities on streamflow are interconnected and complex. Therefore, quantitative investigations of the effects of climate change and human activities on streamflow have been carried out [14,15,16,17,18]. As shown in the studies [15,16,17,18], investigating the characteristics of streamflow and quantifying its driving factors are imperative.

Two main approaches, model-based and statistics-based methods, are widely used for a quantitative evaluation of the effects of climate change and human activities on streamflow [14]. The model-based method uses a hydrological model to quantify the contribution of climate change to the streamflow change. Distributed hydrological models can accurately describe hydrological processes and provide high resolution spatial information that can be analyzed using a geographical information system and remote sensing technology [19]. Modelers can quantify the contribution of climate change to streamflow change by varying the meteorological inputs, given fixed land use/land cover conditions, and they can quantify the contribution of human activities to streamflow change by varying the land use/land cover, given fixed meteorological inputs [11]. This approach may be physically sound; however, collecting and handling data requires effort and can be time-consuming. Furthermore, several studies [2,14,20,21] have shown that a modeling approach can lead to different results between different models because of uncertainty in the model structure and parameter calibration. In addition, the applicability of the hydrological model could be affected in areas of intense human activity because it is difficult to fully describe soil and water conservation, reservoir operation, and water withdrawal status.

The statistics-based method mainly relies on the Budyko hypothesis and the concept of climate elasticity. The original Budyko equations [8] were based on the water and energy balance; thus, the change of catchment water storage is assumed to be zero over a long time period. However, Cavalcante et al. [22] considered water storage in the equation. The Budyko hypothesis provides only a rough approximation of the sensitivity of streamflow to climate change; it is most useful for basins with humidity ratios > 1 and for basins in which the moisture and energy balance are in phase [23]. Ma et al. [24] analyzed eight catchments in the Shiyang River basin in northwest China to investigate changes in annual streamflow over the past 50 years based on the Budyko hypothesis. They estimated that climate change accounted for over 64% of the reduction in the mean annual streamflow, owing to decreased precipitation, while the effect of human activities was limited. Tan and Gan [25] applied the Budyko framework to data from 96 Canadian watersheds and found that climate change increased the mean annual streamflow, while human activities decreased the mean annual streamflow.

The climate elasticity method proposed by Schaake [26] is a widely used approach for investigating the impacts of climate change on streamflow [20,23,27,28,29]. Sankarasubramanian et al. [23] proposed a nonparametric approach, median descriptive statistics, to estimate the climate elasticity directly from the observed data. They also found that the nonparametric estimator had a low bias and was as robust as, or more robust than, watershed model-based approaches for evaluating the sensitivity of long-term streamflow to climate change. Chiew [20] concluded that the nonparametric estimator was simple to use and estimated the elasticity directly from the historical data. Fu et al. [28] extended the nonparametric estimator to two-parameter climate elasticity to assess the climate change effect on annual streamflow. Their application indicated that the modified method could reflect the complicated non-linear relationship among streamflow, precipitation, and temperature and was appropriate to assess the climatic effect on annual streamflow of future climate change scenarios. The nonparametric approach does not require a model assumption or a calibration strategy and can be easily applied in comparative studies to data sets from numerous catchments across large regions [20,23]. This method requires a small amount of data, usually basic meteorological and hydrological data [30]. However, the climate elasticity method still demonstrates uncertainty. For instance, the framework used to quantify the impacts on streamflow are based on the assumption that climate change and human activities are independent [2,30]. In reality, however, the two factors interact, even during the baseline period. The detection of changing points involves subjective judgment, providing an additional source of uncertainty. Recently, many studies have suggested that climate elasticity can also be used to quantify and separate the contributions of climate change and human activities to streamflow, requiring less data than other methods and no parameter adjustments [9,14,30,31,32,33,34].

The attribution analysis could be an important reference for adaptive management of local water resources, especially in areas with decreasing streamflow. Beijing, located in the middle of the Haihe River basin, is the national political, cultural, international communication, and science and technology innovation center of China. The per capita water availability in Beijing is 119 m³, which is only 1/20 of the per capita water availability in China and 1/80 of the average per capita water availability in the world. Beijing is transitioning to a multi-source water supply system to ensure water security, and surface water is an important part of the system. Based on the results of the second water resources investigation and evaluation in Beijing [35], the mean annual amount of river water flowing into Beijing is 2.1 billion m³, which is 1.2 times the amount of water generated by local precipitation in Beijing. In the past 30 years, streamflow into Beijing has decreased, and the water storage and ecosystem health of Beijing have been affected. However, the spatial and temporal changes of streamflow have not been systematically analyzed, and their causes are not well understood. Consequently, the upstream area of Beijing is chosen as our study area. As mentioned above, each analysis method has merits and demerits. Considering the natural condition and available data of the study area, the climate elasticity method is used for the attribution analysis.

