Runoff changes and their potential links with climate variability and anthropogenic activities: a case study in the upper Huaihe River Basin, China

Climate variability and anthropogenic activities are two important factors affecting regional water resources. In northern hemisphere sub-tropical regions, affected by the combined effects of rising temperature and reduced precipitation (e.g. Chen et al. 2007), problems associated with water scarcity are becoming worse. Global warming may give rise to the intensification of regional water cycles (Huntington 2006), resulting in frequent occurrence of natural disasters such as floods (Milly & Dunne 2002) and droughts (Sheffield & Wood 2008). Concurrently, human activities in the form of cultivation, irrigation, drainage, urbanization and flood prevention measures have also induced variability in temporal and spatial distributions of water resources and flow regimes (e.g. Ren et al. 2002). Thus, regional patterns of runoff are complex under coupled impacts of climate variability and human activities.

As for China, observed annual runoff of the six large basins in the eastern regions has generally decreased during the past 50 years, with dramatic reductions observed in northern China (Zhang et al. 2007) which largely impede local economic development. On these grounds, strategies associated with alleviating contradictions between water shortage and growing human demands are highlighted and some preliminary evaluations have been performed (Li et al. 2007; Ye et al. 2013). However, due to the diversity of climatic conditions and human activities (related to local development strategy), runoff changes exhibit high spatial-temporal variability. Even among sub-catchments in a river basin, the main driving force of runoff variation may differ, such as in the Haihe River Basin (Wang et al. 2013). Hence, conducting the quantitative assessment at a local scale would be helpful to establish a wide spectrum of runoff change mechanisms.

The Huaihe River Basin is located in the north and south climate transitional zone. With a total length of 1,000 km, it flows through five provinces (Henan, Hubei, Anhui, Jiangsu and Shandong) controlling a drainage area of 270,000 km². Owing to poor management, problems of water shortage, water pollution and environmental degradation have become increasingly severe in recent decades. Given this, two different mechanisms-based methods have been adopted in this study – statistical analysis-based hydrological sensitivity analysis and process-based hydrological model simulation method – to assess the impact of climate variability and human activities on runoff changes. The catchment located in the upper Huaihe River Basin, together with its five sub-catchments, are selected as a case study. The objectives of this study are: (1) to comprehensively investigate the runoff change from a perspective of varied rainfall–runoff relationship; (2) to systematically analyze climate variability and human activities that have occurred in this region; and (3) to quantitatively estimate the individual contribution of climate change and human activities to runoff variation with two different mechanisms-based methods.

Observed daily discharges of six runoff stations (i.e. DPL, CTG, XIN, HC, XX and HB) and precipitations at 59 gauging sites are used for hydrological analysis (see Figure 1 for geographic distributions). Discharges at the six streamflow gauging stations are converted to catchment runoff by averaging the runoff amounts over the area of each catchment.

Daily meteorological variables at five standard meteorological stations from 1964 to 2010, including mean temperature (T_mean), maximum and minimum temperature (T_max and T_min), wind speed (WS), sunshine duration (SD) and relative humidity (RH), were retrieved from the China Meteorological Data Sharing Service System (http://cdc.cma.gov.cn). Areal rainfall was calculated by interpolating daily precipitations of 59 rain gauges with the inverse distance weighting method (Bartier & Keller 1996), and potential evapotranspiration (PET) was calculated using the Penman–Monteith equation recommended by FAO (Allen et al. 1998).

Socio-economic records for Xinyang City are collected from the ‘Henan Province Statistical Yearbook’. Other geographical data used in the study include: three land cover images (1 km in spatial resolution) for 1985, 1995 and 2000 provided by the Chinese Academy of Sciences; digital elevation data with a 3 arc-second (about 90 m) spatial resolution retrieved from the shuttle radar topography mission (SRTM) digital elevation model (http://srtm.csi.cgiar.org/); and soil data with a 30 arc-second (about 1 km) spatial resolution downloaded from the Harmonized World Soil Database (http://www.isric.org).

