Hydrological responses and adaptive potential of cascaded reservoirs under climate change in Yuan River Basin

Extensive cascaded hydropower exploitation in southwestern China has been the subject of debate and conflict in recent years. The cascaded reservoirs' construction plan and operation policy were proposed based on the analysis of historical climate and hydrological observations. However, global mean annual temperature has increased by 0.8 °C since 1880 and future warming is almost certain according to the Intergovernmental Panel on Climate Change Fifth Assessment Report (IPCC AR5) (Flato et al. 2013). The variation in climate may change the balance of global water circulation and energy cycle, resulting in changes in the magnitude and spatial-temporal distribution of runoff, and even unforeseen changes in agriculture, environment, ecology, urbanization, and many other social-economic systems (Nam et al. 2015). Yuan River is a typical mountainous basin in southwestern China, and considering the huge influences of hydropower systems, it is vital to give a reliable picture of long-term climate change effects and the adaptive potential.

The general circulation models (GCMs) under the Coupled Model Intercomparison Project Phase 5 (CMIP5) are essential tools for simulating future changes in climate (Karl & Trenberth 2003). Several GCMs have been widely applied to predict the hydrological processes and assess the impact of climate changes. Xu & Xu (2012) assessed the performances of simulating the present climate over China based on 18 global climate models' simulations during 1961–2005, and demonstrated models can capture the dominant features of the geographic distributions of temperature and precipitation compared with observations. Among all GCMs under CMIP5, the Community Climate System Model version 4 (CCSM4) is a commonly used GCM consisting of atmosphere, land, ocean, and sea ice components linked through a coupler that exchanges state information and fluxes between the components. Its simulation ability has been confirmed in both global and regional scale, including southwestern China. For example, DeFlorio et al. (2013) demonstrated that the climatological precipitation is more realistic in CCSM4, both in spatial structure and seasonal agreement with observations. Shi et al. (2016) pointed out that although small biases exist, the spatial distributions of Asian surface temperatures and precipitation rates are simulated well by CCSM4. Zhang et al. (2017) indicated that CCSM4 performs well in simulating the spatial distribution of surface air temperature climatology and observational precipitation products in China. As well, the CCSM4 output can model temperature and most precipitation indices, including extreme precipitation (Liu et al. 2015). Climate projections are often used to drive hydrological models for the prediction of runoff responses in the future (Nerantzaki et al. 2015; Neupane & Kumar 2015). Distributed hydrological models have been particularly embraced by researchers, including the MIKE SHE model (Moussoulis et al. 2015), Distributed Hydrology Soil-Vegetation model (Du et al. 2016), Variable Infiltration Capacity model (Leng et al. 2016), and Soil Water Assessment Tool (SWAT) model (Serpa et al. 2015).

Water resource management is facing major challenges to balance conflicting objectives in reservoir operation given the uncertainties associated with climate change (Raje & Mujumdar 2010), which has stimulated the search for new adaptive operation rules to mitigate possible negative impacts of climate changes. Zhou & Guo (2013) proposed an integrated optimization model to develop operation rule curves of Danjiangkou Reservoir in a base period and three future periods under the A2 scenario, one of the four scenarios proposed by the Intergovernmental Panel on Climate Change in its Special Report on Emission Scenarios (IPCC SRES). Alvarez et al. (2014) developed a reservoir management tool to investigate the adaptation strategies of three reservoirs in the Lièvre River watershed to climate change under A2 scenario. Ahmadi et al. (2015) established an adaptive model to revise reservoir operating rules for Karoon-4 Reservoir in Iran as an adaptive strategy to climate under A1, A2, B1, and B2 scenarios. Currently, most of the studies in southwestern China are based on CMIP3 projections. The latest CMIP phase 5 (CMIP5) GCMs have advanced our knowledge of climate variability and climate change based on the previous version CMIP3. Considering the huge influences of hydropower systems in southwestern China, a more reliable picture is required for long-term climate change effects and adaptive potential assessment. In similar studies using CMIP5 data, Vliet et al. (2016) developed a coupled hydrological-electricity modeling framework to propose six adaptation options of a reservoir under Representative Concentration Pathways (RCPs) 2.6 and 8.5. Carvalho-Santos et al. (2017) studied the two-reservoir adaptation to water supply problems of a mountain catchment in northeast Portugal under RCPs 4.5 and 8.5. However, most of these studies were qualitative analyses and only gave the adaptive options. It has been rarely explored to establish an optimal operation model to propose the reservoir adaptive policies.

