Integrated assessment of the impacts of climate variability and anthropogenic activities on river runoff: a case study in the Hutuo River Basin, China

Over the past decades, the runoff in many regions around the world has significantly decreased. Human activities and climate variability are regarded as the two driving factors for the runoff changes (e.g., Nijssen et al. 2001; Bao et al. 2012). Thus, it is very important to understand and quantitatively evaluate the effects of these factors on runoff (Jiang et al. 2012). However, the interactions between these factors have a confounding effect that makes inferring causation difficult to achieve at a sufficient scale (Bogena et al. 2005; Zhang & Schilling 2006). For instance, the conversion of forest to crop land and extensive urbanization are observed to affect the exchange of moisture and heat between terrestrial and atmospheric components and hence change local and regional surface temperatures and rainfall patterns (Wang et al. 2010; Miller & Cotter 2013). Climate change, in turn, exerts a dominant control on the predominant vegetation types, land use practices and their spatial characteristics (Brovkin 2002; Tao et al. 2015). Moreover, a definite coupling is observed between the climatic variables and land surface processes such as consumptive water use in irrigation schemes (e.g., Qian et al. 2013; Awan et al. 2016). In spite of this, most previous studies have assessed impacts of climate variability and/or human activities and associated hydrological changes in isolation (e.g., Li et al. 2009; Jiang et al. 2012; Wang et al. 2013). An implication of the lack of integrated analysis associated with these studies is likely to lead to either over- or under-estimation of the potential effects.

In recent years, a number of studies have pointed out that human activities and climate variability are inter-related with each other and are not readily separable. For example, Liu & Xia (2004), conducted a semi-quantitative analysis to explore the reason for the runoff reduction in the Haihe River Basin, and they noted that interaction between climate variability and human activities is inherent in the study basin and plays an important role in runoff decline. A similar study by Xu et al. (2013) in the Luan River Basin indicated that the sum of all contributions of the driving factors to decrease river flow in the study watershed could be less than 100% due to the cross impacts between the contributing factors. Zheng et al. (2009) studied the impact of climate and land surface changes on streamflow in the headwaters of the Yellow River Basin. They noted that the major uncertainty related to the study is the assumption that climate and land surface are independent factors. More recent studies by Wang et al. (2014), El-Khoury et al. (2015), and Yan et al. (2015) reported that the individual and combined effects of land use and climate variability on runoff are determined by how these factors are defined with different extents of interaction effects.

To assess the hydrological effects of human activities and climate variability, several methods have been used, which can be categorized into four groups: experimental paired catchment method (e.g., Scott et al. 2000), statistical methods (e.g., Tian et al. 2009), climate sensitivity analysis approach (e.g., Liu & Xia 2004), and distributed hydrological models. However, among these techniques, the distributed hydrological models are by far the most appealing tools to carry out impact assessment studies (Mango et al. 2011). These models provide a distributed framework that takes physically observable land surface characteristics and meteorological conditions into account (Refsgaard & Knudsen 1996). Among the distributed hydrological models, the basin-scale Soil and Water Assessment Tool (SWAT) is widely used for evaluating changes in hydrologic regime in response to land use/land cover alterations and climate variability (e.g., Fan et al. 2010; Du et al. 2013; Singh et al. 2015).

As a typical mountainous basin that is heavily influenced by environmental change, the Hutuo River Basin (HRB) is used in this study as a proxy to conduct integrated assessment of the impacts of climate variability and human activities on river runoff. It is located in north China, occupies an area of 15,900 km² and supports 4.29 million people. The Hutuo River is the largest tributary in the mid portions of the Haihe river catchment, which plays a key role in the sustainable development of the economy and ecology of China (Peng et al. 2012). Over the past few decades, the basin has endured influential, frequent droughts and serious water shortages causing severe environmental problems (Liu & Xia 2004). Climatic variability and human activities have negatively affected the hydrological processes and, consequently, water resources in the basin (Tian et al. 2009; Wang et al. 2013). Rapid socio-economic development has led to an increase in water demand, whereas climate variability and land use change have an impact on water availability. However, nonlinear relationships, multiple causation, lack of mechanistic understanding together limit researchers’ ability to diagnose causes (Allan 2004; McIntyre et al. 2014). In order to address and meet these challenges, integrated assessment approaches are needed which take into account the interactive effects of the driving factors on river runoff. The present study addresses this issue through systematic investigation, considering both the separated and complex-coupled natural and human influences.