The objectives of this study are: (1) to detect statistically significant trends and changing points in the annual streamflow into Beijing, (2) to quantify the impacts of climate change and human activities on the streamflow changes into Beijing, by doing (1) and (2), (3) provide a reference for adaptive management of the water resources at different spatial scales.

2. Study Area and Data

2.1. Study Area

Beijing is located between 39°26′–41°04′ N and 115°20′–117°30′ E in the middle reaches of the Haihe River basin, in the northwest part of the North China Plain. There are seven major rivers within Beijing: the Yongdinghe River (Yongdinghe River catchment), Baihe River (Chaobaihe River catchment), Heihe River (Chaobaihe River catchment), Chaohe River (Chaobaihe River catchment), Juhe River (Juhe River catchment), Beiyunhe River (Beiyunhe River catchment), and Jumahe River (Jumahe River catchment). Each is a tributary of the Haihe River. With the exception of the Beiyunhe River, which originates from the Jundu Mountain in Beijing, the other six rivers flow into the territory of Beijing. Therefore, these six rivers that belong to four catchments were chosen for analysis in this study. The Baihe, Heihe, Chaohe, Juhe, and Jumahe Rivers originate from the Hebei Province, and the Yongdinghe River originates from the Shanxi Province and Inner Mongolia Autonomous Region. The total study area is 58,194 km². The Yongdinghe, Chaobaihe, Juhe, and Jumahe River catchments account for 69%, 20%, 8%, and 3% of the study area, respectively.

The study area is affected by human activities, and there are severe ecological problems. With economic development, urbanization, and population growth, demands for agricultural, industrial, and domestic water use have significantly increased since China’s land reform and opening-up policy. In addition, land use/land cover change alters the hydrological processes. Major land use types in the study area are cropland (38%), grassland (29%), forest land (28%), and urban and built-up land (3%).

2.2. Data Sources and Processing

Annual streamflow data at the Bahaoqiao, Xiabao, Sandaoying, Gubeikou, Sangyuan, and Dashadi stations were collected from the national hydrologic annals, published by the Hydrological Bureau of the Ministry of Water Resources. The total streamflow into Beijing consists of the streamflow of the Yongdinghe, Baihe, Heihe, Chaohe, Juhe, and Jumahe Rivers. Because the Xiabao (Baihe River), Sandaoying (Heihe River), Gubeikou (Chaohe River), and Sangyuan (Juhe River) stations are not close to the territory of Beijing, the streamflow into Beijing should be modified with a conversion coefficient considering the drainage area. The Beijing Water Authority [35] calculated the total streamflow into Beijing as

$$Q_T=Q_{BHQ}+1.414\times \left(Q_{XB}+Q_{SDY}+Q_{GBK}\right)+1.307\times Q_{SY}+Q_{DSD}$$

(1)

where Q_T is the total streamflow into Beijing; Q_BHQ is the streamflow at the Bahaoqiao station on the Yongdinghe River; Q_XB is the streamflow at the Xiabao station on the Baihe River; Q_SDY is the streamflow at the Sandaoying station on the Heihe River; Q_GBK is the streamflow at the Gubeikou station on the Chaohe River; Q_SY is the streamflow at the Sangyuan station on the Juhe River; and Q_DSD is the streamflow at the Dashadi station on the Jumahe River.

Daily precipitation records and other meteorological data, including daily mean temperature, wind speed, sunshine duration, vapor pressure, and relative humidity, from 86 national meteorological stations within and around the study area were obtained from the National Meteorological Information Center of the China Meteorological Administration (CMA). The Thiessen polygon method was used to calculate the average precipitation and potential evapotranspiration for each catchment. Potential evapotranspiration represents the integrated effect of climate variables [15]. Based on the meteorological data, potential evapotranspiration was estimated using the Penman-Monteith equation recommended by the Food and Agriculture Organization of the United Nations (FAO) [36]. The Penman-Monteith equation calculates potential evapotranspiration with respect to a uniform reference surface, making it convenient to compare the potential evapotranspiration of different locations. Thus, potential evapotranspiration is calculated as

$$E{T}_p=\frac{0.408∆\left(R_n-G\right)+\gamma \frac{900}{T_{mean}+273}{U}_2\left(e_s-e_a\right)}{∆+\gamma \left(1+0.34U_2\right)}$$

(2)

where ET_p is the daily potential evapotranspiration (mm day⁻¹); R_n is the net radiation at the crop surface (MJ m⁻²day⁻¹); G is the soil heat flux (MJ m⁻²day⁻¹); T_mean is the average daily air temperature at a 2 m height (°C); U₂ is the wind speed at a 2 m height (m s^-1); e_s is the saturation vapor pressure (kPa); e_a is the actual vapor pressure (kPa); (e_s- e_a) is the saturation vapor pressure deficit (kPa); Δ is the slope of the saturated vapor pressure curve (kPa °C⁻¹); and γ is the psychrometric constant (kPa °C⁻¹).

