1 Introduction

Global climate change has resulted in increasingly significant and far-reaching impacts and is a major issue of concern for the international community (Sivakumar 2011; Gosling et al. 2011; Kusangaya et al. 2014). Climate change exacerbates processes within global and regional hydrological cycles, resulting in changes to atmospheric circulation and ice conditions. Consequent changes in precipitation, evaporation, soil moisture, and river runoff, ultimately lead to the spatial and temporal re-allocation of water resources (Li et al. 2010). The fifth assessment report by the Intergovernmental Panel on Climate Change (IPCC) showed that, since 1901, mean precipitation over land areas of the northern mid-latitudes has increased, while between 1880 and 2012, the global mean surface temperature has risen by 0.85 °C (IPCC 2013). These changes will significantly affect temperature, precipitation and runoff; thus, causing an increase in flooding, droughts and other extreme events (Wu et al. 2010; Vaghefi et al. 2014; Spinoni et al. 2015; Wang et al. 2015a).

Jiangsu Province, which is a typical northern mid-high latitude plain area, covers a total area of 102,600 km2, of which 70,000 km2 represent downstream plains of the rivers. Owing to its moderate geographical and climatic conditions, the province represents the second largest economic region in China. The availability of water resources has been an important factor in the sustained economic development of Jiangsu Province (Ma 2007); however, as the area is located in both the north–south climate transition zone and in the marine–terrestrial transition zone, it is extremely sensitive to climate change. Weather and climate related disasters occur frequently and as such the spatial and temporal distribution of water resources is uneven and annual variations are large (Yang and Li 2003; Liu et al. 2009; Sun et al. 2013a; Xia et al. 2014). To date, climate change in Jiangsu Province has primarily affected precipitation (Deng et al. 2004; Qiu et al. 2008; Fu et al. 2013).

Rainfall levels determine the total water budget involved in the land surface water cycle, and variations reflect regional changes in water resources (Gu et al. 2010). In contrast, river runoff is related to social water needs (e.g., irrigation) more directly. Extreme events that affect runoff (e.g., flooding or droughts) can threaten economic development and human safety (Madsen et al. 2014; Tezuka et al. 2014; Wang et al. 2015b). Therefore, the most appropriate approach to assessing regional water issues is from the perspective of runoff. However, few studies have focused on runoff in Jiangsu Province, while fewer still have considered the impact on runoff based on future climate change scenarios. This is due to the lack of closed basins within the region, which significantly complicates the calibration of parameters during the building of hydrological models for plain areas. Climate models are the main tool used to project climate change (Dufresne et al. 2013), while the evaluation of climate change impacts on regional runoff usually involves a nested climate model and hydrological model approach (Wang and Zhan 2015). In this study, we simulated the impact of future climate change on runoff in Jiangsu Province by setting up a method to determine variable infiltration capacity (VIC) large-scale hydrological model parameters for plains, constructing a VIC model for Jiangsu Province, and by nesting multiple Coupled Model Intercomparison Project Phase 5 (CMIP5) climate models under the latest IPCC Representative Concentration Pathway (RCP) emission scenarios. The results of this study will provide support for water resources planning and the sustainable social and economic development of the region.

2 Data and methods

2.1 Study area

Jiangsu Province is located along the eastern coast of China (116.3°–121.95°E, 30.75°–35.33°N) and is an important part of the Yangtze River Delta region (Fig. 1). The terrain mainly consists of plains (having a higher terrain area than any other province in China), which cover an area of 70,000 km2, accounting for more than 70% of the area of Jiangsu Province. Jiangsu is the lowest-lying province in China, with the altitude of most areas being less than 50 m. Hills, which are concentrated in the southwest, account for just 14.3% of the total area. Jiangsu is located downstream of the Jianghuai and Yishusi rivers, while the Yangtze River runs across the southern part of the province. The Taihu, Hongze and Luoma lakes, along with other medium-sized lakes, are all located in Jiangsu, as are the Grand Canal, Tongyu Canal and other tributary rivers. Jiangsu experiences a temperate to subtropical transitional climate and moderate rainfall. Taking the Huaihe River as the boundary, the northern part of the province experiences a warm, humid and semi-humid monsoon climate, while the south experiences a subtropical humid monsoon climate. The mean temperature ranges between 13 and 16 °C, and gradually increases from northeast to southwest. Jiangsu is rich in water resources and the annual runoff is between 150 mm and 400 mm.