The main objective of this study is to explore the separated and interactive impacts of climate variability and anthropogenic activities on river runoff over the representative HRB. To this end, a generic framework, based on a hydrological model (SWAT) and two simple approaches, is proposed. The main research work involved: (1) isolating quantitatively the impacts of three main processes, i.e., direct human activities, climate variability and land use change on runoff; (2) estimating the interactive effect of climate variability and land use change on the river runoff.

SWAT is a conceptual basin-scale, continuous-time semi-distributed hydrological model, developed by the United States Department of Agriculture (USDA) to predict the impact of land management practices on watershed hydrology, agricultural chemicals, and sediment yields (Arnold et al. 1998). SWAT divides a watershed into a number of sub-basins related to stream networks in the basin. Based on the input information, SWAT arranges in each of the sub-watershed the following categories: climate, hydrologic response units (HRUs), groundwater, ponds/wetlands, and the main reach draining each sub-watershed. Then the model captures the underlying physical processes at small lumped land areas within the sub-basin (HRUs) that comprise unique land cover, soil, slope attributes and management combinations. The distributed framework of SWAT serves to conceptualize the relations between climate variability and human activities and their synchronic effects on the hydrological process.

The required input data for the SWAT model, including topographic, climatic, soil and land use data, in this study were collected or processed as follows. Topography of the watershed was defined by digital elevation model (DEM) clipped out from the Shuttle Radar Topography Mission (SRTM) 90 m Digital Elevation Data (http://srtm.csi.cgiar.org/).

60 m Landsat Multispectral Scanner (MSS) and 30 m Thematic Mapper (TM) images for three different dates (26 June 1974, 26 May 1984, and 6 May 2000) were selected to generate land use and land cover data of the study area. The Landsat images were obtained from the Global Land Cover Facility (http://glcf.umd.edu/data/landsat/) and Earth Explorer site (http://earthexplorer.usgs.gov/). The selected images are sufficiently long enough apart from each other (10–16 years) and obtained within the same season (a maximum of 30 days from each other) to detect the potential land use changes and consequent catchment responses.

Observed metrological data at Yuanping and Wutaishan stations during 1961–2000 were collected from the China Meteorological Administration (CMA), including daily records of maximum and minimum temperature, precipitation, wind speed, sunshine duration, and relative humidity. Monthly and annual runoff data of the same period were obtained for Nanzhuang hydrological station (see Figure 1).

In this study, the nonparametric Mann–Kendall test (Mann 1945; Kendall 1975) was used to analyze the monotonic of trend of annual and monthly hydroclimatic time series i.e., precipitation, mean temperature, and observed runoff in HRB for the period 1961–2000. Pettit's (Pettitt 1979) test method was further utilized to investigate the homogeneity of the hydroclimatic data by detecting the break points in the mean of time series. These methods have been proved to be a very useful tool to quantify the significance and magnitude of change in hydroclimatic data. The methodologies of these statistical tests are broadly documented in many literatures (e.g., Partal & Kahya 2006; Salarijazi et al. 2012).

After preparing the required maps (land use, soil, DEM) and database files such as climate and soil characteristics, a new SWAT project was set up for Hutuo catchment. Through delineating the sub-watershed and creating HRU, the SWAT project can simulate the runoff of HRB. However, SWAT is a high parameterized model, and the associated uncertainty must be adjusted within a realistic range (Arnold et al. 2012). In a practical sense, a preliminary step in the model calibration process is to determine the most sensitive parameters to the input changes. In this study, the sensitivity analysis was conducted using the SWAT-CUP software (Abbaspour 2013), which merges different calibration/uncertainty analysis procedures for SWAT.

After determining the sensitive parameters, SWAT calibration was processed at monthly and yearly time steps by matching simulated runoff with the measured runoff from Nanzhuang station. The model was calibrated during the period 1962–1979 and further validated during 1980–1997. Land use maps of 1974 and 2000 were used as the land use input data to run the model for the calibration and validation period, respectively. The performance of SWAT during the calibration and validation periods was evaluated using three common statistical indices: Nash–Sutcliffe efficiency ⁠, relative error (Re), and coefficient of determination (R²) (Moriasi et al. 2007).