Data over the period 1956–2016 were obtained. Figure 1 shows the locations of the catchments and hydro-meteorological stations.

3. Methodology

3.1. Detection of Trends and Changing Points

The nonparametric Mann–Kendall test developed by Mann [37] and Kendall [38] is used to detect the significance of the trends and changing points in the hydro-meteorological time series. This widely used method has the following advantages: (1) it is suitable for non-normally distributed data and censored data that are frequently encountered in a hydro-meteorological time series [39], and (2) it is robust to the presence of a small amount of abnormal data [40].

For the given time series X(x₁, x₂, . . . ,x_n), the statistic S is defined as follows.

$$S={\displaystyle \sum }_{i=1}^{n-1}{\displaystyle \sum }_{j=i+1}^{n}sgn\left(x_j-x_i\right)$$

(3)

where

$$sgn\left(x_j-x_i\right)\{\begin{array}{ll}1& x_j>x_i\\ 0& x_j=x_i\\ -1& x_j<x_i\end{array}$$

(4)

Mann [37] and Kendall [38] showed that the statistic S is approximately normally distributed, with the expected value and variance given by the following.

$$E\left(S\right)=0$$

(5)

$$Var\left(S\right)=\frac{n\left(n-1\right)\left(2n+5\right)}{18}$$

(6)

The standardized statistic is shown below.

$$Z=\{\begin{array}{cc}\left(S-1\right)/\sqrt{Var\left(S\right)}& S>0 \\ 0& S=0\\ \left(S+1\right)/\sqrt{Var\left(S\right)}& S<0\end{array}$$

(7)

In a two-sided test, the null hypothesis (no trend) is accepted when –Z_1-α/2 ≤ Z ≤ Z_1-α/2. Here, Z < –Z_1-α/2 indicates a decreasing trend, and Z > Z_1-α/2 indicates an increasing trend, where α is the significance level [41]. In this study, α is set to 0.05.

To detect the changing points in the hydro-meteorological data X_t = (x₁, x₂,…, x_n), the sequential Mann-Kendall test is used. To apply this method, the cumulative number n_i of samples, where x_i > x_j (1 ≤ j ≤ i), should be calculated. The normally distributed statistic S_k can be calculated via the following formula.

$$S_k={\displaystyle \sum }_{i=1}^k{n}_i\left(k=2, 3, \dots , n\right)$$

(8)

The expected value and variance of the normally distributed statistic S_k are given by the following.

$$E\left(S_k\right)=k\left(k-1\right)/4$$

(9)

$$Var\left(S_k\right)=k\left(k-1\right)\left(2k+5\right)/7$$

(10)

The normalized statistic UF_k is estimated as follows:

$$U{F}_k=\frac{S_k-E\left(S_k\right)}{\sqrt{Var\left(S_k\right)}}\left(k=1, 2, \dots , n\right)$$

(11)

where UF_k is the forward sequence, and the backward sequence, UB_k, is calculated using the same equation but with a reversed series of data. A positive UF_k denotes an upward trend, while a negative UF_k denotes a downward trend. The null hypothesis (no step changing point) is rejected if any of the points in the forward sequence, UF_k, are outside the confidence interval. An intersection point of UF_k and UB_k located within the confidence interval indicates the beginning of a step changing point [42,43].

3.2. Nonparametric Estimator of Climate Elasticity

The concept of streamflow elasticity was originally introduced by Schaake [26] to evaluate the sensitivity of streamflow to climate change [14]. The climate elasticity of streamflow (ε) is defined by the proportional change in streamflow (Q) divided by the proportional change in a climatic variable, such as precipitation or potential evapotranspiration (X), and is expressed as follows. [2]

$$\epsilon =\frac{dQ/Q}{dX/X}$$

(12)

Sankarasubramanian et al. [23] proposed and defined a nonparametric estimator ε at the mean value of the climatic variable. Similar to the study by Zheng [2], an estimator of climate elasticity (ε) of streamflow is expressed as follows.

$$∆Q_i/\overline{Q}=\epsilon ·∆X_i/\overline{X}$$

(13)