Fig. 1
figure 1

The location of Jiangsu Province, meteorological stations, rain gauge stations, soil moisture station (Xuzhou Station) and 13 cities. XZ Xuzhou, LYG Lianyungang, SQ Suqian, HA Huai’an, YC Yancheng, YZ Yangzhou, TZ Taizhou, NJ Nanjing, ZJ Zhenjiang, NT Nantong, CZ Changzhou, WX Wuxi, SZ Suzhou

Full size image

2.2 Meteorological data

Meteorological data was provided by the China Meteorological Data Sharing Network (http://www.escience.gov.cn/metdata/page/index.html) and the Jiangsu Province Hydrology and Water Resources Investigation Bureau (The office of Jiangsu province water resource comprehensive planning leading group 2014). Daily temperature and precipitation data were from 32 meteorological stations and 81 rain gauge stations during 1956–2011. We used the inverse distance weighting method (Lu and Wong 2008; Chen and Li 2012) to interpolate the site data across Jiangsu province into a 0.125°×0.125° horizontal resolution grid.

2.3 Emissions scenarios

Greenhouse gas emissions scenarios are the basis for global and regional climate predictions, and are usually calculated according to a series of assumptions (e.g., population growth, economic development, technological advances, environmental conditions, globalization, and fairness) that correspond to different social and economic development projections. In this study, we used three emission scenarios identified as major future climate change scenarios (Xu and Xu 2012): the high emission RCP8.5 scenario, the medium emission RCP4.5 scenario, and the low emission RCP2.6 scenario (IPCC 2012). RCP8.5 assumes a high population, low technological innovation, slow energy improvement, and slow income growth, all leading to a prolonged high-energy demand, high greenhouse gas emissions, a lack of policy response to climate change, and a predicted radiative forcing of 8.5 W/m2 in 2100. In contrast, RCP4.5 predicts that radiative forcing will stabilize at 4.5 W/m2 in 2100, while in RCP2.6 there is a limited global mean temperature rise of 2.0 °C, post-21st century energy applications for negative emissions, a radiative forcing peak before 2100, and a subsequent reduction in radiative forcing to 2.6 W/m2 in 2100.

2.4 Methods

2.4.1 Climate model filter

The climate models used in this study were chosen based on the long-term historical climate data from 47 CMIP5 models simulated under RCP2.6, RCP4.5, and RCP8.5. The spatial resolutions of the original data in each model were different; therefore, we took a reference period (1970–1999) and used it for comparison with the forecast climate data in this study (2011–2040). Considering the output results of GCM are grid data, we selected bilinear interpolation (Li et al. 2000) to unify the data onto the Jiangsu Province 1°×1° grid. In order to reduce the uncertainty of climate models and projections of climate change, the equidistant quantile matching method (Lu et al. 2014) was applied to revise multi-model ensemble mean monthly data, and then we applied the Delta method (Zorita and von 1999) to generate daily data for the reference period and three future scenarios.

To reduce model uncertainty and select the most appropriate approach (Zhang et al. 2010), we first evaluated the 47 CMIP5 global climate models (GCMs) by comparing the relative error between the simulated mean precipitation of the region across the reference period (1970–1999) and the mean measured precipitation, and the absolute error between the simulated mean temperature and the mean measured temperature (Table 1). It showed that the various models all well simulated the observed precipitation and temperature in Jiangsu province. While CMCC-CESM showed significant errors in precipitation, that of the other models was approximately +50%. For temperature, CNRM-CM5 showed significant error, but that of all other models was approximately +3 °C. We also calculated the correlation of precipitation and temperature of 47 models (Fig. 2). As we can see that the temperature of 47 models had a well correlation to measured data in the reference period, but there was fluctuation in correlation of precipitation between 0.951 and 0.997.