Based on the hydroclimatic trend analysis, the whole runoff time series in HRB during 1961–2000 was divided into two periods: natural period (1962–1971) and impacted period (1991–2000). Compared with the average runoff in the baseline period (1962–1971), runoff variability in the impacted period (1991–2000) was mainly due to three factors: direct human activities, climate variability, and land use change. In this study, the direct human activities refer to the direct withdrawal of water from river channels for irrigation, industrial and domestic usage (hereafter referred to as D.HA), while climate variability refers to changes in precipitation and temperature. To evaluate the impacts of these factors on river runoff, five scenarios were considered as follows:

• SR.A: Climate conditions of 1962–1971 and land use map for 1974

• SR.B: Climate conditions of 1962–1971 and land use map for 2000

• SR.C: Climate conditions of 1991–2000 and land use map for 1974

• SR.D: Climate conditions of 1991–2000 and land use map for 2000

• SR.E: Observed runoff during period 1991–2000.

In the first scenario (SR.A), simulation was carried out for the baseline period 1962–1971 with land use for 1974. For the following two scenarios (SR.B and SR.C), the calibrated SWAT model was run based on a one-factor approach (Li et al. 2009), i.e., changing one factor at an instant while holding the other invariant. SR.B and SR.C were constructed to quantify the individual impacts of the land use change and climate variability on runoff, comparing with SR.A. In SR.D, the simulation was carried out for 1991–2000, changing both climate and land use inputs corresponding to 1962–1971, and thus serving to estimate combined (land use + climate variability) effects comparing with SR.A. With the reconstructed runoff under SR.D and the coincident observed runoff series, it is possible to quantify the impact of D.HA, which is considered by the fifth scenario (SR.E).

Land use change in any region occurs gradually over a period of time, and thus starting and ending years could satisfactorily represent the potential alteration that has occurred in each land use category. Therefore, only the land use maps of 1974 and 2000 were considered by hydrological modeling in this study.

The above results suggested that climate in the HRB during 1961–2000 was getting warmer and drier. Compared to 1962–1971, the annual mean temperature in 1991–2000 increased by 2.5 °C, while the precipitation and runoff obviously decreased by 29% and 72%, respectively. Furthermore, winter and spring months exhibited the highest increase (temperature) and decrease trends (precipitation, runoff) among the other months.

The results suggest that SWAT performance was very good during the calibration period and satisfactory during the validation period, according to Moriasi et al. (2007). However, the simulated runoff during the validation period at a yearly time step (Figure 8(b)) was evidently greater than the measured values in most years. This could be, therefore, reliable for supporting the hypothesis (SR.E) that isolates the extent of direct human influence on runoff from the other factors based on the differences between simulated runoff (without the human modifications effect) and observed (with human modifications).

Table 6 shows the individual and combined impacts of land use and climate variability as well as the isolated impact of D.HA on annual mean runoff depth, obtained under the five scenarios. Compared with SR.A, the observed runoff during the impacted period decreased by 98.37 mm, which represented the total runoff change due to the integrated effect of D.HA, climate, and land use change. From Table 6 it can be found that the runoff reduction caused by land use change, climate, and the combined change accounted for 2.06%, 45.30%, and 47.84% of the total change (−98.37 mm), respectively. While the contrast between the observed runoff during 1991–2000 and reconstructed runoff under SR.D indicates D.HA impact, which accounted for 52.16% of the total change. It should be pointed out that the average annual runoff reconstructed for the baseline period (231.84 mm) was slightly greater than the observed runoff during the same period (230.07 mm), which indicates that there might be some human activities within HRB affecting the runoff during this period.

It should be noted that the combined impact of climate and land use change on runoff (47.84%) was greater than the sum of their individual impacts (47.36%), due to an interaction between the two factors. However, no clear inference about the implication of the climate–land use interactive effect on runoff can be achieved from the total runoff change (−98.37 mm), due to the large runoff reduction caused by D.HA.