On the basis of previous studies, the purpose of this study is to update the understanding of climate change impacts on hydropower systems and to quantify the future potential in reservoir operation optimization. A systematic framework is proposed for climate change assessment and adaption, including bias correction, hydrological simulation, and adaptive operation. The future runoff responses to climate change are simulated and predicted based on bias-corrected GCM outputs. Finally, the adaptive operation chart model is developed to balance hydropower generation and ecological requirements of the cascaded reservoirs' system under climate change.

Yuan River is located in southwestern China (east: 100.01°–105.67°, north: 22.45°–25.53°) with a total length of 692 km and a drainage area of 34,629 km². This river has several tributaries, including Lixian River, Tengtiao River, Nanxi River, Panlong River, and Nanli River. Principal flow regulation capabilities within the basin are provided by two main cascaded reservoirs, namely, Gasa (GS) Reservoir and Madushan (MDS) Reservoir (Figure 1). MDS Reservoir is located in the downstream of GS Reservoir. Currently, GS and MDS Reservoirs are mainly utilized for hydropower generation, flood control, and ecological conservation. The main parameters of these two reservoirs are listed in Table 1. The conventional operating charts of GS and MDS Reservoirs are shown in Figure 2. There are five zones in the operation chart of GS Reservoir, including the reduced output zone (power output defined as 0.5Nr), guaranteed output zone (Nr), 1.5 times increased output zone (1.5Nr), 2 times increased output zone (2Nr), and full capacity output zone (Nrr), from the bottom to the top. There are three zones in the operation chart of the seasonal MDS Reservoir, including the reduced output zone, guaranteed output zone, and full capacity output zone from the bottom to the top.

Geospatial, runoff, meteorological, and climate data were collected in this study. Geospatial data included digital elevation model (DEM), land use and soil information, where the DEM data produced by the United States Geological Survey (USGS) (100*100 m) were used for defining stream, area, and boundary of sub-basins; land use data were developed by the Chinese Academy of Sciences (westdc_lucc_1km_China.asc) in 2000 with a spatial resolution of 1 km; soil data in 30 arc-second were derived from the Harmonized World Soil Database 1.2 (Fischer et al. 2008). The historical meteorological data were used to analyze the past climate features of Yuan River Basin, to calibrate the hydrological model, and to downscale the GCM outputs to a finer resolution. The observed variables included daily precipitation, maximum and minimum air temperature, solar radiation, wind speed, and relative humidity, which were collected at six meteorological stations including 56,768, 56,856, 56,875, 56,966, 56,985, and 56,968 (id number defined by China Meteorological Administration) from 1981 to 2007. The locations of these stations are shown in Figure 1. For the future climate projections, we analyzed the daily precipitation, temperature, maximum and minimum near-surface air temperature for the period 2015–2100 based on CCSM4. Then, the future projected temperature and precipitation of CCSM4 were bias corrected and spatial downscaled using BCSDd method. The monthly runoff data collected at the entry of the GS and MDS Reservoirs from January 1981 to December 2007 were used for the calibration and validation of the SWAT model. The future monthly runoff simulated from the SWAT model during the period of 2015–2100 as the inflow of GS-MDS Reservoir was applied to force the operation chart model.

IPCC AR5 has introduced a new way of developing scenarios. These scenarios span the range of plausible radiative forcing scenarios, and are called RCPs (Takemura 2012). There are four RCP scenarios, including RCP2.6, RCP4.5, RCP6.0, and RCP8.5, which are defined and named according to their total radiative forcing in 2100 relative to pre-industrial values. Here, RCPs 2.6, 4.5, and 8.5 were used to reflect the low, medium, and high forcing levels, respectively, as shown in Table 2.