Table 1 Precipitation and temperature error in CMIP5 global climate models
Full size table
Fig. 2
figure 2

Correlations of annual precipitation (blue bars) and temperature (red bars) of 47 models over Jiangsu province in reference period (1970–1999)

Full size image

Since the parameterization process of many models are the same or belonging to the same category, it may lead to some similarities between results of these models and the number of models with consistent results may not be fully representative of the credibility and reliability of the conclusions (Zhao et al. 2013). If all these models in this agency are chosen, it certainly will bring about greater deviation on estimation of future climate change and runoff. Therefore, we chose hierarchical clustering analysis method (Pennell 2010) to select relatively independent models based on the independence between models of precipitation simulation. Specific steps are as follows:

  1. 1.

    Calculate errors between annual average simulated and observed values in every grid point (n) and each model (m), and normalize by the standard deviation (σ) of observed values to get single-model error e n,m :

    $$e_{n,m} = {{\left( {f_{n,m} - o_{n} } \right)} \mathord{\left/ {\vphantom {{\left( {f_{n,m} - o_{n} } \right)} {\sigma_{n} }}} \right. \kern-0pt} {\sigma_{n} }}$$
    (1)

    where, f represents the value of model simulation; o represents the observed value; n=1,2,…, N, N represents grid points of covering the area; m=1,2, …, M, M represents the amount of models. e n,m could be showed by vector quantity in space, e m  = (e 1,m , e 2,m , ···, e N,m ).

  2. 2.

    Calculate multi-model average error ē:

    $$\bar{e} = \frac{1}{M}\sum\limits_{m = 1}^{M} {e_{m} }$$
    (2)
  3. 3.

    Eliminate affection of errors ē, and get error vector d m :

    $${\mathbf{d}}_{m} = {\mathbf{e}}_{m}^{ * } - r \cdot {\bar{\mathbf{e}}}$$
    (3)

    where, (·)* is vector after standardization; r is correlation coefficient of m and ē. After eliminate ē, pertinence of m and ē is 0.

  4. 4.

    Calculate correlation coefficient matrix r i,j , take gradation cluster analysis:

    $$r_{i,j} = corr\left( {\text{d}_{i} ,\text{d}_{j} } \right)$$
    (4)

    where, i,j = 1, 2, …, M. we got average correlation coefficient matrix for four seasons r i,j according to the above steps, and used the weighted pair-wise average distance algorithm in Interactive Data Language(IDL) program language to take gradation cluster analysis. Considering the similarity of precipitation simulation results, hierarchical cluster analysis was selected to divide 47 models into five categories; and the most representative models from each category are selected according to their best simulation results of the current climate. Suppose these representative models which were relatively independent reflect the level of various types of model group in basin precipitation simulation. The result was shown in Fig. 3.

    Fig. 3
    figure 3

    Cluster analysis results and adopted models for the 47 Coupled Model Intercomparison Project Phase 5 (CMIP5) global climate models

    Full size image

Of the 47 models, the 5 models chosen in this study were CCSM4, CSIRO-Mk3-6-0, GISS-E2-R, HadGEM2-ES, and IPSL-CM5A-LR (Table 2). For these models, the error in mean precipitation ranged between −7.0 and 26.9%. Simulations were more accurate for the summer and autumn, and showed greater uncertainty for spring and winter. The simulation error for annual mean temperature was small and ranged between −1.8 and 0.5 °C. With regard to season, the temperature simulation under GISS-E2-R had a positive bias, while the other four models had negative bias.

Table 2 Climate model simulation results (2011−2040)
Full size table

Furthermore, according to the correlation analysis results of Fig. 2, we got the correlation of five representative models (Table 3). As can be seen from the data in the table, the precipitation of five models had a well correlation with the gauges and the correlation of temperature was very good.