To estimate quantitatively the climate–land use interactive effect, the proposed approach primarily requires the outputs of the scenarios SR.A, SR.B, SR.C, and SR.D. Table 7 shows the annual runoff response to isolated and interactive impacts of land use and climate variability. It can be seen that the total runoff decline due to land use and climate variability was 47.06 mm obtained by the difference between SR.A and SR.D. The single impacts of climate variability (ΔQ_C) and land use (ΔQ_L) were responsible for 94.3% (SR.A–SR.C) and 4.33% (SR.A–SR.C) of the total simulated change (47.06 mm), respectively. Based on Equation (5), the exacerbated impact of land use due to climate variability (ΔQ_L^C) was estimated by 5.33% (SR.C–SR.D) of total change. The contrast between the individual and amplified effect of land use change (Equation (6)) represents the land-climate interactive effect (ΔQ_I). The interactive effect of land use change and climate variability accounted for 1.02% of the total simulated change.

Results from Table 6 suggest that runoff decline was mainly caused by D.HA. As a result of urbanization and farmland expansion during the two sub-periods 1974–1984 and 1984–2000, the water demand by industry, domestic usage, and irrigation has intensified, subsequently resulting in a sharp runoff decline in HRB. This finding is relatively consistent with Tian et al. (2009), who suggested that the abrupt decline in HRB was driven mainly by human activities, and especially agricultural water use.

Climate variability is also a dominant driving factor for runoff decline in HRB. Human activities usually affect runoff regimes suddenly and directionally, whereas climate variability affects runoff regimes periodically and lastingly (Miao et al. 2011). The most influential human activities in HRB occurred during 1980–2000. However, runoff has had a remarkable decreasing trend since the late 1960s in conjunction with precipitation decrease (Figure 4). This may infer that climate variability has a lasting negative impact on runoff. The results obtained based on the second approach (Table 7) suggest that climate variability is the predominant contributor to runoff reduction in HRB (>90%) when the impact of D.HA is isolated. The large variability in observed runoff (−96.60 mm) corresponding to the variability in rainfall (−10.33 mm) leads to an underestimation of the potential impact of climate variability.

The results show that the annual runoff in HRB is influenced more by D.HA and climate variability compared to land use change. This finding is in line with Ma et al. (2010), who found that the indirect impact of land use change was less than direct abstraction of water and climate variability in the Miyun Reservoir catchment (a nearby catchment). Although the deforestation in HRB (−26.06%) is expected to increase surface runoff (Stednick 1996), the reconstructed runoff under impact of land use showed a significant decreasing trend. The heavily direct water abstraction most likely ameliorates the effect of deforestation on surface runoff. However, in general, the impacts of land use change on hydrological regimes is not easy to quantify because they are the result of interplay between various processes, such as climatic factors, vegetation characteristics, site-specific conditions, and management factors (Conway 2001).

The impacts’ separation approach was successfully managed to isolate the effect of D.HA, land use, and climate variability. In spite of this, a clear inference about the implications of land use change, climate variability, and the involved interactions were not provided owing to the large variability imposed by D.HA. It is hence useful to isolate the impacts of land use change and climate variability from that of D.HA.

The results from Table 6 show that the combined effect of land use and climate leads to runoff reduction greater than the sum of their individual impacts. One factor is most likely the exacerbated impact of the other factor, hence resulting in a synergistic effect. This hypothesis is further confirmed by the results from Table 7, which indicate that climate variability has clearly amplified the impact of land use by 23.67%.

Saha et al. (2012) found that evapotranspiration over land control is responsible for about 10–20% of the inter-annual variability in rainfall over central India. In this study, the calculation shows that land use change decreased the annual evapotranspiration over the basin by 2.98 mm. This constitutes about 71% of the annual rainfall decrease that has occurred during the impacted period. Moreover, as shown by Figure 10, land use change has dampened the evapotranspiration flux back to atmosphere through an antagonistic interaction estimated to be 2.10%. This possibly had an influence on cloud formation and precipitation over the basin. The temporal trend of precipitation and temperature could be influenced by the large-scale human activities in land surface, such as farmland expansion, deforestation, and urbanization (Wang et al. 2010; Tao et al. 2011; Miller & Cotter 2013). If this is also valid in HRB, the significant alterations in the precipitation and temperature time series (Figure 5) are presumably influenced by the extensive human activities and the imposed land use changes that occurred during past decades. On the other hand, the analysis presented here indicates that the grassland and farmland areas in HRB were increasing significantly (Table 1) over the study period, while the precipitation was decreasing (Figure 4). From this it may also be deduced that the rainfall decrease over the basin is most likely caused by long-term changes in land use, in particular the conversion of forest to grassland and farmland (Lambin 1997). As the canopy interception in the basin decreased with the decrease in forest, a higher decrease of evapotranspiratable water is suggested, as can be seen from Figure 10.

