Assessment of Potential Climate Change Effects on the Rice Yield and Water Footprint in the Nanliujiang Catchment, China

The Nanliujiang catchment is one of major rice production bases of South China. Irrigation districts play an important role in rice production which requires a large quantity of water. There are potential risks on future climate change in response to rice production, agricultural irrigation water use and pollution control locally. The SWAT model was used to quantify the yield and water footprint (WF) of rice in this catchment. A combined method of automatic and manual sub-basin delineation was used for the model setup in this work to reflect the differences between irrigation districts in yield and water use of rice. We validated our simulations against observed leaf area index, biomass and yield of rice, evapotranspiration and runoff. The outputs of three GCMs (GFDL-ESM2M, IPSL-CM5A-LR and HadGEM2-ES) under three RCPs (RCP2.6, 4.5, 8.5) were fed to the SWAT model. The results showed that: (a) the SWAT model is an ideal tool to simulate rice development as well as hydrology; (b) there would be increases in rice yield ranged from +1.4 to +10.6% under climate projections of GFDL-ESM2M and IPSL-CM5A-LR but slight decreases ranged from −3.5 to −0.8% under that of HadGEM2-ES; (c) the yield and WFs of rice displayed clear differences in the catchment, with a characteristic that high in the south and low in the north, mainly due to the differences in climatic conditions, soil quality and fertilization amount; (d) there would be a decrease by 45.5% in blue WF with an increase by 88.1% in green WF, which could provide favorable conditions to enlarge irrigated areas and take technical measures for improving green water use efficiency of irrigation districts; (e) a clear rise in future grey WF would present enormous challenges for the protection of water resources and environmental pollution control in this catchment. So it should be to improved nutrient management strategies for the agricultural non-point source pollution control in irrigation districts, especially for the Hongchaojiang and Hepu irrigation districts.


Introduction
Over the past decades, the global climate has undergone significant changes with global warming [1]. Agriculture is the largest freshwater consumer, accounting for more than 70% of the world's freshwater [2] and is extremely sensitive to climate change [3]. The change in the global climate will undoubtedly influence the future freshwater supply and food security by exerting pressure on agriculture and water resources, which presents a significant challenge to society [4]. It is necessary to physically-based semi-distributed hydrologic model, which has been widely used to assess the impact of prospective climate change on the water resources and agricultural production of various basins across the world [23,24].
There are many works regarding yield and WF of upland crops production focused on dry-farming areas [25][26][27], but few quantitatively studies have been done on the yield and WF of rice production in South China, our understanding on climate change in response to rice production and water resources utilization in this area is still insufficient. In this study, the SWAT model is used to assess the yield and water footprint of rice in the Nanliujiang Catchment. The Catchment is an important double-cropping rice production region and makes an important contribution to grain consumed within South China. Irrigation districts play an important role in rice production, and cropping therein requires large amounts of water for irrigation, which is potentially at stake under future water stress. Furthermore, the provincial agricultural department plans to continue enlarging irrigated areas of rice paddy in the next 15 years. Therefore, the question arises whether future climate change may be potentially harmful to rice production, agricultural irrigation water management as well as pollution control locally. In addition, one needs a tool that has good adaptability to simulate crop production including fertilization and irrigation, and assess crop yield and water use under future climate change. Consequently, the aims of this study were (1) to evaluate performance of the SWAT model; (2) to evaluate the future climate of the Nanliujiang catchment; (3) to assess rice yield under current and future climate conditions; and (4) to investigate quantitatively the WF of rice production under current and future climate conditions.