Table 3 Correlation of five representative models between the GCM model outputs and measurements
Full size table

We also analyzed the cumulative distribution functions (CDF) of 47 models (Fig. 4). As shown in Fig. 4a, monthly precipitation of 47 models was between 3.54 and 4.78 mm, and that of five representative models was between 3.54 and 4.78 mm at 25% probability. Monthly precipitation of 47 models was between 1.46 and 3.94 mm, and that of five representative models was between 1.84 and 2.72 mm at 50% probability. Monthly precipitation of 47 models was between 0.65 and 2.51 mm, and that of five representative models was between 0.81 and 1.58 mm at 75% probability. Figure 4b shows that monthly temperature of 47 models was between 17.84 and 24.06 °C, and that of five representative models was between 19.86 and 21.13 °C at 25% probability. Monthly temperature of 47 models was between 11.31 and 18.13 °C, and that of five representative models was between 13.30 and 15.59 °C at 50% probability. Monthly temperature of 47 models was between 4.16 and 12.12 °C, and that of five representative models was between 6.73 and 10.65 °C at 75% probability. As we can see in the above figures, five representative models included models which were close to measured data and models which can reflect change intervals of 47 models, and it reflected the variation range of 47 models as a whole.

Fig. 4
figure 4

The CDFs of average monthly precipitation (a) and temperature (b) of measured (black solid) and 47 models (color solid and grey dashed) data over Jiangsu province in reference period (1970–1999). The color solid lines represent five models (CCSM4, CSIRO-Mk3-6-0, GISS-E2-R, HadGEM2-ES, and IPSL-CM5A-LR), the grey dashed lines represent the other models

Full size image

2.4.2 Setup and validation of the VIC model

Numerous hydrological models have been developed for impact assessment studies (Xu et al. 2008; Sun et al. 2013b; Lespinas et al. 2014), with the VIC model being one of the most widely used (Chang et al. 2014). The VIC model is a gridded large-scale hydrological model, in which each grid independently follows the energy balance and water balance in order to simulate hydrological and physical processes. Its most important feature is the introduction of the storage capacity curve concept of the Xin’anjiang model (Shu et al. 2008), which describes the uneven distribution of sub-grid soil moisture. Liang et al. (1996) explained the model principles in detail.

The parameters of VIC model can be divided into four categories: geography, vegetation, soil and hydrological. The first three are determined when the model is built and remain constant. However, due to the complexity of watershed runoff, it is difficult to determine hydrological parameters directly and so these must be calibrated using the measured hydrological data of each individual basin. The unique climate and geography of Jiangsu imposes new requirements for determining hydrological parameters. As few closed basins are available to verify the hydrological model, it is difficult to calibrate hydrological parameters by site-measured data. Formulas used to determine the hydrological parameters have typical national wide parameter transfer formula (NPTF) (Lu et al. 2010) and southern part of China parameter transfer formula (SPTF) (Lu et al. 2013).

NPTF is a hydrological gridding parameter formula that is based on nationwide 43 typical basins, established by structuring multiple regression equation using hydrological parameter, basin soil and climate factors. 43 typical basins were selected over different climatic regions in China, and they were divided into two sections. The first part was 35 basins which are used to calibrate hydrological parameter and establish hydrological parameter formula; the other 8 basins were used to verify the formula to ensure the independence of the basin. 17 factors of soil and climate in the basin were chosen to build the multiple regression equation through stepwise regression. Three kinds of regression models were considered in regression analysis: multivariate linear regression model (LIN), multivariate square root model (SQRT), and multivariate logistic model (LOG). The optima land the most significant model was finally selected as the parameter formula. SPTF aims to rebuild gridding parameters formula in southern China because NPTF doesn’t consider the impact of the special climate, soil and other factors in north-south differences and some factors like landform and vegetation feature. This formula was established based on 49 typical basins and 25 factors of climate, soil, vegetation and landform which are selected from south areas of China, then the VIC model hydrological gridding parameter formula was established in southern China. The multiple regression hydrological gridding parameter formula was built through stepwise regression, it can improve the nationwide VIC model gridding parameter formula and enhance VIC model simulation precision.

However, typical basins of these two sets of formula are not found in Jiangsu Province as they generally represent mountainous watershed. Parameters for use in plain areas require further testing and cannot be directly selected. Therefore, in this study, we used two sets of gridding formula (NPTF and SPTF) to calculate hydrological parameters (Table 4); however, we used rational analysis and fine-tuning to determine the exact parameters used, which were verified by comparison with data from the Jiangsu Water Resources Bulletin (JWRB). The hydrological parameters estimated, which varied spatially (Fig. 5), were as follows:

Table 4 Mean regional hydrological parameters calculated by NPTF and SPTF
Full size table
Fig. 5
figure 5

The distribution of five hydrological parameters of VIC over Jiangsu Province with a 0.125°×0.125° resolution: a B (saturated water content capacity curve); b Ds (the proportion of the maximum daily base flow when base flow non-linear growth occurs); c Dsmax (base flow during maximum daily flow); d D2 (the thickness of the second layer of the soil); e D3 (the thickness of the third layer of the soil)

Full size image
  1. 1.