Further insight indicates that the driving factors to runoff decrease in HRB are also determined by modification interactions. The climate variability over HRB during 1961–2000 caused a significant increase (>50%) in the annual evapotranspiration (Figure 10). The land surface state includes variables such as soil moisture and temperature, snowpack and other surface water stores, and the state of vegetation. Of these, soil moisture is perhaps the most important (Dirmeyer & Studies 1995). As the transpiration from land surface increases with temperature, the soil wetness decreases and suggests a higher infiltration rate. Land use variations play a vital role in local water cycle changes, especially for the water movement within the soil layer (Tan et al. 2015). The changes in land use and land cover that occurred in HRB most likely caused an increase in infiltration and a decrease in surface flow and under the changed climatic conditions contributed to more runoff decline. Accordingly, the increasing (decreasing) trend of temperature (rainfall) in the basin most likely exacerbated the impact of land use change on runoff. Moreover, the high temperature increment after 1986 likely increased the irrigation water demand (Awan et al. 2016), and hence intensified the direct withdrawal of water. It can be seen from Figure 9 that the contributions of D.HA to runoff decrease in January, February, November, and December are particularly high, corresponding to the high temperature increment observed in these months (Figure 6).

Major sources of uncertainty in the simulation from the SWAT model may arise from the meteorological input data, model parameters and Landsat imagery classification and model land use inputs involved. The meteorological data were from a very limited number of stations, which might not be of sufficient coverage for a mountainous catchment area of 15,900 km². Also, there are some unavoidable errors in the calibration of model parameters. Moreover, there are still some limitations associated with assessing the interactive effect of climate variability and anthropogenic activities in the present work. In fact, the reciprocal influences between human and climate systems occur within a wide range of complex mechanisms (Brovkin 2002). However, in this study, these mechanisms were only briefly addressed. Further study should consider these uncertainties and quantify their impacts on the final analysis results.

The primary objective of this study was to explore the complexities that are associated with multi-driver analyses, by examining both the separated and interactive natural–human influences on river runoff. A generic framework-based hydrological model (SWAT) and two approaches were developed and utilized to quantitatively estimate the individual, combined, and interactive impacts of three main processes, i.e., direct human activities, climate variability, and land use change on runoff. The analysis was conducted for the HRB, for the period 1961–2000, which has been heavily influenced by environmental change.

Climate variability alone has a negative lasting effect on runoff in HRB, while direct human activities mainly cause the abrupt runoff reduction. Land use change has less effect on runoff, but it imposes significant water-related direct human activities. The combined effect of land use and climate on runoff is observed to be more pronounced compared to their individual impacts in the basin.

The analysis of the interactive effect indicated that the driving factors on river runoff are multiplicative rather than additive. Climate variability clearly exacerbates the impacts of land use change on runoff through a noticeable synergistic effect. Moreover, it could be found that the driving factors interact to impact runoff through chain effects, whereby one factor tends to increases the magnitude of another factor. Thus, the interactive effect of environmental change drivers should be given more attention when assessing the potential impacts and predicting future changes.

Despite the uncertainties involved, the results of this study provide an insight about the linear and nonlinear responses of the runoff to environmental change in the study region. Based on understanding of the interaction effects between human-induced changes in the land surface and climate system, risks and opportunities can be summarized. Risks include ineffective conservation management; but opportunities also arise by which the negative impacts of climate variability can be reduced through appropriate land management as an adaptation measure. The proposed framework could be useful for the assessment of the potential effect of adaptation measures to cope with global change particularly in regions with high climate sensitivity and rapid socio-economic development.

This work was jointly supported by the National Natural Science Foundation of China (51379057), the Fundamental Research Funds for the Central Universities (2015B14114), National ‘Ten Thousand Program’ Youth Talent, QingLan Project, and a Project funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD).