Runoff sensitivity to climate and land-use changes: A case study in the Longtan basin, Southwestern China

Under changing environments, the water cycling process and its formation characteristics are increasingly sensitive to the changes in global climate and human activities (Chenoweth et al. 2011). Atmospheric warming intensives the movement of water molecules as well as the movement of evaporation, rainfall and soil moisture. Finally, it affects the spatial and temporal distribution of water resources (Şen 2009). On the other hand, human activities affect the process of infiltration, soil moisture, and surface runoff, based on the variation in land-use patterns (Green et al. 2007). Differences between water cycle processes and water resource allocation caused by climate change and human activities are prone to induce the problems of extreme hydrometeorological events (droughts and floods) and water pollution (Wu & Tan 2012). Therefore, related subjects and studies of climate change (at regional or global scales) and the changes in water resources have become issues of increasing concern to experts and scholars in recent years (Chen et al. 2012; Dittmer 2013; Kim et al. 2014; Jiang et al. 2015; Zhou et al. 2015; Wu et al. 2017; Gusev et al. 2018; Yan et al. 2019).

Longtan River basin, located in the upper reaches of the Hongshuihe River, is the main stream of the Xijiang River in the Pear River system, China. After the construction and operation of the Longtan hydropower dam, regional climate has changed. Similarly, Miller (2005) noticed that reservoir area will influence the patterns of rainfall, wind, and temperature and found that the change from a natural river channel to a reservoir around the Three Gorges Dam decreases upward motion. Also, it further increases evaporation and surface temperature, cools the lower atmosphere, and increases rainfall as well as sinking air mass. Similar changes are shown in the study area. At the same time, with the change in vegetation coverage during the construction of the Longtan hydropower dam (2001), some changes occurred in the hydrological circulation process and water environment in the upper reaches (Green et al. 2007). Meanwhile, according to the existing measured runoff data, the study basin has been in a state of long-term runoff decline (2001–2010) (Huang et al. 2017). Therefore, future water resources will face many uncertainties (e.g., randomness, fuzziness, gray), and new challenges will arise for the sustainability and management of water resources as well as socioeconomic development in the study catchment. However, runoff-sensitivity research in the study catchment is scarce, although watershed managers indeed call for such study.

Based on the above-mentioned analysis, this paper focused on sensitivity analysis of runoff to climate and land-use changes, and tried to identify the major impact factors for runoff variability in this basin. The goal was implemented through a combination of the Soil and Water Assessment Tool (SWAT) model and the research approach of ‘climate scenarios-hydrological simulation-response analysis’. First, temperature and rainfall were selected as the main influencing factors according to the related researches (Huang et al. 2017; Mo 2018). Second, a satisfactory SWAT model was set up to simulate the hydrological process under different scenarios of climate and land-use change. Finally, the sensitivity of runoff to climate and land-use changes in the Longtan basin was quantified. The purpose of this paper was to further understand the influence mechanism of runoff variations, and provide a scientific basis for the possible subsequent challenges brought by climate change and changes in land-use patterns in the future. In general, the result of this paper can help policymakers and planners to propose suitable management practices, which can cope with the quantitative impacts of climate and land-use changes on runoff in this basin or other similar catchments.

The Longtan watershed, which is a leading reservoir of southwestern China, was selected as the study site for sensitivity analysis of runoff to climate and land-use changes. This catchment stretches from latitudes of 23°11′–27°01′N and longitudes of 102°14′–107°32′E, and extends across an area of 98,500 km² around Yunnan Province, Guizhou Province, and Guangxi Zhuang Autonomous Region (Figure 1). Tian-e hydrologic station is the outlet to the Longtan reservoir, which can be used not only for hydropower generation but also for water supply and flood control (Shi 2014). The average elevation of the Longtan basin is 1,450 m, with a valley elevation of 23 m in the southeast region and a plateau with an elevation of 3,358 m in the northwest area.

The climate of this catchment is sub-subtropical with a hot-wet summer season and cold-dry winter season. The average annual temperature for 1959 to 2013 ranges from 12.3 °C to 21.3 °C and the mean annual rainfall increases from 760 mm in the western region to 1,860 mm in the eastern area, where flood season rainfall (April to October) accounts for 89% of the annual precipitation. The main land uses of the study area are forestland and grassland, with 34.83% vegetation cover. The soil texture of loam accounts for 60% of the soil types.

The SWAT model, a semi-distributed and physically based model, was designed to predict the impact of land management practices on water, sediment, and agricultural chemical yields in large complex watersheds (Arnold 2012). The SWAT model divides a watershed into sub-basins and further subdivides each sub-basin into a number of hydrological response units (HRUs). Taking HRUs as the basic unit, hydrological components are simulated and aggregated for each sub-basin, and then routed to the basin outlet throughout the channel network to obtain the hydrological components based on water balance equation (Arnold et al. 1998). In this paper, SWAT was based on an interface of ArcGIS software, which provided an easy linkage of SWAT CUP for model calibration and uncertainty/sensitivity analysis (Abbaspour et al. 2007). Therefore, the required spatial data (digital elevation model (DEM), land-use, and soil type maps) and temporal data (meteorological data) were either raster or vector data sets.

DEM data for the Longtan watershed (Figure 1) was downloaded from the Geospatial Data Cloud website (http://www.gscloud.cn) and had a spatial resolution of 90 m. Land-use data in 2010 and soil map in 2000 were both provided by the Resource and Environment Data Cloud Platform of China (http://www.resdc.cn) and had a resolution of 1:1,000,000. At the same time, the reclassified land-use types of forest (main land use type), grassland, water body, urban area, bare land, paddy, and cultivated land were assigned as FRST, PAST, WATR, URBN, BALD, PADY, and AGRL, respectively (Wang et al. 2014). Meanwhile, the 35 different soil types were reclassified into seven soil types in the study catchment according to the FAO classification (Chesworth et al. 2008). These soils were Haplic Alisols (40.74%), Chromic Cambisols (31.03%), Humic Acrisols (15.44%), Dystric Cambisols (5.39%), Rendzic Leptosols (4.82%), Cumulic Anthrosols (2.28%), and Ferric Lixisols (0.30%).

Daily average, maximum and minimum temperatures, wind speed, relative humidity and precipitation data for the 25 meteorological stations (Figure 1) were obtained from the China Meteorological Data Sharing Service System (http://data.cma.cn) and met the requirements of data quality control, consistency check and record correction and recheck, covering the 55-year period from 1959 to 2013. Solar radiation data were simulated via the weather generator and other obtained climate data. Meanwhile, for the missing climate data at some stations over a short time, this paper used ArcGIS software to consider the correlation between meteorological stations for interpolation and supplementation, in order to improve the simulation accuracy of the SWAT model (Liston & Elder 2006).

Streamflow data consisted of the measured inflow after dam construction and the runoff deduced from the near-by Tian-e hydrological station before dam construction. Runoff data from 1959 to 2013 could be used for model calibration and validation.

According to the previous studies by Mo (2018) and Huang et al. (2017), temperature and rainfall change rates were about 0.13 °C/10a and 23.2 mm/10a, respectively. Considering the possible extreme weather and the local weather conditions, a 20% change will occur in rainfall in the near future. Meanwhile, temperature variety was related to precipitation change based on CMIP5 simulation results (Zhou et al. 2014) and a 2 °C change was assumed for temperature, which fitted the temperature circumstances in the study area (Huang et al. 2017; Mo 2018). Finally, 25 climate change scenarios were constructed based on their temporal and spatial variation characteristics and the situation hypothesis method (Jones et al. 2006; Somura et al. 2009), illustrated as Table 2. It was further simulated in the SWAT model.

According to the measures of conversion of cropland to forest and grassland in western China by Li (2002) and corresponding change models by Li & Wu (2002), four land-use type scenarios (SL1, SL2, SL3 and SL4) were set in this paper based on land-use data of 2010 (Supplementary material, Figure S1), topographic slope and vegetation types. Scenario 1 (SL1, ‘S’ means sensitivity, ‘L’ represents land-use type) assumed that all the current cultivated land would be converted to forestland and that the remaining land-use types would remain constant. In this situation, the forest coverage was 63%, and the grassland coverage rate was 26.69%. Scenario 2 (SL2) assumed that 14.29% of the current cultivated land would be changed to grassland, with 48.73% forest coverage and 40.96% grassland. Scenario 3 (SL3) assumed that all the current forestland would be converted to grassland for grazing and no forest land would remain. Scenario 4 (SL4) assumed that all the current grassland would be transformed to cultivated land, leaving no grassland in the Longtan basin. Simulation results of scenario 1 and scenario 2 reflect the hydrology response to regular policies of conversion of cropland to forest and grassland, and scenario 3 reflects the possible land-use change of overgrazing, which calls for a great demand for grassland. Scenario 4 demonstrates land-use varying from rangeland to bare ground, which is a common trend for farming in the study area.