When comparing historical simulation results from GCM to observations, comparison often shows that simulations tend to be biased wet/dry, cool/warm, with biases varying by location, season, and variable (Allen & Ingram 2002; Dibike & Coulibaly 2005). To reflect the regional climate change features, the BCSDd method (Thrasher et al. 2012) was performed to establish empirical relationships between GCM-resolution climate variables and local climate and to reproduce the regional climate features. BCSDd consists of two main steps, including bias correction and spatial downscaling. (1) In bias correction, a quantile mapping of daily temperature and precipitation from GCMs to observations re-gridded to the coarse model resolution is used to identify and remove bias (Panofsky & Brier 1958). Bias correction matches the statistical moments of observations and GCM output covering a common time period (e.g., 20th century), and accordingly adjusts for biases in GCM output for projected time periods (e.g., 21st century) by assuming a constant model bias. (2) In spatial disaggregation, daily bias-corrected precipitation and temperature are spatially disaggregated to the fine-resolution grid by interpolation using the synographic mapping system (SYMAP) algorithm (Hietala & Larson 1979), and then applying fine-resolution spatial anomaly patterns derived from the observations. The anomalies calculated as the correction factors between the re-gridded observations and coarsened projected GCM outputs were spatially interpolated to the fine-resolution grid; for precipitation, the fine-resolution daily observations multiplied by the correction factors are calculated as the fine-resolution daily simulations and for temperature, the fine-resolution daily observations plus the correction factors are the fine-resolution daily simulations.

SWAT requires as input hydro-meteorological data, a land-cover map, a soil map, and a DEM; after data compilation, ArcSWAT version 510 (Mehdi et al. 2015) running on an ArcGIS 9.3.1 platform was used for watershed delineation and sub-basin discretization. In Yuan River Basin, 107 sub-basins were delimited and then divided into multiple HRUs according to the land cover, soil types, and slope classes, as shown in Figure 3. The slope of the Yuan River Basin with a range from 0 to 30% accounts for more than 90% of the whole basin. Thus, due to the catchment's rather flat topography, single slope option was chosen to simplify the basin slope into two categories: 0–1% and 1–99.99%.

The SWAT Calibration and Uncertainty Program (SWATCUP) enables sensitivity analysis, calibration, validation, and uncertainty analysis of the SWAT model (Arnold et al. 2012), and Sequential Uncertainty Fitting 2 (SUFI2) of SWATCUP was used for this study (Abbaspour et al. 2007). The global sensitivity analysis integrated within SUFI2 was applied to test 21 SWAT hydrologic parameters, for surface runoff simulation in parallel with the calibration procedure, shown in Table 3. The derived new parameter values obtained from calibration and confirmation analyses were incorporated with the SWAT database for further simulations. The coefficients of determination (R²) (Herman et al. 2015) and Nash–Sutcliffe efficiency (NSE) (Nerantzaki et al. 2015) were used to evaluate model performance, respectively. The hydrological simulation was considered to be acceptable if R² > 0.5 or NSE> 0.5 mean.

The Tennant method (Tennant 1975) was used to determine the minimum ecological flow of the downstream area of the GS and MDS reservoirs. It is one of the classical hydrological methods developed to compute environmental flow. The Tennant method is simple as it requires no field work and is based on a single hydrologic statistic (mean annual discharge) (Palmer et al. 2008), which can be employed to calculate future ecological flow with predicted runoff under climate change. The commonly used threshold varied from 10% (poor or minimum condition) to 60–100% (optimum range) (Tharme 2003). Based on the investigation of key aquatic species of Yuan River Basin and their preference analysis (Wen et al. 2016), 60% of monthly average runoff was identified as the minimum ecological flow requirement.

The adaptive operation charge model of cascaded reservoirs was subjected to the following constraints.

• (1)
  Water balance constraint:

  

  (4)
• (2)
  Water storage constraint:

  

  (5)
• (3)
  Outflow constraint:

  

  (6)
• (4)
  Turbine release constraint:

  

  (7)
• (5)
  Power output constraint:

  

  (8)

  where m is the index of reservoir or hydropower station, ⁠; t is the index of time step; w and q are the inflow and outflow of reservoir (m³/s), respectively; S is the storage of reservoir (m³); I is the water loss of reservoir (m³); u is the turbine release of hydropower station (m³/s); N is the power output of hydropower station (KW); min and max are the minimum and maximum limits of variables, respectively. The values of model inputs are all shown in Table 4.

Dynamic programming successive approximation was applied to solve the chart model (Shi et al. 2015). The monthly guiding level of operation chart was taken as the decision variable.

The CCSM4 temperature and precipitation future projections were bias corrected and spatial downscaled to the grid scale of 0.5° × 0.5° using the BCSDd method. Figure 4 shows the downscaled CCSM4 projected temperature and precipitation for the future. Figure 5 indicates the spatial distribution of deviation of historical GCM simulations and BCSDd downscaled results to climate observations. The bias in CCSM4 simulated precipitation is mainly distributed in 10–40 mm. After downscaling, the variation of bias decreases to 5–20 mm. The improvements in the temperature downscaling are more significant. The variation of bias is 1–4 °C in CCSM4 simulations and 0.0–0.6 °C after downscaling.

The SWAT model was calibrated and validated using the inflow of GS and MDS Reservoirs for the period 1981–2000 and 2001–2007, respectively. The optimal value of parameters used for the calibration procedures are shown in Table 3. A significant correlation is observed between observed and simulated runoff with R² > 0.75 and NSE> 0.75 (before removing seasonal cycle) with R² > 0.4 and NSE> 0.4 (after removing seasonal cycle), as shown in Figure 6 and Table 5. The simulated and observed monthly average flow are 143.06 and 143.66 m³/s for GS Reservoir, and 295.16 and 296.44 m³/s for MDS Reservoir, respectively. However, most of the high flows (>500 m³/s for GS Reservoir and >1,000 m³/s for MDS Reservoir) are underestimated by 10–50%. The calibration strategy prioritizes the overall simulation accuracy, also the high flow simulation is always the weak point of the SWAT model (Ning et al. 2012).

The annual average precipitation of the Yuan River Basin is 1,102.86 mm for the period 1981–2007, and is expected to increase by 2.56–4.65% for the period 2015–2100. Figure 7(a) shows the future changes in decadal average projected precipitation under three RCP scenarios based on CCSM4 of the Yuan River Basin from 2015 to 2100 with respect to that from 1981 to 2007. After 2030, the annual precipitation may exceed the historical average level, and increases with increasing strength of the radiative forcing at a rate ranging from 5.83 mm/10a (RCP2.6) to 11.05 mm/10a (RCP4.5) and then to 18.15 mm/10a (RCP8.5).

Figure 7(b) shows the future changes in monthly average projected precipitation. The monthly precipitation could increase by 0.01–11.1 mm in spring (MAM), summer (JJA) and October. In particular, the Yuan River Basin may experience more serious flood events in the wet period in the future, since the monthly precipitation is expected to rise by 7.82–11.11 mm in June, the main flood period of the basin. Meanwhile, a decrease in the monthly precipitation is observed in winter months (DJF), September and November. The most significant reduction in monthly rainfall (2.56–4.06 mm) is observed in September, indicating that the Yuan River Basin may become dryer in the dry period in the future, and the transition between flood and drought may become faster and more obvious. In addition, the change in monthly precipitation is not sensitive to the radiation intensity.

Figure 8 shows the future changes in decadal and monthly average projected temperature based on CCSM4 of the Yuan River Basin from 2015 to 2100, respectively. The decadal average temperature is expected to increase by 0.80–2.22 °C in the future. Not surprisingly, the increasing rate ranges from 0.18 °C/10a in RCP2.6 to 1.33 °C/10a in RCP8.5. The monthly average temperature will also increase in a similar way, and the largest increase (0.80–2.50 °C) is detected in April and May.