The rank-based non-parametric method, i.e. Mann‒Kendall (MK) test (Mann 1945; Kendall 1975), was employed for monotonic trend detection due to its advantages of non-restriction on data distribution and robustness against the interference of outliers. Significant trends are identified when the absolute value of the statistic Z exceeds 1.96 or 2.576, corresponding to 5% and 1% significance level, respectively.

For change point detection, the cumulative curve of precipitation and runoff method was employed. Normally, the cumulative curve of precipitation and runoff follows a straight line before the change point occurs, and the point of non-conformity is recognized when the consistency of hydrological series is broken (shown as the variation in the gradient of the cumulative curve). Moreover, the order cluster analysis method (OC) was employed to detect the change point. The long time series can be divided into two parts by the possible change point, and the point which makes the sum of the square deviation of the two divided series least is identified as the change point. The detailed procedure for this test can be found in Jiang et al. (2011). For both test methods, pre-change period and post-change period are identified on the basis of detected change points.

Therefore, the separate effects of the climate change and human activities can be obtained with this framework if or is known. In this study, quantitative assessments are implemented by calculating or with the hydrological model simulation method and hydrological sensitivity analysis method, respectively.

The physically based distributed hydrology soil and vegetation model (DHSVM) was employed for hydrological simulation. At each time step, simultaneous simulations of energy and water balances are carried out in every grid cell within the watershed, and runoffs generated on the basis of saturation and infiltration excess mechanism according to a user-defined infiltration capacity. Meanwhile, individual grid cells are hydrologically linked through a quasi-three-dimensional saturated subsurface transport scheme. A detailed description of DHSVM can be found in Wigmosta et al. (1994).

For model implementation, DHSVM was run at a 3-hour time step over a digital elevation model (DEM) with 200-m grid spacing. Daily forcings (precipitation, maximum and minimum temperature, RH and WS) are disaggregated to 3-hour intervals by procedures provided by Nijssen et al. (2001). For model performance evaluation, two statistics are employed: coefficient of efficiency (NSCE) (Nash & Sutcliffe 1970) and BIAS.

Similarly, relevant calibration of the model was conducted in the baseline period with two criteria: BIAS and correlation coefficient (CC).

The runoff is potentially influenced by the combined effects of climate variability and human activities, shown as non-stationary responses of runoff to precipitation (Zhang et al. 2011). In this section, runoff change patterns are analyzed from the perspective of the varied relationship between runoff and precipitation, including the trends of long-term series, cumulative curves and the decadal runoff coefficient (RC).

RC is calculated for the baseline and post-change periods, respectively, for each catchment. Table 3 summarizes the statistics of RC for all six catchments. For DPL, CTG, XX and HB catchments, RC regularly oscillated with the lowest value in the 1990s, indicating the least runoff generation in unit precipitation during that period. For the XIN catchment, RC increased slightly in the 1980s in comparison with the baseline period, then decreased in the following two decades (1990s and 2000s). RC in HC catchment showed unique variations with consistently higher values in the whole post-change periods than in the baseline period.

Climate variability and anthropogenic activities are regarded as two main driving forces that result in runoff changes. In this section, we analyze runoff-related climatic variables, historical land cover data, and local socio-economic indices, trying to explore their potential links with runoff changes for the whole HB catchment.

Precipitation and PET, which govern regional water and energy budgets, respectively, are two crucial factors influencing runoff patterns. Since most catchments experienced their first breakpoints during the 1970s (Figure 3), precipitation and PET in the period before 1980 are used as benchmarks, and statistics in the three changed decades (1980s, 1990s and 2000s) are compared with the benchmarks so as to detect decadal climate variability.