The obtained coefficient of determination statistics R² (0.75 for annual calibration during 1987–1998, 0.88 for annual validation during 1999–2013) represent good consistency between the observed and simulated data and indicated low error variance. Furthermore, the NSE values were 0.85 for annual calibration and 0.88 for annual validation (Table 3). On the other hand, the calculated R² (0.94 for monthly calibration, 0.91 for monthly validation) and NSE (0.94 and 0.81) represented good matches between the monthly observed and simulated runoff (Figure 2) (Moriasi et al. 2007).

Runoff variability under different climate change scenarios (Table 4) showed that streamflow decreased when catchment rainfall decreased or temperature increased. Generally, for every 1 °C increase in the temperature of the basin, the average annual runoff decreased by 9.9 mm, and for every 10% increase in rainfall in this basin, the average annual runoff increased by an average of 96.3 mm. Specifically, from Table 4, the annual runoff decreased by 9.3 mm and 9.5 mm when the temperature increased by 1 °C and 2 °C, respectively. Correspondingly, decreases in temperature of 1 °C and 2 °C caused increases of 10.5 mm and 10.4 mm in streamflow, respectively, which were both based on the same rainfall. In contrast, mean annual runoff had a positive correlation with precipitation. Specifically, the increased scenarios for rainfall (10 and 20%), based on the same temperature, had induced streamflow increase of 98.1 mm and 99.7 mm, respectively, and runoff showed a decrease of 95.5 and 92 mm when rainfall was reduced by 10 and 20%, respectively. Obviously, runoff can be concluded to be more sensitive to the changes in rainfall than in temperature.

Evaporation variations related to the changes in temperature and rainfall (Table 5) showed that, on the one hand, for every 1 °C increase in temperature, the average annual evaporation increased by 9.3 mm; on the other hand, for every 10% increase in rainfall, the mean annual evaporation increased by an average of 11.53 mm. Specifically, evaporation maintained the same change trend as temperature and increased by 8.6 and 9 mm, respectively, when the temperature increased by 1 °C and 2 °C, while evaporation decreased by 9.9 mm and 9.8 mm when the temperature decreased by 1 °C and 2 °C, respectively, based on the same rainfall conditions. Similar results can be observed for the changes in rainfall. Evaporation increased or decreased with the increase or decrease in precipitation (increased 9.9 mm and 8.3 mm for 10 and 20% increase in rainfall, decreased 12.4 mm and 15.5 mm for 10 and 20% decrease in rainfall, respectively). In view of these results, evaporation was more sensitive to the changes in rainfall than in temperature.

The change amplitude of four land-use scenarios from the baseline period (2010) is shown in Table 6.

Compared to the simulation results (Table 7) of 2010 land-use type (Figure 3, baseline land cover), the mean annual surface runoff under the SL3 scenario reached a maximum value of 196 mm. Then, the increase in cultivated land resulted in an increase of 192.4 mm (SL4) in surface runoff and also increased for the conversion of cultivated land to forestland (177.5 mm) and grassland (182.4 mm). Similarly, the decrease in cultivated land (SL2) and the increase in forestland (SL1) contributed positively to water yield (520.3 mm for the SL1 scenario and 496.7 mm for the SL2 scenario), and the disappearance of forestland (SL3) or grassland (SL4) reduced water yield, compared to the land-use types of 2010. Meanwhile, an increase in evaporation was also observed when the vegetation rate increased (SL1 to SL3), such as the transformation among cultivated land, forestland, and grassland. Conversely, evaporation decreased to 575.7 mm when grassland was replaced by cultivated land (SL4). At the same time, the conservation index and runoff change rate were 0.225 and 10.02% for the SL1 scenario and 0.208 and 5.03% for the SL2 scenario, respectively. These values were largely affected by vegetation types, i.e., the SL2 scenario forestland and grassland. Runoff decreased 4.96% and 7.97% under the SL3 and SL4 scenarios, respectively, which may have been due to the decreases in forestland and grassland.

Monthly runoff variation under the four land-use scenarios (Table 8) demonstrated that dry season runoff (November to April) under the SL1 scenario ranked first, with a maximum value of 18.9 mm, followed by the SL2 and SL4 scenarios, and the SL3 scenario had the lowest monthly flow with a minimum value of 8.1 mm. In contrast, runoff under the SL3 scenario had the greatest surface discharge in the flood season, reaching a maximum value of 187.9 mm and the SL1 scenario produced the minimum surface runoff (merely 158.7 mm).