Thus, the elasticity of streamflow ε can be regarded as the linear regression coefficient between ∆X_i/$\overline{X}$ and ∆Q_i/$\overline{Q}$. Then, the elasticity of streamflow can be estimated as [2]:

$$\epsilon =\frac{\overline{X}}{\overline{Q}}·\frac{{\displaystyle \sum }\left(X_i-\overline{X}\right)\left(Q_i-\overline{Q}\right)}{{\displaystyle \sum }{\left(X_i-\overline{X}\right)}^2}={\rho }_{X,Q}·C_Q/C_X$$

(14)

where ρ_X,Q is the correlation coefficient of X and Q, and C_X and C_Q are the coefficients of variation of X and Q, respectively. These can be calculated as shown below.

$${\rho }_{Q,X}=\frac{{\displaystyle \sum }\left(Q_i-\overline{Q}\right)\left(X_i-\overline{X}\right)}{\sqrt{{\displaystyle \sum }{\left(Q_i-\overline{Q}\right)}^2·{\displaystyle \sum }{\left(X_i-\overline{X}\right)}^2}}$$

(15)

$$C_Q=\frac{\sqrt{{\displaystyle \sum }{\left(Q_i-\overline{Q}\right)}^2/n}}{\overline{Q}}$$

(16)

$$C_X=\frac{\sqrt{{\displaystyle \sum }{\left(X_i-\overline{X}\right)}^2/n}}{\overline{X}}$$

(17)

3.3. Attribution of Climate Change and Human Activities on the Streamflow

For a given catchment, the streamflow can be modeled as a function of climate change (C) and human activities (H) [30]:

$$Q=f\left(C, H\right)$$

(18)

where Q represents streamflow; C represents the integrated effects of climate variables; and H represents the integrated effects of human activities. Therefore, changes in streamflow can be approximated as [14]

$$∆Q=∆Q_C+∆Q_H$$

(19)

where ΔQ is the total change in streamflow; ΔQ_C is the change in streamflow owing to climate change; and ΔQ_H is the change in streamflow caused by human activities.

The total streamflow change is determined as

$$∆Q=Q_{obs,2}-Q_{obs,1}$$

(20)

where Q_obs,1 and Q_obs,2 are the observed annual streamflow before and after the changing point, respectively.

Precipitation (P) and potential evapotranspiration (E₀) are the dominant controls on the mean annual water balance [8,44]. Changes in the mean annual precipitation and potential evapotranspiration can lead to changes in the annual streamflow, and the relationship can be calculated as [2]

$$∆Q_C=\left({\epsilon }_P·∆P/P_{obs,1}+{\epsilon }_{E_0}·∆E_0/E_{0 obs,1}\right)·Q_{obs,1}$$

(21)

where ε_P and ε_E0 are the precipitation and potential evapotranspiration elasticities of the streamflow. Here, ΔP and ΔE₀ are the changes in precipitation and potential evapotranspiration and are determined as

$$\epsilon ∆P=P_{obs,2}-P_{obs,1}$$

(22)

$$∆E_0=E_0{}_{obs,2}-E_0{}_{obs,1}$$

(23)

where P_obs,1, P_obs,2, E_0obs,1, and E_0obs,2 are the annual precipitation and potential evapotranspiration before and after the changing point.

4. Results

4.1. Trends and Changing Points of the Annual Streamflow

The Mann-Kendall test was applied to detect the trends and changing points of the annual streamflow over the period 1956–2016. The UF value of the total streamflow into Beijing has been consistently lower than −1.96 since 1968. From then onwards, the streamflow began to decline. The value of the Mann-Kendall statistic ($Z=-7.52$) indicated a statistically significant downward trend of the total streamflow into Beijing during the past six decades, as listed in

Table 1. Trends and changing points of the annual streamflow.

Station	Area (km2)	M-K test		t-test Changing Point	Detected Changing Point	Mean Annual Streamflow (Million m3)		Change Ratio (%)
		Z	Changing Point			Baseline Period	Changed Period
Total streamflow		−7.52	1982–1983	1982	1982	2830	872	−69%
Bahaoqiao	39,339	−8.32	1982–1983	1983	1983	1185	207	−83%
Xiabao	4015	−8.06	1980–1982	1983	1982	236	102	−57%
Sandaoying	1600	−3.74	1983–1988	1985	1985	138	78	−44%
Gubeikou	4701	−5.08	1994–1995	1994	1994	362	175	−52%
Sangyuan	375	−4.29	1994–1995	1996/1998	1995	83	27	−68%
Dashadi	4446	−5.99	1971–1980	1979	1979	700	208	−70%