Study Area
The Nanliujiang catchment (21 • 35 54 -22 • 52 32 N, 109 • 00 03 -110 • 23 12 E) is located in the south of Guangxi province, China. It borders the Darong Mountain on the North and the North Bay on the South. The 9569 km 2 catchment has a unique leaf-shaped physiographic characteristic with a drop of more than 1250 m, high-lying northeast, and low southwest ( Figure 1). The Nanliujiang catchment belongs to the subtropics monsoon climate region, with an average annual temperature of 22 • C. The winter is mild with less rainfall, and the summer is hot and humid with abundant rainfall. The annual precipitation varies from 1596 mm in the northern mountains to 1804 mm on the southern coast. Having a proper soil and climate conditions, the catchment becomes one of the most important grain production bases in South China. The principal crops in this area are double-cropping rice. Early rice is usually planted in mid-March, until harvest in early July. Late rice is planted in mid-July and harvested around mid-November. The rainfall is heaving during the rainy season, about two-thirds of which falls during May-September; the maximum runoff usually occurs in July and August. However, the precipitation is less than the water requirement for rice growing in spring and autumn. Thus, irrigation is highly necessary for triggering rice growth and producing high yields. At present, the total cultivated land areas were expanded to 269.8 thousand hectares, approximately 70% of which were cultivated with rice ( Figure 2a). There are 18 irrigating areas with a total area of 179.7 thousand hectares (Figure 2b) [28].

Description of the SWAT Model
The SWAT model is a physically semi-distributed hydrologic model, which has been widely used worldwide because of its powerful functionality for simulating water yield, sediment, water quality in basins [29,30], as well as predicting the impact of human activity and climate change on hydrology [31].
In the SWAT, the river basin is divided into subbasins, and each subbasin is further divided into several Hydrologic Response Units (HRUs). The hydrological component of SWAT calculates the daily water balance for each HRU and takes into account precipitation, evapotranspiration, soil moisture, surface runoff and percolation. The water balance equation is represented as where SW is the soil water content; i is the time T (days) for the simulation period, P day , R surf , ET, W seep and Q gw are daily precipitation, surface runoff, evaporation, percolation and return flow (mm), respectively. The plant growth module that is incorporated into SWAT is a biogeochemical-based EPIC model [32,33]. The daily heat unit accumulation is used to regulate plant growth. Plant growth and crop yield are constrained by temperature, water, atmospheric CO 2 concentration and fertilizer application.
Plant biomass production is modeled by simulating leaf area development, radiation interception and conversion of intercepted light into dry biomass with the radiation-use efficiency of a specified plant. The equations are expressed as follows: where bio is the total dry biomass production in the growing season (kg·ha −1 ); RUE i is the radiation use efficiency on day i, kg·ha −1 ·(MJ·m −2 ) −1 ; H day is the incident total solar (MJ·m −2 ); k is the light extinction coefficient; and LAI is the leaf area index. The total leaf area index is calculated as follows: where fr LAImx,i and fr LAImx,i−1 are the fraction of the plant's maximum leaf area index for day i and i − 1, respectively; LAI mx is the maximum leaf area index of the plant; fr PHU is the fraction of potential heat units accumulated for the plant on a given day in the growing season; fr PHU,sen is the fraction of the growing season (PHU) at which senescence becomes the dominant growth process. The crop yield is determined by the total crop biomass, and the harvest index and is calculated as follows: where Yield is the crop yield (kg·ha −1 ); bio g is the aboveground biomass on the day of harvest (kg·ha −1 ); HI is the potential harvest index on the day of harvest; HI min is the minimum harvest index allowed for the plant; γ is the water deficiency factor. The research shows that the change in the atmospheric CO 2 concentrations has a large influence on the radiation use efficiency [34]. The formulas expressing the relationship between the radiation-use efficiency and CO 2 concentrations have been incorporated into SWAT for climate change studies, and the equation is as follows: where CO 2 is the concentration of carbon dioxide in the atmosphere (ppm); r 1 and r 2 are shape coefficients; and RUE is the radiation use efficiency, kg·ha −1 ·(MJ·m −2 ) −1 .