    B represents the saturated water content capacity curve, which acts on D1 (0.1 m) and D2, and has a great impact on runoff. In Jiangsu Province the differences between topography and vegetation on the 0.125° × 0.125° grid is small; therefore, we chose NPTF for the initial calculation and did fine-tuning.

  2. 2.

    Ds represents the proportion of the maximum daily base flow when base flow non-linear growth occurs. This parameter is related to soil texture, and considering the low-lying topography of Jiangsu, lateral flow is slow, we chose the relatively small NPTF for the initial calculation.

  3. 3.

    Dsmax represents the base flow of the maximum daily flow. Again, given the low-lying topography of Jiangsu, lateral flow is slow; therefore, we chose the relatively small SPTF for the initial calculation and we performed fine-tuning.

  4. 4.

    Ws represents the proportion of underlying soil moisture and the maximum soil moisture when base flow nonlinear growth occurs. Because of the low-lying topography, groundwater outflow in the region is mostly smooth and linear, with nonlinear outflow rare; therefore, Ws was set to 1.

  5. 5.

    D2 represents the thickness of the second layer of the soil. Research in the Qinhuai River Basin of Jiangsu Province showed that the surface soil moisture content in fields around the Yangtze River ranged between 120 and 150 mm, while saturated water contents ranged between 160 and 200 mm. Soil porosity, as calculated based on soil properties, was 0.49, and the effective soil thickness was 0.32–0.40. This thickness corresponds to D1 + D2 in the VIC model; therefore, the appropriate range for D2 was 0.22–0.30, and we selected SPTF for the initial calculation and then performed fine-tuning.

  6. 6.

    D3 represents the thickness of the third layer of the soil. In Jiangsu, the mean groundwater depth is 1–2 m; therefore, we selected NPTF for the initial calculation and then made appropriate adjustments.

The annual runoff of Jiangsu Province from 1956 to 2009, as modeled by the VIC, was similar to the measured values published by JWRB (Fig. 6). The mean values were 265 and 262 mm, respectively, while the relative error was only −1.4%; however, the data simulated by VIC was smaller than that published by JWRB. Both datasets fit well with annual discharge processes and their coefficient of determination came to 0.88. The annual mean absolute relative error was 18%, but there were 6 drought years (1978, 1994, 1966, 1967, 1973 and 1976) in which relative error was greater than 40%. The main reason for these significant differences was that the calculation methods of the VIC model for rainfall-runoff transformation relations in dry years was different from that used by the JWRB. The VIC model shares the storage capacity curve concept with the Xin’anjiang model, it assumes that precipitation fulfills interception vegetation firstly, and then all the rest are used to calculate runoff. The VIC model separates water source when calculating runoff, the upper layer soil produces direct runoff and upper and lower soil water infiltration, while the subsoil generates base flow. While in JWRB, the Jiangsu underlying is divided into Plains and mountains, and then according to land attribute is subdivided into construction land, water area and agricultural land (includes paddy field and dryland). The analysis was separately carried out from construction land, water area, paddy field and dryland according to their runoff characteristics.

Fig. 6
figure 6

Comparison of the 1956–2009 annual surface runoff depth (mm) simulated by the variable infiltration capacity (VIC) model (dashed line) and the values published by the Jiangsu Water Resources Bulletin (solid line). Er relative error, Dc annual runoff coefficient of determination

Full size image

At the same time, the process of soil moisture (0–100 cm) of Xuzhou station, Jiangsu province during 1981–1999 (Fig. 7) simulated by VIC model showed a good agreement between the simulated and measured soil moisture. The average value of measured data was 328 mm and that of simulated data was 316 mm, the relative error was −3.72%, and the coefficient of determination was 0.52. To verify the simulation of VIC model in space, we compared the annual runoff simulated by VIC and that in each city from Jiangsu province published by Jiangsu Water Resources Bulletin during 1980–2000 (Fig. 8). When we calculated the surface runoff of 13 cities, we first calculated the proportion of the grid in each region based on GIS tools, and then computed the average runoff depth of the region using the area ratio weighting method. As shown in the figure, two sets of data were close and the average error of 13 cities was 8%, the maximum gap reached 24% in Suzhou and Taizhou, the next reached 15% in Huai’an, others were under 10%. Overall, the model established in this paper was rational and feasible.