. Figure 2a shows that the intersection point of the UF and UB curves was outside the confidence interval, based on the Mann-Kendall test. Thus, the confidence interval exceeded 95%. For more than one intersection point of the UF and UB curves, the detection of the changing point should be combined with other methods. In addition to the Mann-Kendall test, the moving t-test was also employed to further investigate the streamflow time series changing points (Figure 2h) in this study. An abrupt change in the total streamflow into Beijing was detected in 1982 by the integrated application of the Mann-Kendall test and moving t-test [45]. The detected changing points divided the hydro-meteorological time series into two periods. The first period represents the baseline period, when human impacts are assumed to be negligible, and the second period, referred to as the changed period, is associated with the influence of human activities [10,16]. The streamflow data for the baseline period were compared with data from the changed period. The streamflow into Beijing declined by 69% from the baseline period to the changed period.

The UF values for the annual streamflow at the six stations stayed below zero for most of the period. As listed in Table 1, the values of the Mann-Kendall statistic, Z, varied from −3.74 to −8.32; all of these values are lower than the critical value of −1.96. The largest absolute value of Z occurred in the Bahaoqiao station, highlighting that the annual streamflow in the Yongdinghe River has experienced the most significant downward trend during the past six decades. Figure 2b–g and Figure 2i–n show the UF, UB, and t-curves of the Bahaoqiao, Xiabao, Sandaoying, Gubeikou, Sangyuan, and Dashadi stations, respectively. Table 1 lists the detected changing points and change ratios of the mean annual streamflow. The change ratio is defined as the percent change in the mean annual streamflow before and after the changing point. The changing points at each of the stations, with the exception of the Gubeikou and Sangyuan stations, occurred in the early 1980s, consistent with the changing point of the total streamflow into Beijing. The Bahaoqiao station exhibited the largest change in streamflow, while the Yongdinghe River appeared to be the most influential with regard to the total streamflow into Beijing.

4.2. Climate Elasticity of the Streamflow

To assess the impacts of climate change on streamflow, the climate elasticity of the streamflow with respect to annual precipitation and potential evapotranspiration was calculated using Equation (14) and the annual precipitation and potential evapotranspiration data from 1956 to 2016. The elasticity values for each of the six rivers flowing into Beijing and the overall elasticity values are listed in

Table 2. Streamflow depth, precipitation, potential evapotranspiration, and climate elasticity over the different periods.

River	Hydrological Station	Data Period	Long-Term Mean Value (mm)			Climate Elasticity
			Q	P	E0	εP	εE0
Total streamflow		1956–1981	48	501	863	1.66	−3.73
		1982–2016	15	475	854	2.14	−4.03
		1956–2016	29	486	858	2.45	−3.77
Yongdinghe River	Bahaoqiao	1956–1982	30	467	890	1.18	−3.65
		1983–2016	5	439	886	2.30	−4.75
		1956–2016	16	451	888	2.19	−3.89
Baihe River	Xiabao	1956–1981	59	516	841	1.12	−1.94
		1982–2016	25	515	799	0.24	−1.81
		1956–2016	40	515	817	0.75	−0.47
Heihe River	Sandaoying	1956–1984	87	564	789	2.49	−2.36
		1985–2016	49	543	738	2.37	−1.59
		1956–2016	65	552	760	2.54	−1.27
Chaohe River	Gubeikou	1956–1993	77	596	736	2.67	−2.89
		1994–2016	37	561	712	2.51	−1.37
		1956–2016	54	576	722	2.94	−3.11
Juhe River	Sangyuan	1956–1994	222	751	786	2.39	−1.11
		1995–2016	72	675	787	4.45	−8.39
		1956–2016	168	724	786	3.05	−1.77
Jumahe River	Dashadi	1956–1978	157	662	811	2.07	−3.71
		1979–2016	47	600	815	2.93	−4.14
		1956–2016	89	623	814	3.00	−4.37

. Figure 3 shows the streamflow time series at each station over the period 1956–2016.

The mean annual total streamflow into Beijing decreased from 2830 million m³ (48 mm) during the baseline period to 872 million m³ (15 mm) during the changed period. Generally, streamflow positively correlates with precipitation and negatively correlates with potential evapotranspiration. This was observed in the elasticity values, with precipitation elasticity, ε_P, and potential evapotranspiration elasticity, ε_E0, calculated as 2.45 and −3.77, respectively, over the long-term period 1956–2016. These elasticity values indicate that a 10% decrease in precipitation is associated with a 24.5% decrease in streamflow, while a 10% decrease in potential evapotranspiration is associated with a 37.7% increase in streamflow. Thus, the total streamflow into Beijing was more sensitive to potential evapotranspiration than to precipitation.