Data and SWAT Model Setup
The data required in this paper include the following: (a) a DEM map provided by the Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences; (b) a land use map ( Figure 2a) and (c) a soil map obtained from the Data Center for Resources and Environmental Sciences, Chinese Academy of Sciences (d) daily meteorological data from 1995 to 2013 obtained from the China Meteorological Science Data Sharing Service Network, consisting of daily precipitation, maximum and minimum air temperature, relative humidity, sunshine duration and wind speed for 5 meteorological stations (  (Table 1) provided by the Inter-Sectoral Impact Model Inter-comparison Project (ISI-MIP), which had been linearly interpolated and verified by statistical bias the probability distribution [35]. CO 2 was taken as a constant for each RCP (RCP2.6, CO 2 = 443, RCP4.5, CO 2 = 487 ppm, RCP8.5, CO 2 = 540 ppm, against CO 2 = 370 ppm currently) [8,24]. In plain agricultural regions with strong disturbance by human activity, the construction of ditch system projects profoundly alter the natural water flow paths and original sub-catchments [36,37]. Irrigation return water and runoff yield are mainly drained by the drainage canal system [38], making the irrigation district become an artificial subbasin [39,40]. Consequently, plain areas should not be divided into subbasins the watershed but the irrigation boundary controlled by drainage canal systems [41]. To show the real irrigation situation as closely as possible and reflect the differences between irrigation districts in yield and water use of rice, we used a combined method of automatic and manual delineation in this study. With the traditional watershed delineation method, the Nanliujiang catchment was divided into 26 subbasins according to the watershed (Figure 2c). Regarding the subbasins which contain both mountainous and plain, using manual delineation, 19 subbasins (1-2, 4-14, 16, 19-23) were redivided into 58 subbasins the boundaries of irrigation districts. Thus, a total number of 64 subbasins were identified for the catchment (Figure 2d), and 881 HRUs were defined therein. Depending on irrigation district location, irrigated agriculture was considered in subbasins 3, 6-7 and 24-59, and rain-fed cultivated land was considered in the others.
For each type of crop, the management practice was specified, such as planting, tillage, fertilization, irrigation and harvesting, among others. All operations were scheduled by the heat unit the cropping calendar. Fertilizers were applied with no effects of nitrogen and phosphorus stress through automatic fertilization application option in the SWAT model. The major sources of irrigation are reservoirs and river reaches (only irrigation district 4, 10 and 11 shown in Figure 2b). Irrigation water was extracted from the river reach or reservoir of a subbasin in response to the irrigation district location. Irrigation practices were accounted for in the simulations through automatic irrigation demand at the designated subbasins (it can be considered as sufficient irrigation without water stress), which aims to quantify the potential water requirement for rice growth under the influence of future climate change.

Crop Water Footprint
The water footprint (WF) is the total amount of fresh water needed to produce the goods and services related to a certain consumption pattern, which consists of green, blue, and grey WF [2]. The total WF of crop production can be calculated using the following equation: where WF is the total WF of crop production (m 3 ·kg −1 ); and WF green , WF blue and WF grey are the green, blue and grey WF (m 3 ·kg −1 ), respectively. WF green is calculated as follows: where WF green is the green WF (m 3 ·kg −1 ); W green is the amount of green water use per unit area (m 3 ·ha −1 ); Y is the crop yield per unit area (kg·ha −1 ); T is the length of the crop growing season (d); and ET green is the daily green water evapotranspiration (mm), which is the minor factor between the daily crop evapotranspiration ET and the daily effective precipitation P eff .
ET green = min ET, P e f f (8) where ET is the daily evapotranspiration (mm) and P eff is the daily effective precipitation (mm). Effective precipitation is the fraction of the total precipitation that is available to the crop and does not run off [42]. Based on this definition, P eff is calculated by using the SWAT model as follows [6]: where P is rainfall, RO is surface runoff and DP is deep percolation. W blue is calculated as follows [2]: where WF blue is the blue WF (m 3 ·kg −1 ); W blue is the amount of blue water use per unit area (m 3 ·ha −1 ); and ET blue is the daily blue water evapotranspiration (mm). In irrigated systems, ET blue is the difference between ET and ET green , and ET blue = 0 in rain-fed systems [6,43].
WF grey is defined as the freshwater required to assess the load of pollutants during crop growth, and it expresses the degradative water use [7].
where WF grey is the grey WF; AR is the product of the fertilizer application rate in the field per unit area (kg·ha −1 ); and a is the leaching-runoff fraction. Nitrogen is the main pollutant considered for most water footprint studies, which is estimated due to its potential pollution effect via nitrate leaching into ground and surface waters, here a = 0.14 [43]; and c max is the maximum acceptable concentration (10 mg·L −1 ) and c nat is the concentration in natural water (0.1 mg·L −1 ) [44].