The climate projections were employed to drive the calibrated SWAT hydrological model to predict the runoff responses to climate changes of the Yuan River Basin in the future. The upstream area of GS Reservoir is unregulated. The inflow of MDS Reservoir is composed of the release of GS Reservoir and the runoff flowing into the GS-MDS river reach. Figure 9 shows the decadal mean simulated average runoff change under RCPs 2.6, 4.5, and 8.5 based on CCSM4 of the GS and MDS Reservoir Basin during the period of 2015–2100, respectively. Generally, the upward variation trend is detected in the future decadal mean runoff. For the GS Reservoir Basin, the decadal average discharge is expected to increase by 2.93%, 5.36%, and 10.02% under RCPs 2.6, 4.5, and 8.5, respectively, compared to the historical baseline of 140.20 m³/s. For the MDS Reservoir Basin, the future average runoff is expected to increase by 0.79%, 4.04%, and 8.31% under RCPs 2.6, 4.5, and 8.5, respectively, compared to the historical mean discharge of 292.62 m³/s. In addition, the stream flow will be reduced after the 2060s under RCPs 2.6 and 4.5.

The boxplots of monthly mean simulated runoff under RCPs based on CCSM4 of the GS and MDS Basin from 2015 to 2100 compared with that from 1981 to 2007 are shown in Figures 10 and 11, respectively. Generally, the differences between wet and dry periods become more pronounced. The runoff in the flood season (June–September) will increase by 16.52–36.87% and 29.10–55.03% for the GS and MDS Basin, respectively. However, the dry period is expected to be much drier than before, since the monthly runoff will decrease by 8.41–15.72% and 22.60–29.63% for the GS and MDS Basin, respectively. The largest reduction is observed in October and November. The range of the monthly runoff variation increases as the radiative forcing intensity increases from RCP2.6 to RCP8.5, that is, the high forcing level may increase the possibility and the intensity of flood and drought. In addition, the temporal distribution of runoff is not sensitive to the variation of radiation intensity. These findings are consistent with the results in future precipitation analysis. The largest reduction is observed between September (the ending of wet period) and October (the beginning of dry period). For the GS Basin, October is decreased by 41.15–65.29% compared with September, while there is only 9.30% reduction for the historical scenario. For the MDS Basin, October is decreased by 52.31–69.08% compared with September, while there is only 15.86% reduction for the historical scenario. The same situation occurs between May (the ending of dry period) and June (the beginning of wet period). It can be concluded that the transition between flood and drought may become faster and more obvious in the future.

Figure 2 shows the adaptive operation chart of GS and MDS reservoirs generated from adaptive operation chart model under climate change. It indicates new regulation policies that allow for a better adaptation to the future flow regime under climate change for cascaded reservoirs' system in Yuan River Basin. Compared with the convention operation chart, the adaptive charts reduce the guiding level to produce more energy output for both GS and MDS reservoirs. Figure 12 shows the monthly adaptive operation results of cascaded reservoirs' system guided with the operation chart. It can be seen that GS Reservoir and MDS Reservoir have similar storage with different inflow under three scenarios, respectively. GS Reservoir and MDS Reservoir operate smoothly throughout the year. The only exception is that the monthly storage is higher in September of MDS Reservoir, which has small storage volume, due to the lesser outflow than inflow in this period.

The adaptive operation chart model without ecological penalty function was used (termed as nonpenalty-adaptive scenario) to explore the tradeoffs between hydropower and ecological conservation in the future. Future runoff following current operation rules was also simulated as the baseline condition (termed as non-adaptive scenario). Table 6 shows the results of non-adaptive (historical convention), nonpenalty-adaptive, and adaptive operation, respectively. For the future, the adaptive operation technique could significantly improve the ecological condition of the Yuan River Basin by reducing 52.66–70.77% of the total ecological streamflow shortage, at a cost of 2.09–4.54% decrease in total power generation of the GS-MDS cascaded hydropower system compared to that of the non-adaptive operation model. More water resources are allocated to satisfy the ecological demands, and the total spills and average storage are reduced by 5.81–13.18% and 1.54–8.63%, respectively, compared to the baseline scenario. In addition, the average storage changes by −5.72 to 3.98% for GS Reservoir, and is reduced by 19.24 to 22.96% for MDS Reservoir. Compared to the nonpenalty-adaptive model that prioritizes the hydropower objective, severe ecological deterioration could be avoided in the adaptive operation model, where the ecological runoff shortage is reduced by 71.16–83.01% at a cost of 11–13.97% decrease in the total power generation amount.