Evaluating each river separately, the precipitation elasticity of the streamflow varied from 0.75 to 3.05, and the potential evapotranspiration elasticity of the streamflow ranged from −0.47 to −4.37. The streamflow in the Yongdinghe, Chaohe, and Jumahe Rivers was more sensitive to potential evapotranspiration than to precipitation, whereas the streamflow of the Baihe, Heihe, and Juhe Rivers was more sensitive to precipitation than to potential evapotranspiration.

4.3. Impacts of Climate Change and Human Activities on the Streamflow

Based on the general framework described in Section 3.2 and the estimated streamflow elasticities (Table 2), the impacts of climate change on the streamflow were quantified.

Table 3. Quantitative assessment of the impacts of climate change and human activities on the streamflow.

River	ΔQ (mm)	ΔP (mm)	ΔE (mm)	ΔQP (mm)	ΔQE (mm)	ΔQC (mm)	ΔQC/ΔQ	ΔQH (mm)	ΔQH/ΔQ
Total streamflow Yongdinghe River	−33 −25	−26 −28	−8 −5	−6.1 −3.9	1.8 0.6	−4.3 −3.4	13% 13%	−28.9 −21.5	87% 87%
Baihe River	−33	−1	−42	−0.1	1.4	1.3	−4%	−34.7	104%
Heihe River	−38	−22	−52	−8.4	7.2	−1.2	3%	−36.5	97%
Chaohe River	−40	−35	−24	−13.3	7.9	−5.4	14%	−34.5	86%
Juhe River	−150	−76	2	−68.8	−0.8	−69.6	46%	−80.7	54%
Jumahe River	−111	−62	4	−44.0	−3.4	−47.4	43%	−63.2	57%

lists the estimated contributions of climate change and human activities to the streamflow in the six rivers. Using Equation (21), with ε_P = 2.45 and given the observed 26 mm decrease in precipitation, the total streamflow depth should decrease by 6.1 mm. At the same time, because ε_E0 = −3.77, the observed 8 mm decrease in potential evapotranspiration should result in a 1.8 mm increase in the total streamflow depth. The cumulative effect of the changes in precipitation and potential evapotranspiration predicted by Equation (21) is a 4.3 mm decrease of the total streamflow depth or approximately 13% of the observed streamflow reduction. The remaining 28.7 mm decrease in the total streamflow is attributed to human activities based on Equation (19). Thus, human activities were estimated to have accounted for 87% of the total streamflow change.

As listed in Table 3, the combined effects of precipitation and potential evapotranspiration changes caused the streamflow to decrease in all rivers, except the Baihe River. Decreased precipitation was the major climatic factor associated with the decreasing streamflow trends in each of the rivers, except the Baihe River, where the increase in streamflow owing to decreased potential evapotranspiration outweighed the decrease in streamflow owing to decreased precipitation. The overall dominance of the precipitation change effects on the streamflow is a reasonable result because the change of the streamflow was mainly controlled by precipitation variability rather than potential evapotranspiration variability in the water-limited study region [6,46]. While precipitation had more influence than changes in potential evapotranspiration, as shown below, human activities had more influence on the streamflow than climatic factors.

As described in other studies [30,41,47,48], there is no remarkable climate change in the Haihe River basin, especially the Yongdinghe and Chaobaihe River catchments. In this study, the combined impacts of climate change and human activities caused a 56 to 978 million m³ streamflow decrease, with the Yongdinghe River experiencing the largest decline. The reduced precipitation caused decreases in the streamflow depth ranging from 0.1 mm to 68.8 mm, and the changes in potential evapotranspiration caused streamflow depth changes ranging from −3.4 mm to 7.9 mm. Therefore, climate change contributed −4% to 46% of the streamflow drop, and human activities were responsible for 54% to 104% of the streamflow decline. The contribution ratios of human activities to the streamflow change were larger than 80%, except for the Juhe and Jumahe River catchments. Human activities were the dominant cause of the streamflow decline across all catchments.

5. Discussion

5.1. Rationality of the Attribution Analysis

To reduce uncertainty of the climate elasticity method in this study, the Mann–Kendall test and moving t-test were jointly used to detect trends and changing points of the annual streamflow. An attribution analysis was performed. In this study, human activities accounted for the majority of streamflow reduction in the Yongdinghe River catchment, approximately 87%, in agreement with previous studies [41,49].

Table 4. Previous attribution studies on the streamflow change in the Chaobaihe River catchment.