For the hydrological sensitivity analysis method, calibration was conducted in all six catchments during each baseline period, and Table 4 lists relevant fitted parameters and calibration results. The fitted n varied between 1.29 and 1.61, reflecting different properties of the six catchments. With sensitivity coefficients β uniformly larger than γ, it reveals that the change in runoff is more sensitive to precipitation than to PET within this region. The calibration results for all catchments are generally satisfactory according to the two criteria (BIAS and CC).

Individual impacts of climate variability and human activities on runoff are quantitatively assessed for each catchment using the hydrological model simulation method and the hydrological sensitivity analysis method, respectively, and the results are presented in Table 5. For DPL catchment, the results of both methods are comparable; both showed that climate variability was the dominant factor. The closeness of the results by the two methods indicates the effectiveness of both methods in evaluating the individual impacts of climate variability and human activities on runoff. According to results given by the hydrological sensitivity analysis method, the reduction of runoff for the entire HB catchment during the whole post-change period (1974–2010) was mainly attributed to climate variability, accounting for 60%, while human activities accounted for the remaining 40%. Similar patterns are found in another two main stream catchments, CTG and XX, for which annual runoff changes are largely determined by climate variability as well. This suggests that even though there are dozens of water conservancy constructions, their impacts on runoff changes do not exceed those of climate variability at the decadal scale. This is quite different from the adjacent basins, such as the Yellow River Basin and the Haihe River Basin, where human activities influence runoff greatly (Li et al. 2007; Wang et al. 2013). The possible reason is that precipitation in the Yellow River Basin and Haihe River Basin can hardly satisfy local water demand even in their wet years, while the HB catchment is generally self-sufficient in wet years and the demands of water diversion in the form of irrigation are not strong. But in dry years, when precipitation cannot largely meet the requirements, more water would be withdrawn through these conservancy facilities. As shown in Table 5, DPL, HC, XX and HB catchments suffered more impacts of human activities in their driest decade of the 1990s than in the wet decades (1980s and 2000s).

Human interferences in the XIN catchment are more intensive than in the other catchments. Table 5 shows that human activities contributed 98% of runoff reduction during the post-change period, while climate variability accounted for 2%. The regulation of Xiangshan Reservoir was possibly responsible for most of the runoff reduction in this catchment. With a controlling area of 72.8 km², the reservoir governs 27% of the XIN catchment area, and its storage capacity accounts for 56% of mean annual runoff. Apart from irrigation (Table 2) and hydropower generation (Figure 8), the stored water also serves to meet domestic usage, industrial consumption and ecological landscape requirements of the XIN County (more than half of its area is outside the XIN catchment). With the development of the economy in XIN County, the water demands would increase gradually. For instance, according to the local survey, the total abstraction from Xiangshan Reservoir in 2010 was up to 0.44 × 10⁸m³, equivalent to 29% of mean annual runoff of this catchment.

In contrast to the dramatic runoff reduction in the XIN catchment, the runoff of the HC catchment experienced unusual increases. For the post-change period, climate variability and human activities contributed 64% and 36% of runoff increments, respectively. The water resources allocation in this catchment is rather complex; since the Pohe Reservoir (inside the HC catchment) can hardly meet local requirements, two large reservoirs outside the catchment (i.e. Wuyue Reservoir and Nianyushan Reservoir) contribute considerable amounts of water to this region. The large irrigation canal conveys water from the Wuyue Reservoir (Figure 8), and for Nianyushan Reservoir, approximately 0.5 × 10⁸m³ of water on average was transported from the lateral channels of its western irrigation canal (assessed from local official websites, http://xxgk.xinyang.gov.cn/huangchuan/slj). Owing to the lack of detailed observations of these reservoirs, it is hard to quantify the specific effect of each one.