Simulation of Future Crop Yield and Water Footprint
To quantify the impact of climate change on the crop yield and WF of rice in the Nanliujiang catchment, the bias-corrected output of climate variables (precipitation, temperature, and CO 2 ) from GCMs were fed into the calibrated SWAT model, and thus rice yield and WF were calculated for future periods. Here, the reference period (2001-2010) was ten years, and therefore the future reference period of the same length (2041-2050) was used for improved comparability. It was assumed that there will be no change in land use for the project area in the future. Climate variables in the future, such as wind speed, relative humidity, and sunshine hours, were also assumed to be the same as the reference period.

LAI and Biomass
The crucial parameters that pertain to crop growth were calibrated on a daily time step to minimize the simulated and observed residuals for LAI and BIOM. The related parameters of early and late rice were displayed their initial and final values in Table 2. It can be seen that the values of BIO_E, FRGRW2, LAIMX2 and BLAI showed greater differences between early and late rice, which were caused mainly by rice varieties and climate variations in growing season. LAI and BIOM were adjusted observed measurements in the Hepu irrigation district from 2005-2007 ( Figure 3). The performance indicators of calibration in this study were Nash-Sutcliffe (NS), root mean square error (PBIAS). Figure 3 and Table 3  where WFgrey is the grey WF; AR is the product of the fertilizer application rate in the field per unit area (kg·ha −1 ); and a is the leaching-runoff fraction. Nitrogen is the main pollutant considered for most water footprint studies, which is estimated due to its potential pollution effect via nitrate leaching into ground and surface waters, here a = 0.14 [43]; and cmax is the maximum acceptable concentration (10 mg·L −1 ) and cnat is the concentration in natural water (0.1 mg·L −1 ) [44].

Simulation of Future Crop Yield and Water Footprint
To quantify the impact of climate change on the crop yield and WF of rice in the Nanliujiang catchment, the bias-corrected output of climate variables (precipitation, temperature, and CO2) from GCMs were fed into the calibrated SWAT model, and thus rice yield and WF were calculated for future periods. Here, the reference period (2001-2010) was ten years, and therefore the future reference period of the same length (2041-2050) was used for improved comparability. It was assumed that there will be no change in land use for the project area in the future. Climate variables in the future, such as wind speed, relative humidity, and sunshine hours, were also assumed to be the same as the reference period.

LAI and Biomass
The crucial parameters that pertain to crop growth were calibrated on a daily time step to minimize the simulated and observed residuals for LAI and BIOM. The related parameters of early and late rice were displayed their initial and final values in Table 2. It can be seen that the values of BIO_E, FRGRW2, LAIMX2 and BLAI showed greater differences between early and late rice, which were caused mainly by rice varieties and climate variations in growing season. LAI and BIOM were adjusted observed measurements in the Hepu irrigation district from 2005-2007 ( Figure 3). The performance indicators of calibration in this study were Nash-Sutcliffe (NS), root mean square error (PBIAS). Figure 3 and Table 3

Discharge and Evapotranspiration
The first years (1995-1997) used to warm-up the model, and the data from 1998 to 2006 were used for calibration and those from 2006 to 2013 for validation. The 12 hydrological parameters were selected for 3 hydrological stations (Hengjiang, Bobai and Changle), and their optimal values for SWAT calibration are shown in Table 4. The results showed that the accuracy of the simulated value for the monthly streamflow at the Nanliujiang catchment in the calibration and verification period affectively met the standard (Figure 4). PBIAS of the simulated streamflow at the three stations ranged from 2.8 to 11.7 and from 3.5 to 13.0, respectively, and NS of the simulated streamflow at the three stations ranged from 0.85 to 0.94 and from 0.73 to 0.82, respectively (Table 5).
A direct relationship has been between crop development and evapotranspiration [45], and thus the evapotranspiration is also necessary for calibration in this work. However, the data of the actual evapotranspiration is hard to be measured or recorded, and therefore the measured monthly pan evaporation data from 2000 to 2011 for 3 stations (Yulin, Bobai and Hepu) may be a good source of information. Here, the values of the actual evapotranspiration were estimated by using a simple conversion method. Xie et al., 2009 [46] examined the ratio between the actual evapotranspiration and pan evaporation in the Dongjiang catchment, which is near to the Nanliujiang catchment and shows similar climatic characteristics. The actual evapotranspiration was estimated by multiplying the measured pan evaporation with the ratio.   Notes: r indicates the existing parameter value is multiplied by (1 + a given value), v indicates the default parameter is replaced by a given value.