Catchment	Hydrological Station	Study Period	Method	Contribution		Source
				Climate Change (%)	Human Activities (%)
Baihe	Xiabao	1956–2005	Budyko hypothesis	16.5	83.5	Xu, et al. [11]
Heihe	Sandaoying	1956–2000	Distributed time-variant gain model	30	70	Xu, et al. [11]
Chaohe	Daiying	1956–2005	Distributed time-variant gain model	22.1	77.9	Xu, et al. [11]
Baihe	-	1961–2001	Distributed time-variant gain model	29.6	70.4	Wang et al. [47]
Chaohe	-	1961–2001	Distributed time-variant gain model	31.4	68.6	Wang et al. [47]
Baihe	Zhangjiafen	1986–1998	SIMHYD model	37.5	62.5	Zhan et al. [51]
Chaohe	Daiying	1957–2000	Two-parameter model method	46	54	Wang et al. [30]
Chaohe	Daiying	1957–2000	Hydrological sensitivity analysis method	34	66	Wang et al. [30]
Chaohe	Daiying	1957–2000	Climate elasticity method	35	65	Wang et al. [30]

lists the results of previous studies that attributed the streamflow changes to climate change and human activities in the Chaobaihe River catchment. Comparing the results of this study (Table 3) with those of the previous studies listed in Table 4, our estimate of the main driving force for the streamflow reduction in the Chaobaihe River catchment was consistent with previous estimates. However, there was a difference in the contribution of human activities to the streamflow changes. We attribute these differences to methodological choices; different hydro-meteorological stations, different study periods, and different methods of analysis were used in each study. The results are reasonable and could be a reference for the adaptive management of water resources. Nevertheless, previous studies [9,32,33,34,50] performed an attribution analysis using different methods in other regions. To compare and verify our results, other methods should be evaluated to further quantify the impact of climate change and human activities on streamflow.

5.2. Streamflow Change Response to Human Activities

Downward trends of the total streamflow into Beijing have been observed since the early 1980s, despite the non-significant decreasing trends in precipitation and potential evapotranspiration. The streamflow decline implies that streamflow could be affected by human activities in addition to climate change. In this study, we assume that the primary human activities affecting the streamflow into Beijing are water withdrawal and land use/land cover change. The early 1980s marked the beginning of China’s land reform. The reform motivated farmers to manage their lands more productively. This was accompanied by rapidly increasing agricultural water use [14]. Meanwhile, with the concurrent improvement of water infrastructure, the area of effective irrigation expanded, despite a slight decrease in the cropland area. The increase in agricultural irrigation triggered high evapotranspiration, hence decreased streamflow [52]. Figure 4 shows the land use categories for the study area in the summer of 1980 and 2015.

Urban and built-up land increased by 47% in the study area, and their distributions corresponded to the regions with a higher population density and GDP (Figure 5 and Figure 6). Dense population and socio-economic development may imply high industrial and domestic demand for water resources. Increasing demands for agricultural, industrial, and domestic water use have led to over-exploitation of surface and groundwater. Withdrawal from surface water would reduce the streamflow directly. Over-exploitation of groundwater leaves a gap to be filled by surface runoff, thus reducing streamflow [10]. A series of soil and water conservation practices have been carried out in the study area since the 1980s, reducing streamflow by increasing evapotranspiration and improving water conservation [17,52]. Although land use changes, such as an increase of forest and grassland, are not distinct in Figure 4, the degree of coverage has grown. In addition, the low efficiency of water use in the study area is considered to be responsible for the sharp decline in streamflow.

Because of the lack of long-term water consumption data, population and GDP maps (Figure 5 and Figure 6) in the study area were evaluated to determine how they might drive the human impact on streamflow. Sun et al. [50] showed that the population density correlated to a higher impact of the human activities on the streamflow. Figure 5 and Figure 6 show that the GDP and population density are concentrated in the Yongdinghe and Chaobaihe River catchments. The impacts of human activities were the most intensive in these two catchments, consistent with the results in Table 3. Massive water withdrawals could considerably reduce the streamflow of the Yongdinghe and Chaobaihe River catchments. In the Juhe and Jumahe River catchments, the population was small owing to the comparatively inhospitable natural conditions. Therefore, human activities were low. Human activities contributed to only 54% and 57% of the streamflow reduction in the Juhe and Jumahe River catchments, respectively.