Although close results were derived by the hydrological model simulation and hydrological sensitivity analysis method in the DPL catchment, which indicates the feasibility of the two methods at this basin scale, the effects of modeling uncertainty in each evaluation method should not be ignored. For the hydrological model simulation method, a major concern is the rainfall–runoff simulation uncertainty introduced by heavy data requirements for the distributed modeling. As shown in Figure 10, the simulation bias of streamflow discharge is −0.81% for the calibration period and 0.84% for the validation period, equivalent to less than 3 mm of runoff. The bias is rather small when comparing with the volume of runoff changes caused by climate variability and human activities. As the bias of the simulated water amount is the major consideration in the assessment of the impacts of climate variability and anthropogenic activities, such a small bias caused by modeling uncertainty would not affect our analysis results significantly.

The hydrological sensitivity analysis method requires only meteorological inputs and therefore is more easily manipulated, though it has some distinct drawbacks such as the oversimplified modeling structure in which seasonal variability of hydrological factors is ignored. Nonetheless, this limitation does not influence the application of the hydrological sensitivity analysis method substantially. As shown in Table 5, the DPL catchment was evaluated with the two methods and the difference between each derived attribute result was less than 6% (e.g. for the 1990s, the hydrological model simulation method shows climate variability is responsible for 83% of runoff changes, while the figure is 78% for the hydrological sensitivity analysis method). The bias generally has little impact on the role of climate variability and human activities. The hydrological sensitivity analysis method also provides an alternative for the assessment in data-scarce basins where the hydrological model simulation method cannot work well.

Using meteorological data at one meteorological observation site, precipitation data at 59 gauging sites and streamflow discharge data at six gauging stations during 1964–2010, runoff changes in the upper Huaihe River basin at HB together with five sub-catchments (i.e. DPL, CTG, XIN, HC and XX) were analyzed using the method of trend detection, the precipitation and runoff cumulative curves method, and the OC analysis method. Then, the potential impacts of climate variability and human activities within HB catchment on runoff were analyzed with two quantitative evaluation methods: the DHSVM-based hydrological model simulation method; and the hydrological sensitivity analysis method. The investigation results show the following:

1. A long-time series of runoff for the whole HB catchment showed no significant trends but fluctuated considerably at annual scale. Among its five sub-catchments, except the HC catchment in which runoff even increased under the condition of decreased precipitation, annual runoff generally decreased during 1964–2010, with varying degrees of reduction. Breakpoint for the two headwater catchments (DPL and CTG) both occurred in 1990, relatively later than other catchments, for which the first breakpoints emerged in the 1970s. Decadal variation of RC indicated the varied runoff response to precipitation between the pre-change and post-change periods.

2. Generally decadal drying and wetting alternations are found in the HB catchment, with the lowest amounts observed in the 1990s for most sub-catchments. Comparing with the period before the 1980s, seasonal precipitation in the 1980s, 1990s and 2000s presented a higher variability, with considerable increments of summer precipitation in the 1980s and 2000s. Annual PET significantly decreased during 1964–2010 with certain increments in the 1990s which was possibly caused by the rising SD. These findings suggest that the 1990s was the driest decade both in water and energy conditions. Meanwhile, investigations on land cover changes and socio-economic indices imply that more water would be abstracted and consumed in the post-change period.

3. Quantitative evaluation on separated effects of climate variability and human activities showed that for the whole HB catchment, climate variability played a dominant role, and runoff changes in the other four catchments DPL, CTG, HC and XX were also mainly induced by climate variability. As for the XIN catchment, runoff change was mainly attributed to human activities, accounting for 98%. Generally, runoff in most catchments suffered more reduction in dry decades (1990s) than wet decades (1980s and 2000s) due to enhanced impacts of climate variability and human activities. These results reflected the complexity of climate variability and human activities induced impacts on runoff regime, such as differences among catchments and decades. In these upstream catchments of the Huaihe River Basin, the runoff change should be highlighted and policy makers should make reasonable management interventions to guarantee the sustainable utilization of local water resources.

The authors are grateful for the valuable suggestions from the two anonymous reviewers. The National Natural Science Foundation of China (Project No. 41371050), the Fundamental Research Funds for the Central Universities of China (Grant No. 2014B35914) and the 111 Project (B08048) are also gratefully acknowledged.