Rice Yields
Root mean squared errors (RMSE) and mean relative error (MRE) were used to evaluate the SWAT performance. The parameters were calibrated measured yield data from two stations (Yulin and Hepu) for the period from 1999-2012. Table 7 showed acceptable performance by the model, with PBIAS values ranging from −13.13 to 7.90 and RMRE values from 0.24 to 0.83, and the MRE between the observed and simulated yield was less than 15%. All the results were considered acceptable for calibration of the model. Figure 4 showed the simulated and measured rice yields at Hepu station from 1999 to 2012.

Crop Yield and Water Footprint in the Reference Period
The rice yield in the Nanliujiang catchment was estimated using the validated SWAT model for the reference period (2001-2010), as shown in Figure 5a. The estimated total, green, blue and grey WF of rice in the 2000s calculated by Equations (6)-(12) is shown in Figure 5b-e respectively. Statistics (mean) of the yield and WF values are also shown in the figures. A combination of water use, soil and fertilization amount determined the spatial distribution characteristics of rice production in the catchment. Figure 5a shows that the yields of irrigated rice were higher than those of rain-fed rice. Since the rice in the rain-fed area was grown under rain-fed conditions during the growing period, there was a rapid decline in yield relative to that grown in irrigation districts under the same climate conditions. Rice yields varied from 4761 to 5264 kg·ha −1 in the irrigated area and from 3035 to 3378 kg·ha −1 in the rain-fed area. In the irrigated subbasins, the rice yields in the southern catchment were usually larger than that in northern catchment, and the rain-fed subbasins shared a similar characteristic. Hepu and Hongchaojiang irrigation districts (subbasin 24, 48-59) showed the maximum yield per acre in all irrigation districts, vary from 5174 to 5264 kg·ha −1 . The Maolin irrigation district (subbasin 31) showed the minimum yield per acre at 4761 kg·ha −1 .
The average total WF of rice production in the study area was 1064 m 3 ·t −1 (Figure 5b), which was higher than the national average (971 m 3 ·t −1 ) but below the global average (1450 m 3 ·t −1 ) [47]. The largest share of the total WF corresponded to the green WF, accounting for 44.8%, followed by blue and grey WFs, which accounted for 29.1% and 26.1%, respectively (Figure 5d The values of the WF of rain-fed subbasins with low rice production were higher than those of irrigated subbasins, confirming that a greater reduction in yield can result in large water footprints [6].

Future Climate Change
Since the south long-narrow catchment experiences a spatial and temporal variation in precipitation and temperature, the projection of future climate differs in different stations, which can be observed in the results provided by the different GCMs and RCPs. Future average precipitation and maximum and minimum temperature in the 2040s are shown in Table 8