Fig. 7
figure 7

Comparison of simulated (red dashed) and Measured (black solid) soil moisture (mm) for depths of 0–100 cm at Xuzhou station during 1981–1999. Er relative error, Dc coefficient of determination

Full size image
Fig. 8
figure 8

Comparison of the 1980–2000 annual average surface runoff depth (mm) simulated by the VIC model (grey bars) and published by the Jiangsu Water Resources Bulletin of 13 cities (black bars)

Full size image

3 Results and Analysis

3.1 Projected precipitation and temperature

3.1.1 Temporal variations

Projected changes in precipitation, as calculated using the five models and three scenarios, showed that future precipitation will differ little from the base year, ranging from −7.7 to 10%; in contrast, the mean annual temperature over the next 30 years was predicted to increase by 0.81–1.82 °C (Fig. 9).

Fig. 9
figure 9

The change trends of Annual mean precipitation (a) and temperature (b) in Jiangsu Province during 2011–2040 based on the results of five models (CCSM4, CSIRO-Mk3-6-0, GISS-E2-R, HadGEM2-ES, and IPSL-CM5A-LR) and three Representative Concentration Pathway (RCP) scenarios (2.6, 4.5, 8.5)

Full size image

3.1.2 Spatial variations

Projected precipitation, simulated using the five models and three scenarios, showed an increasing trend generally, with the rate of increase becoming gradually larger from south to north (Fig. 10a). Overall, precipitation in the Yishusi Basin was predicted to increase by more than 2%; although, western and eastern coastal regions increased 3–5%. In the Huaihe River Basin, precipitation was predicted to increase by 1–3%. In the Yangtze River and Taihu basins, potential increases were not obvious; although, in the southern part of the Taihu Basin, precipitation was predicted to decrease slightly. Projected temperatures in Jiangsu Province showed a significant increasing trend (Fig. 10b), with the rate of increase becoming gradually larger from southeast to northwest. Temperature was predicted to increase most significantly (by more than 1.32 °C per year) is the coastal border area between the Yishusi and Huaihe River basins.

Fig. 10
figure 10

Spatial variations in annual mean precipitation (a) and temperature (b) for Jiangsu province during 2011–2040 based on the results of five models (CCSM4, CSIRO-Mk3-6-0, GISS-E2-R, HadGEM2-ES, and IPSL-CM5A-LR) and three Representative Concentration Pathway (RCP) scenarios (2.6, 4.5, 8.5)

Full size image

3.2 Runoff changes over the next 30 years

3.2.1 Seasonal changes

Figure 11 showed the future (2010–2040) relative changes of monthly runoff depths compared with reference period (1970–1999) under RCP2.6, RCP4.5 and RCP8.5 scenarios. The histogram represented the average forecast results of five models and the error lines represented the variation range of five models. As can be seen that the relative changes of monthly runoff depths showed large difference under three scenarios. The average runoff depths of five models tended to decrease in February, October and November, while the other months mainly tended to increase. The relative change range of runoff depths in December was from 44 to 75%, while other months were from −20 to 15%. As can be seen from the error lines under three scenarios, the consistency of predicted changes of five models was poor and the results still had great uncertainty, such as the ranges of monthly runoff depths in January under three scenarios were −52.21 to 52.84% (RCP2.6), −22.93 to 26.38% (RCP4.5) and −20.94 to 68.42% (RCP8.5), while the average changes were −9% (RCP2.6), −2% (RCP4.5) and 10% (RCP8.5).