Figure 13 shows the monthly hydropower production and ecological runoff shortage of GS and MDS reservoirs for the future. The adaptive monthly hydropower production of GS Reservoir is expected to increase by 31.60, 21.30, and 25.90 million kW·h during May–September and decrease by 16.00, 14.20, and 9.40 million kW·h during October–April under RCPs 2.6, 4.5, and 8.5, respectively. The monthly hydropower production of the MDS hydropower station will decrease by 13.30, 14.60, and 15.70 million kW·h under RCPs 2.6, 4.5, and 8.5, respectively. The only exception is that the monthly hydropower production increases by about 14.50 million kW·h in October due to the increasing requirement of the ecological streamflow in this period.

Figure 14 shows the monthly ecological runoff shortage of GS and MDS reservoirs for the future. Under current rules, runoff shortage may occur in the downstream area of GS Reservoir in July–October in the future. However, after adaptive operation, it could be reduced by 64.81% under RCP 2.6 and totally eliminated under RCPs 4.5 and 8.5. However, the problem of ecological runoff shortage in the downstream area of MDS Reservoir is most pronounced in October, but less severe in January–May. Correspondingly, the adaptive ecological water shortage will be reduced by 55.60%, 65.20%, and 66.20% in October and 43.60%, 35.30%, and 33.80% from January to May under RCPs 2.6, 4.5, and 8.5, respectively.

Global warming is altering the global climate and hydrological cycle, thus the cascaded hydropower system needs to be changed correspondingly to better adapt to the new environment. This study aims to assess the climate change impacts on the hydropower system of the Yuan River and to quantify the future potential in operation optimization of Gasa-Madushan (GS-MDS) Reservoir system. A systematic framework is proposed for climate change assessment and adaption including bias correction, hydrological simulation, and adaptive operation. The SWAT model was applied to simulate the hydrological process, and then the BCSDd downscaled CCSM4 projections were used to drive the calibrated SWAT model to predict the future (2015–2100) runoff under RCPs 2.6, 4.5, and 8.5, respectively. An adaptive operation chart model was proposed to improve the hydropower system performance and quantify the potential optimization under climate change. Compared with the baseline condition (1981–2007), the warming and wetting trend was detected across the Yuan River with 0.80–2.22 °C increase in annual average temperature and 2.56–4.65% growth in annual precipitation. Also, the monthly average runoff may increase by 2.93%, 5.36%, and 10.02% for the GS Reservoir Basin, and by 0.79%, 4.04%, and 8.31% for the MDS Reservoir Basin under RCPs 2.6, 4.5, and 8.5, respectively. The Yuan River Basin may experience more serious flood events in the wet period and drought events in the dry period, and the transition between flood and drought may become faster and more obvious in the future. For the future, the adaptive operation technique could significantly improve the ecological condition of the Yuan River Basin by reducing 52.66–70.77% of the ecological streamflow shortage, at a cost of 2.09–4.54% decrease in total power generation of the GS-MDS cascaded hydropower system compared to that of the non-adaptive operation model.

This research is funded by National Natural Science Foundation of China (U1765201, 51609061), the Fundamental Research Funds for the Central Universities (2018B11314), National Key R&D Program of China (2018YFC0407902), the CRSRI Open Research Program (CKWV2016370/KY), Open Fund Research of State Key Laboratory of Hydraulics and Mountain River Engineering (SKHL1621), the Jiangsu Province Ordinary University Graduate Student Research Innovation Project (2016B05327), a project funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD). We would like to thank the United States Geological Survey (USGS), Chinese Academy of Sciences (CAS), International Union of Soil Sciences (IUSS) and China Metrological Data Sharing Service for providing geospatial and metrological data. Also, we acknowledge the World Climate Research Program's Working Group on Coupled Modeling and the climate modeling groups for producing and making available their model output.