Future Climate Change
Since the south long-narrow catchment experiences a spatial and temporal variation in precipitation and temperature, the projection of future climate differs in different stations, which can be observed in the results provided by the different GCMs and RCPs. Future average precipitation and maximum and minimum temperature in the 2040s are shown in Table 8   Rice is a C 3 plant, and therefore an increase in CO 2 can, in theory, increase its yield; however, yield is controlled by a combined effect of precipitation, temperature and CO 2 . A higher temperature can shorten the period of rice growth (according to the heat unit approach), and temperature stress can reduce the accumulation rate of dry materials, both of which may lead to low production. Figure 6 shows the potential rice yield in the 2040s under 9 climate change scenarios. On average, rice yield would increase by +3.0% in the future relative to the 2000s, ranging from −3.5% (HadGEM2-ES, RCP8.5) to +10.5% (GFDL-ESM2M, RCP4.5). Both GFDL-ESM2M and IPSL-CM5A-LR provide an increasing yield by different degrees under all RCPs, but HadGEM2-ES instead depict a decreasing yield under all RCPs ( Figure 6). GFDL-ESM2M shows an increased yield up to +10.9% for irrigated rice and +10.3% for rain-fed rice under RCP4.5 in the 2040s relative to the 2000s, due to increasing precipitation and CO 2 . HadGEM2-ES provide the lowest rice yield due to the highest temperatures compared with the other models under all RCPs (Table 8), because rice yield will be largely impacted by temperature stress as well as a shortened growing period. Rice is a C3 plant, and therefore an increase in CO2 can, in theory, increase its yield; however, yield is controlled by a combined effect of precipitation, temperature and CO2. A higher temperature can shorten the period of rice growth (according to the heat unit approach), and temperature stress can reduce the accumulation rate of dry materials, both of which may lead to low production. Figure  6 shows the potential rice yield in the 2040s under 9 climate change scenarios. On average, rice yield would increase by +3.0% in the future relative to the 2000s, ranging from −3.5% (HadGEM2-ES, RCP8.5) to +10.5% (GFDL-ESM2M, RCP4.5). Both GFDL-ESM2M and IPSL-CM5A-LR provide an increasing yield by different degrees under all RCPs, but HadGEM2-ES instead depict a decreasing yield under all RCPs ( Figure 6). GFDL-ESM2M shows an increased yield up to +10.9% for irrigated rice and +10.3% for rain-fed rice under RCP4.5 in the 2040s relative to the 2000s, due to increasing precipitation and CO2. HadGEM2-ES provide the lowest rice yield due to the highest temperatures compared with the other models under all RCPs (Table 8), because rice yield will be largely impacted by temperature stress as well as a shortened growing period. The spatial distributions of the average rice yields of the 3 GCMs under 3 RCPs in the 2040s were shown in Figure 7. It can be seen that the spatially clustered areas with high rice production are distributed in the northern (subbasin 3, 6, 7, 25-39), middle (subbasin 40-47) and southern (subbasin 24, 48-59) study area, which are the important irrigation districts and grain production base in South China. This finding can be attributed to strong factors such as a smooth topography, favorable soil The spatial distributions of the average rice yields of the 3 GCMs under 3 RCPs in the 2040s were shown in Figure 7. It can be seen that the spatially clustered areas with high rice production are distributed in the northern (subbasin 3, 6, 7, 25-39), middle (subbasin [40][41][42][43][44][45][46][47] and southern (subbasin 24, 48-59) study area, which are the important irrigation districts and grain production base in South China. This finding can be attributed to strong factors such as a smooth topography, favorable soil quality and proper irrigation facilities. The average rice yield in the irrigation districts will be increased by 239 kg·ha −1 under RCP2.6, 380 kg·ha −1 under RCP4.5 and 199 kg·ha −1 under RCP8.5 relative to the 2000s, respectively. Subbasins with low rice yield are distributed in mountainous and hilly areas with rain-fed farming due to a shortage of water conservancy facilities. The mountainous subbasins with larger yields will be subbasin 14-16, 18-21 and 64, which profit from advantages in climate conditions. The average rice yield will be increased by 138 kg·ha −1 under RCP2.6, 223 kg·ha −1 under RCP4.5 and 105 kg·ha −1 under RCP8.5 relative to the 2000s, respectively. However, their rice yields will be no larger than 3400 kg·ha −1 . These mountainous subbasins show a rugged topography with poor soils, mountainous cultivation could not only destroy mountain vegetation but also lead to soil erosion and environmental pollution. Therefore, reasonable adjustment and optimization of the agricultural structure is of great necessity in this area in the future. Mountain areas may be more adaptable to planting economic forests that can generate better economic returns, such as longan, litchi, mango and citrus. However, their rice yields will be no larger than 3400 kg·ha −1 . These mountainous subbasins show a rugged topography with poor soils, mountainous cultivation could not only destroy mountain vegetation but also lead to soil erosion and environmental pollution. Therefore, reasonable adjustment and optimization of the agricultural structure is of great necessity in this area in the future. Mountain areas may be more adaptable to planting economic forests that can generate better economic returns, such as longan, litchi, mango and citrus. shows that the average total WF of irrigated rice in the 2040s will decrease by different degrees under all scenarios. On average, the total WF will be reduced by 2.3% under RCP2.6, 4.9% under RCP4.5 and 5.4% under RCP8.5 relative to the 2000s, respectively. A projected rise in rice yield along with a decrease in blue water may be the main cause of this phenomenon. Due to the rise in precipitation during the growing season, rice has more green water to meet the increasing evapotranspiration demand triggered by climate change, leading to a substantial increase in green water demand and a reduction in blue water use. Hence, blue WF shows a significant decrease by 43.0% under RCP2.6, 48.3% under RCP4.5 and 45.2% under RCP8.5, respectively. Accordingly, green WF shows an increase from 207 m 3 ·t −1 to 394 m 3 ·t −1 under RCP2.6, 395 m 3 ·t −1 under RCP4.5 and 382 m 3 ·t −1 under RCP8.5, respectively. The amount of nitrogen fertilizer application will significantly increase with rice yield growth in the future, whereas the gray WF is likely to slightly increase by 4.0% under RCP2.6, 4.1% under RCP4.5 and 1.0% under RCP8.5, respectively, because rice yield improvements have helped to reduce obvious growth in the gray WF. Figure 8b shows that the mean total WF of rain-fed rice under 9 scenarios in the 2040s showed no obvious upward or downward trend. A slight decline in green WF and rise in gray WF may be the main causes. The green WF shows a small decrease by 4.0% under RCP2.6, 6.7% under RCP4.5 and 7.0% under RCP8.5, respectively. The amount of fertilizer applied to rain-fed rice will be important for growth along with rice yield growth in the 2040s, and the gray WF is likely to increase under all scenarios by 13.9% under RCP2.6, 17.1% under RCP4.5 and 10.2% under RCP8.5, respectively.  