Fig. 11
figure 11

Relative changes (%) of monthly runoff depths in the future (2011–2040) compared with reference period (1970–1999) under three scenarios in Jiangsu Province. The colored bars are the average results of five models (CCSM4, CSIRO-Mk3-6-0, GISS-E2-R, HadGEM2-ES, and IPSL-CM5A-LR) under three scenarios: RCP2.6 (blue bars), RCP 4.5 (red bars), and RCP 8.5 (green bars). Error lines represent the change range of five models

Full size image

3.2.2 Spatial variations

The predicted changes in the annual mean runoff of Jiangsu Province from 2011 to 2040 differed both spatially and temporally according to the model and RCP scenario employed (Figs. 12, 13, 14; Table 5). Under RCP2.6, model CCSM4 showed the most widespread and significant positive changes, while model GISS-E2-R consistently predicted reductions in runoff across the province (Fig. 12). Model GISS-E2-R also predicted widespread reductions in annual mean runoff depths under both RCP4.5 (Fig. 13) and RCP8.5 (Fig. 14). Under RCP4.5, only model HadGEM2-ES predicted consistent increases in runoff depth across the province, with the other models predicting both increases and decreases, depending on location. The most significant changes in runoff depth were predicted under RCP8.5 (Fig. 14). Changes in the Yishusi Basin showed an increasing trend in model CCSM4, HadGEM2-ES and IPSL-CM5A-LR under RCP2.6 and RCP4.5, in model CCSM4, CSIRO-Mk3-6-0 and HadGEM2-ES under RCP8.5. While changes in the Taihu Basin showed a decreasing trend in model GISS-E2-R and IPSL-CM5A-LR under three RCPs, HadGEM2-ES under RCP2.6 and CCSM4 under RCP4.5. The Huaihe River and Yangtze River basins contained both increasing and decreasing runoff depth predictions.

Fig. 12
figure 12

Spatial variations in the relative change of annual mean runoff depth (%) in Jiangsu Province from 2011 to 2040 using the Representative Concentration Pathway (RCP) 2.6 scenario and models a CCSM4, b CSIRO-Mk3-6-0, c GISS-E2-R, d HadGEM2-ES, and e IPSL-CM5A-LR

Full size image
Fig. 13
figure 13

The same as Fig. 12, but for RCP 4.5

Full size image
Fig. 14
figure 14

The same as Fig. 12, but for RCP 8.5

Full size image
Table 5 Relative changes (%) in annual mean runoff depths in the Jiangsu basins (2011–2040)
Full size table

By averaging the results of the five models, our results predicted that overall, and for all three scenarios, runoff depths in the northern part of Jiangsu Province will follow a clear increasing trend over the next 30 years, while runoff depths may reduce slightly in the south (Fig. 15a). An analysis of the likelihood of an increasing trend (Fig. 15b; Table 6) supported this result, with the highest probability of increasing runoff depth occurring in the northern parts of the province. The annual mean runoff variation for the four basins under the three scenarios over the next 30 years ranged from approximately −6 to 15%. Overall, the annual mean runoff depths in the Taihu Basin were predicted to change by −6 to 3%, the Yangtze River and Huaihe River basins were predicted to increase by 0–12%, and the runoff depths in the Yishusi Basin were predicted to increase by 6–15%.

Fig. 15
figure 15

Mean relative change in annual mean runoff depths and level of increased likelihood. a Mean relative change (%) in annual mean runoff depths and b the level of increased likelihood (deep red very likely; light red likely; orange more likely than not; yellow about as likely as not; light green unlikely; deep green very unlikely) in four watersheds using three Representative Concentration Pathway (RCP) scenarios (2.6, 4.5, 8.5) from 2011 to 2040

Full size image
Table 6 Definitions of estimated result possibilities
Full size table

Based on RCP2.6 alone, runoff depths in the Taihu Basin may fall in the future, with the decreases predicted to be less than 3%, while the runoff depths in the other river basins were predicted to increase by 3–12%. Under RCP4.5, the runoff depths in the Taihu and Yangtze River basins were predicted to fall by −6 to −3%, while those of the Yishusi and Huaihe River basins were shown to be more likely to increase, with the increases predicted to be in the range of 0–3 and 6–9%, respectively. Under RCP8.5, the runoff depth in the Yishusi Basin was predicted to increase by 12–15%. The runoff depths in the other three basins were also shown to be likely to increase, in general by 3–12%; although, the increase in the Taihu Basin was predicted to be less than 3%.