The simulation results for different climate and land-use change scenarios had great uncertainties, as did model calibration and validation. However, sensitivity ranking of the parameters CN₂, ALPHA_BF, GW_DELAY, GWQMN, and GW_REVAP was consistent with the results of other studies in similar catchments (Lv et al. 2014; Yang et al. 2017; Huang et al. 2018). On the other hand, although agreement between monthly observations and simulations was achieved, peak discharge for dry months was not a good fit and was underestimated before 2002 or overestimated after 2002. This outcome may have resulted from the construction of the Longtan hydropower dam, which further influenced the condition of the underlying surface. Alternatively, the peak mismatch can be attributed to CN₂, which assumed a unique relationship between cumulative rainfall and runoff in the same antecedent moisture conditions (Khoi & Suetsugi 2014). However, the objective of this study was not to predict floods. Therefore, this mismatch in peak flow can be ignored in this paper.

For the maximum variations in runoff and evaporation due to climate change in the section ‘Sensitivity analysis under climat change’, the mean annual runoff and evaporation decreased by 199.5 mm (173.5 mm) and 16.7 mm (41 mm) when temperature increased (decreased) by 1 °C to 2 °C and rainfall decreased by 20%; values increased by 182.1 mm (214.6 mm) for runoff and 33 mm (2.4 mm) for evaporation when temperature increased (decreased) by 1 °C to 2 °C and rainfall increased by 20% on average. Similar results can be obtained from Kong & Liang (2007), Wang et al. (2010), and Xu et al. (2018), who all indicated a high relation coefficient between rainfall and runoff. This study illustrated that in a karstic basin (Huang et al. 2017), rainfall directly infiltrated underground because of the development of surface rock cracks, fissures, and underground channels. Then rainfall further flowed elsewhere and gathered at the surface. The structure of poor storage capacity in karstic fissures was the major reason for the high correlation between precipitation and runoff. However, evaporation was small in the underground karst environment due to the shallow soil layer and high infiltration (Peng & Wang 2012). This situation may further lead to a relatively small impact on runoff variation when temperature increased. Based on the possible runoff response to rainfall and temperature change, reservoir managers can restore more water on poor rainy days, or release more rain water on rainy days.

Runoff responses to the four land-use scenarios were generally in agreement with the research literature, with some differences. For the annual time scale, deforestation and plantation (SL1 and SL2 scenarios) usually increased evapotranspiration, leading to less runoff (D'Almeida et al. 2010). Conversely, deforestation and urbanization (SL3 and SL4 scenarios) decreased evaporation and led to increased runoff (Dunkell et al. 2011). Relevant studies (Huang et al. 2012) have shown that damage to surface vegetation impaired rain water interception, collection, and holding capacity, and further increased surface discharge. Therefore, the decrease in forest coverage in the SL3 scenario (runoff reached 196.0 mm) would lead to serious soil and water losses in the study area. Regarding the response of monthly runoff to the four land-use scenarios, increased forest cover (SL1 and SL2 scenarios) clearly led to reduced floods, and reduced forest cover caused increased floods (Sahin & Hall 1996). Hence, during the rainy seasons, the conversion to forestland or grassland could prolong the infiltration process of surface runoff and effectively increase the soil water content. Meanwhile, the prolonged infiltration process can store the excess precipitation to recharge for water shortages in the dry season, and finally increase the available water resources and alleviate the drought effect. Obviously, high vegetation coverage under the SL1 scenario could greatly reduce surface runoff and weaken peak discharge during the flood season, which could avoid or alleviate flood disasters, to some extent. According to the runoff sensitivity to land-use scenarios, the reservoir planner should continue to address the measures of conversion of cropland to forest and grassland in western China. Particular attention should be paid to improving land cover to make a relative steady runoff change process.

This paper focuses on sensitivity analysis of runoff to climate and land-use changes, which is vital to provide deeper and better technical support for the watershed management and making ecology strategies. The main innovations include: (1) Constructing climate and land-use change scenarios based on the major impact factors for runoff variability, in the karstic area. (2) Simulating the hydrological process under different scenarios of climate and land-use change at different scales (annual and monthly). (3) Identifying the major impact factors for runoff variability based on quantitative analysis via the research approach of ‘climate scenarios-hydrological simulation-response analysis’.

The results of this study will be useful for understanding the potential impact of climate and land-use changes on runoff for similar catchments and meet the strong demand for proper measures aimed at global warming adaptation and soil and water conservation and protection.

The authors would like to thank the National Natural Science Foundation of China (Grant No. 51569003), the Natural Science Foundation of Guangxi Province (Grant No. 2017GXNSFAA198361) and the Innovation project of Guangxi Graduate Education (Grant No. YCBZ2018023).

The Supplementary Material for this paper is available online at https://dx.doi.org/10.2166/wcc.2020.196.