Figure 8a shows that the average total WF of irrigated rice in the 2040s will decrease by different degrees under all scenarios. On average, the total WF will be reduced by 2.3% under RCP2.6, 4.9% under RCP4.5 and 5.4% under RCP8.5 relative to the 2000s, respectively. A projected rise in rice yield along with a decrease in blue water may be the main cause of this phenomenon. Due to the rise in precipitation during the growing season, rice has more green water to meet the increasing evapotranspiration demand triggered by climate change, leading to a substantial increase in green water demand and a reduction in blue water use. Hence, blue WF shows a significant decrease by 43.0% under RCP2.6, 48.3% under RCP4.5 and 45.2% under RCP8.5, respectively. Accordingly, green WF shows an increase from 207 m 3 ·t −1 to 394 m 3 ·t −1 under RCP2.6, 395 m 3 ·t −1 under RCP4.5 and 382 m 3 ·t −1 under RCP8.5, respectively. The amount of nitrogen fertilizer application will significantly increase with rice yield growth in the future, whereas the gray WF is likely to slightly increase by 4.0% under RCP2.6, 4.1% under RCP4.5 and 1.0% under RCP8.5, respectively, because rice yield improvements have helped to reduce obvious growth in the gray WF. Figure 8b shows that the mean total WF of rain-fed rice under 9 scenarios in the 2040s showed no obvious upward or downward trend. A slight decline in green WF and rise in gray WF may be the main causes. The green WF shows a small decrease by 4.0% under RCP2.6, 6.7% under RCP4.5 and 7.0% under RCP8.5, respectively. The amount of fertilizer applied to rain-fed rice will be important for growth along with rice yield growth in the 2040s, and the gray WF is likely to increase under all scenarios by 13.9% under RCP2.6, 17.1% under RCP4.5 and 10.2% under RCP8.5, respectively. The spatial distributions of the average total, green, blue and gray WF of the 3 GCMs under 3 RCPs in the 2040s are shown in Figure 9. The WFs of rice production exhibit cleat differences among subbasins under 3 RCPs, mainly due to the differences in rice yields and water use levels. The rice yields and water use differ between subbasins and irrigation districts because of their variances in climatic conditions, soil quality and fertilizing amount caused by climatic conditions.  Figure 9l for gray WF). For instance, Figure 9a,e,i show that the subbasins in the northern, middle and southern catchment have a lower green, gray and total WF, representing the subbasins located in plains planted with irrigated rice. The irrigation districts located in the southern catchment usually have higher WFs than other irrigation districts due to the high precipitation and temperature. Hongchaojiang irrigation district (subbasin 56, 59) and Longmen irrigation district (subbasin 27) show the maximum and minimum total WF among all the irrigation districts, respectively. The spatial distribution of green WF is consistent with the distribution of precipitation. Due to abundant rainfall, the proportion of green WF in the Hepu (Subbasin 24, 48-55 and 57-58) and Hongchaojiang irrigation districts is relatively high. Farmers should take technical measures to increase green water use and reduce irrigation water use as more as possible. The southern irrigation districts also have the largest gray WF in the catchment, which increase by 10.8% under RCP2.6, 13.2% under RCP4.5 and 8.3% under RCP8.5, respectively. It will increase the pressure of water resource protection and water pollution control in the downstream The spatial distributions of the average total, green, blue and gray WF of the 3 GCMs under 3 RCPs in the 2040s are shown in Figure 9. The WFs of rice production exhibit cleat differences among subbasins under 3 RCPs, mainly due to the differences in rice yields and water use levels. The rice yields and water use differ between subbasins and irrigation districts because of their variances in climatic conditions, soil quality and fertilizing amount caused by climatic conditions. Although the values of total WFs are different under the 3 RCPs as well as the green, blue and gray WFs, their distributions under different RCPs are similar (Figure 9a vs. Figure 9e vs. Figure 9i for total WF; Figure 9b vs. Figure 9f vs. Figure 9j for green WF; Figure 9c vs. Figure 9g vs. Figure 9k for blue WF; Figure 9d vs. Figure 9h vs. Figure 9l for gray WF). For instance, Figure 9a,e,i show that the subbasins in the northern, middle and southern catchment have a lower green, gray and total WF, representing the subbasins located in plains planted with irrigated rice. The irrigation districts located in the southern catchment usually have higher WFs than other irrigation districts due to the high precipitation and temperature. Hongchaojiang irrigation district (subbasin 56, 59) and Longmen irrigation district (subbasin 27) show the maximum and minimum total WF among all the irrigation districts, respectively. The spatial distribution of green WF is consistent with the distribution of precipitation. Due to abundant rainfall, the proportion of green WF in the Hepu (Subbasin 24, 48-55 and 57-58) and Hongchaojiang irrigation districts is relatively high. Farmers should take technical measures to increase green water use and reduce irrigation water use as more as possible. The southern irrigation districts also have the largest gray WF in the catchment, which increase by 10.8% under RCP2.6, 13.2% under RCP4.5 and 8.3% under RCP8.5, respectively. It will increase the pressure of water resource protection and water pollution control in the downstream area. So the fertilizers application should be regulate scientifically in the future so that grey WF can be kept at a proper level.