Cropland is associated with water withdrawals for agriculture. Yang and Tian [48] found that when farmland exceeded 25% of the catchment area, runoff in the Haihe River basin declined significantly. In this study, the cropland in the Yongdinghe River catchment occupied an area of 18149 km², accounting for 45% of the catchment area. The mean annual water withdrawal from the surface water for agricultural use was approximately 600 million m³, accounting for 80% of the total surface water withdrawal in the Yongdinghe River catchment. The cropland in the Chaobaihe River catchment occupied an area of 2724 km², accounting for 25% of the catchment area. The mean annual water withdrawal from the surface water for agricultural use was approximately 33 million m³, accounting for 90% of the total surface water withdrawal in the Chaobaihe River catchment. The Cetian Reservoir, with a water storage capacity of 580 million m³ located in the Yongdinghe River, is the only large reservoir in the study area upstream of Beijing. Reservoir storage impounds water and allows more evaporation to occur, causing the streamflow to decrease [52]. The Cetian Reservoir regulates the streamflow of the Yongdinghe River to meet a range of social, economic, and environmental demands, mainly agricultural irrigation, flood control, and ecological water demand.

5.3. Adaptive Management of Water Resources

A quantitative analysis using the climate elasticity method can enhance our understanding of the streamflow changes into Beijing and the streamflow response to climate change and human activities over time. The adaptive management of water resources should be considered at different spatial scales to achieve harmony of resources, ecology, and socio-economics. The excess emphasis placed on human water use crowds out ecological water use that might cause ecological system degradation. The ecological water demand can be estimated as the minimum amount of water to prevent the river from running dry and allow aquatic biology to survive [14]. To maintain ecosystem health, we suggest that not only water for living and production, but also ecological water demand should be guaranteed in the future. This could be achieved by a reservoir regulation strategy in the Yongdinghe River catchment. Since the early 1980s, numerous soil and water conservation measures, such as an afforestation and planting grass, have been implemented. The proportion of forest and grassland area exceeds 50% of the total area in the Chaobaihe, Jehu, and Jumahe River catchments. Conservation measures prevented the loss of water and soil erosion; however, these may be responsible for the significant reduction of streamflow. Therefore, soil and water conservation measures should be practiced more reasonably and moderately. In addition, because of low water use efficiency in the study area, enforcement of water conservation is required. Cropland accounted for 40% of the study area; thus, water infrastructure should be increased to improve irrigation water use efficiency. For regions with a high population density and socio-economic development, especially the upstream area of the Yongdinghe River catchment, there are growing demands for industrial and domestic water use. The use of water saving devices and unconventional water resources, such as rainwater resources and recycled water, should be promoted.

The streamflow into Beijing has experienced a statistically significant decline since the 1980s, impacting the city’s water supply. In recent years, Beijing has had to rely on water diversion and overexploitation of groundwater to meet water demand. Enhanced cooperation between the upstream and downstream areas of the catchment is required to achieve integrated watershed management. Among the four catchments, the Yongdinghe and Chaobaihe River catchments provide the majority of Beijing’s surface water. The drainage area of these two catchments accounts for 90% of the total study area. The mean annual streamflow of these two catchments over the period 1956–2016 was 1.2 billion m³, 72% of the total streamflow into Beijing. Compared to the other catchments, the Yongdinghe and Chaobaihe Rivers have more influence on the total streamflow into Beijing, and their limited water resources have a significant impact on sustainable water resource use and environmental protection in Beijing. Therefore, it is important to continue to investigate the impacts of climate change and human activities in these two catchments.

6. Conclusions

With the Mann-Kendall test and t-test, the total streamflow into Beijing, as well as the six rivers, experienced a statistically significant decline over the period 1956–2016. Almost all the changing points occurred in the early 1980s. The mean annual total streamflow into Beijing declined by 69% from the baseline period to the changed period. The precipitation and potential evapotranspiration elasticities of the total streamflow into Beijing were 2.45 and −3.77, respectively, indicating that the response of the total streamflow was more sensitive to potential evapotranspiration than to precipitation. The precipitation elasticity of the streamflow of each of the six rivers varied from 0.75 to 3.05, while the potential evapotranspiration elasticity ranged from −0.47 to −4.37.

Climate change and human activities accounted for 13% and 87% of the reduction in the total streamflow into Beijing, respectively. The catchments were subjected to intensive human activities owing to the social and economic developments since the 1980s. Human activities were the dominant factor affecting the streamflow decline for all catchments. However, there were considerable differences in the contribution rates of the human activities among four catchments owing to various degrees of water withdrawal, land use/land cover change, and soil and water conservation practices. Human activities accounted for over 80% of the streamflow reduction in the Yongdinghe and Chaobaihe River catchments, with a higher density of the GDP and population and a larger area of cropland than those of the other catchments. Beijing is faced with a water resource shortage; therefore, enhancing the cooperation between upstream and downstream management to achieve integrated watershed management is required.