4 Discussion

The results of this study highlight the different possible responses of runoff depth in Jiangsu Province to different forecasted climate change scenarios. The results contain some uncertainty, which mainly arises from: (1) the selection of the climate models and the uncertainty within the climate scenarios. In particular, while simulations of current climate may be good, there may be some errors for future climate simulations; (2) error within the VIC model used to simulate runoff. While the simulated runoff data were shown to closely match the data of the Jiangsu Province Water Resources Bulletin, deviations from actual runoff may still occur. However, despite these uncertainties, we believe the climate change trends simulated in this study to be highly accurate. Furthermore, our results are consistent with related research studies, which have shown that: (1) runoff in the Huaihe River Basin may increase from 2011 to 2060 (Zhang et al. 2014); (2) the runoff in the flood season may increase, while non-flood season runoff may decrease in the Yangtze River Basin (Ju et al. 2011), and; (3) from 2021 to 2050, the Taihu Basin runoff will increase, especially during flood seasons (Liu et al. 2010).

The predicted changes in Jiangsu Province surface runoff between 2011 and 2040 will have a significant impact on Jiangsu Province and it is critical that responses to these changes be considered. In particular: (1) the Yishusi Basin is the most water-stressed area of Jiangsu Province and its flood control engineering requirements are currently low; however, we should consider how future increases in runoff may increase flood challenges, and how flooding may impact on water quality; (2) While runoff in the Yangtze River Basin will likely increase in the future, it remains important to consider flood control during droughts and during the non-flood season. In particular, regional flooding or droughts may impact upon agricultural production; (3) Increasing runoff in the Huaihe River basin may exacerbate flooding. Polluted water from flood discharge will result in water environmental problems and may have an extremely negative effect on the quality of drinking water and on fishery production; (4) Higher predicted temperatures will seriously impact on the water environment, and may cause an increase in the series blue-green algae problems (Shang et al. 2010; Xie et al. 2016) suffered in the Taihu Basin. Furthermore, cyanobacteria stench and toxins may seriously affect the living environment of aquatic organisms and human life.

5 Summary and Conclusions

In this study, we used the VIC model to simulate Jiangsu Province runoff from 2011 to 2040, and by employing five CMIP5 global climate models, we were able to assess the impact of climate change on predicted runoff. The main findings of this study showed that:

  1. (1)

    Annual mean precipitation under the five models and three scenarios showed a fluctuating upward trend, as compared with the reference period (1970–1999), and ranged between −7.7 and 10%. The rate of increase was shown to become gradually larger from south to north, except in the southern part of the Taihu Basin, which showed a slight decrease.

  2. (2)

    Annual mean temperature was projected to increase between 2011 and 2040, as compared with the reference period (1970–1999), with predicted increases ranging from 0.8–1.8 °C. Temperature increases were particularly pronounced after 2030. The rate of increase was shown to become gradually larger from southeast to northwest, with the largest rise in temperature predicted to be 1.34 °C.

  3. (3)

    Increasing runoff depths were highest under the RCP8.5 emission scenario and lowest under RCP4.5. With the exception of February, October, November (where runoff depths were predicted to fall significantly between 2011 and 2040), monthly mean runoff depths were predicted to increase, with the most significant increases occurring in December.

  4. (4)

    Spatial distributed runoff depths variations ranged from −6 to 12%. The likelihood of increasing runoff depths in the four basins increased from south to north. The largest increases in runoff depths were predicted by RCP8.5. Based on both RCP4.5 and RCP2.6, runoff depths in the Taihu Basin may see future reductions, but that in the other three basins were predicted to increase.

The predicted changes in runoff depth in Jiangsu Province will have significant implications for flood management, agricultural production, fisheries, water resource security, water quality, and the heath of aquatic ecosystems; therefore, the results of this study will provide an important reference for policymakers planning for the future water resources of Jiangsu Province. Furthermore, we conducted a case study with the VIC model, it showed high consistency with measured values and provides new reference values for the study of other plain regions, for which the underlying conditions are very complex.