Conclusions
In this study, we used the SWAT model to quantify the yield and WFs of double cropping rice within the Nanliujiang catchment, China. We validated the simulations against observed leaf area index, biomass, runoff, evapotranspiration and rice yield in sequence. The impact of potential

Conclusions
In this study, we used the SWAT model to quantify the yield and WFs of double cropping rice within the Nanliujiang catchment, China. We validated the simulations against observed leaf area index, biomass, runoff, evapotranspiration and rice yield in sequence. The impact of potential climate change on the yield and WF of rice was analyzed in the 2040s under 3 GCMs with 3 RCPs. The results showed that: (a) the SWAT model is an ideal tool to simulate rice development and WF under given and future climate conditions; (b) yield simulation showed significant variations under different future climate scenarios. It showed increased rice yields ranged from +1.4 to +10.6% under the GFDL-ESM2M and IPSL-CM5A-LR models but slightly decreased yields ranged from −3.5 to −0.8% under the HadGEM2-ES model; (c) both the yield and WF of rice exhibited obvious differences among subbasins, with characteristically high values in the south and low values in the north, mainly due to differences in climatic conditions, soil quality and fertilization amounts; (d) blue WF would significantly decrease by 45.5% and green WF would show a great increase by 88.1%, meaning green water would contribute the largest part in the WF of rice production in the catchment. These findings would be good news for local water resource development and utilization, and it would provide favorable conditions to expand the area under irrigation and take technical measures for improving green water use efficiency in irrigation districts; (e) the gray WF would increase under potential climate change, suggesting an increased requirement for fertilizer application in the future. The finding suggested that it was necessary to improved fertilization management strategies for the agricultural non-point source pollution control in irrigation districts, especially for the Hongchaojiang and Hepu irrigation districts. This work will also delineate some helpful information for the management and protection of water resources as well as irrigation planning.