Assessment of drought risk for winter wheat on the Huanghuaihai Plain under climate change using an EPIC model-based approach

ABSTRACT Climate change-induced drought poses a serious negative impact on global crop production and food security. The Huang Huai Hai (HHH) Plain, one of the most important grain production areas in China, is heavenly stricken by drought. Motivated by formulating drought risk prevention strategies that adapt to climate change on the HHH Plain, therefore, the present study aims to quantitatively evaluate the winter wheat drought risk under multiple climate scenarios using the Environmental Policy Impact Climate (EPIC) model. Based on the well-validated EPIC model, the drought hazard intensity (dHI), physical vulnerability (pV), and drought risk (dR) of the HHH Plain from 2010 to 2099 are assessed. Temporally, the dR showed an increasing trend in the long term, the high dR areas increased by 0.63% and 1.18% under the RCP4.5 and RCP8.5 scenarios, respectively. Spatially, dR showed a pattern of high in the south and low in the north whether under RCP4.5 or RCP8.5 scenario. Comparatively, the dR was 0.211 under the RCP4.5 scenario which was slightly higher than that under the RCP8.5 scenario, i.e. 0.207. The Huanghuai Plain agricultural subregion will be a high dHI-pV-dR region. The temperature increase might be the main factor affecting the wheat drought risk.


Introduction
Drought seriously threatens the stable development of global agriculture and food security (Mishra and Singh 2010;Lesk, Rowhani, and Ramankutty 2016;Pandey et al. 2012). Worse still, drought has shown an increasing trend under the influence of climate change (Harrison et al. 2014;Xu et al. 2013). It was projected that global drought-affected agricultural areas may significantly increase by the end of the twenty-first century (Huang et al. 2015). Therefore, it is urgent to quantitatively assess the drought risk of crops under climate change, which creates important scientific value for drought risk adaption and management strategies formulation under climate change. Such study is of great significance to ensure food security in China and even the world.
Wheat is one of the world's major crops contributing approximately 20% of global dietary calories and protein (Portmann, Siebert, and Döll 2010;Shiferaw et al. 2013). However, most wheat growing areas are threatened by drought (Yue et al. 2015;Zampieri et al. 2017). Drought caused 75% of the global wheat harvested area to reduce production, with approximately an average reduction of 0.29 t×ha −1 (Kim, Iizumi, and Nishimori 2019). The Huanghuaihai Plain (HHH Plain) plays a vital role in ensuring China's food security where producing approximately 75% of China's total wheat (Wu et al. 2006;Chang et al. 2020). However, it is also one of the areas with the worst agricultural drought in China, causing a serious negative impact on winter wheat production (Hu et al. 2021). It was observed that precipitation showed a significant decreasing trend, but the frequency of heat waves increased during the winter wheat growing season in 1960-2006(Tan et al. 2010. Therefore, 91% of meteorological observation stations on the HHH Plain showed an increasing trend in the drought index during the winter wheat growing seasons (Ti et al. 2018;Hu et al. 2021), and drought contributed 70% of the winter wheat production loss in the eastern region in 1996-2012 . It also showed that climate change from 1961 to 2003 resulted in a 45.3 kg ha −1 a −1 drop in wheat growth potential in the northern region of the HHH Plain (Chao et al. 2010). Simulations indicated that the mid-to-high drought vulnerability of winter wheat will increase under any RCPs scenario on the HHH Plain (Li et al. 2015b). Furthermore, it was expected that drought will result in a 2%-6% reduction in wheat production in this region under the global warming targets of 1.5°C and 2°C (Chen, Zhang, and Tao 2018). Therefore, the HHH Plain will be a high-risk area for wheat drought risk in China and the world (Yue et al. 2018;Zhang et al. 2015a).
The previous studies warned us that the HHH Plain may face a very high risk of agricultural drought and lead to a substantial reduction in grain production as a result of climate change. At the same time, the HHH Plain was required to undertake 16.45 billion kilograms of additional grain production capacity, accounting for 32.9% of the country's new production capacity in the document passed by the State Council in 2009 (Plan for National Increased Grain Production Capacity by 100 Billion Catties (2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019)(2020)). This puts forward an urgent research task, that is, the need to accurately understand the risk dynamics of wheat drought caused by future climate change. Findings of such research are vital information for putting forward reasonable adaptation strategies to climate change with ensuring China's future food security. Therefore, it is very urgent to further strengthen the research on agricultural drought risk under the influence of climate change in this region.
Crop mechanism models are effective tools for simulating the impact of climate change on crop production (Matthews et al. 1997). Based on crop mechanism models, such as Environmental Policy Impact Climate (EPIC) (Guo et al. 2020;Yue et al. 2018;Zhang et al. 2015a), CropSyst (Li et al. 2015a), CERES (Röttera et al. 2011;Zhang et al. 2019), AquaCrop model (Zarei, Shabani, and Mahmoudi 2021a, 2021b, 2021c, and Sirius (Richter and Semenov 2005), several studies have quantitatively assessed the impacts of rice, corn and wheat by simulating the response of crops to meteorological factors, especially drought, that is, the expected yield loss rate under future climate change scenarios (e.g. Guo et al. 2020;Zhang et al. 2019;Richter and Semenov 2005). These researches showed that the crop mechanism model is an important tool to assess crop drought risk under future climate change scenarios. However, research on wheat drought risk estimation under the influence of climate change on the HHH Plain using crop model is still rare.
Based on the above, this article aims to reveal the spatial pattern and temporal evolution of winter wheat drought risk under the RCP4.5 and RCP8.5 scenarios on the HHH Plain to provide a scientific basis for agricultural drought risk management and climate change adaptation. On this basis, this article achieves the following research goals: (1) develop a wheat drought risk assessment model; (2) use the EPIC model to simulate winter wheat yield loss and corresponding water stress under selected RCP scenarios; and (3) calculate and analyze the drought risk and its temporal and spatial distribution of winter wheat on the HHH Plain from 2010 to 2099.
Most of the region is a warm temperate, humid or semi-humid area with a temperate monsoon climate. The average annual precipitation from 1980 to 2010 was 734.9 mm (Li et al. 2015b), and precipitation was concentrated in July to September. The precipitation during the growth period of winter wheat in this region is only approximately 130 mm (Zhang et al. 2008), which is far below the water demand of 400-500 mm for winter wheat production (Liu et al. 2009). Thus, the growth of winter wheat on the HHH Plain is vulnerable to drought stress (Zhao and Liu 1999). Heat and soil conditions differ greatly as a result of the large coverage of the HHH Plain (Rashid et al. 2019). Therefore, the HHH Plain is divided into 4 agricultural regions (National Agricultural Regionalization Committee 'Comprehensive Agricultural Regionalization of China ' Compilation Group, 1981), including Region-1 Yanshan Taihang Mountain piedmont plain agricultural region, Region-2 Jiluyu low-lying plain agricultural region, Region-3 Huanghuai Plain agricultural region, Region-4 Shandong Hilly agricultural and forestry region ( Figure 1). Table 1 shows the data used in the study including historical meteorological observation data (1960-2011), soil property data (2000), field management data (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009), winter wheat production data (1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011), winter wheat harvest area data (2010) and meteorological prediction data of RCP scenarios (2010-2099). Among which, the projected meteorological data of the selected RCP scenarios driven by the GFDL-ESM2M model was provided by the Inter-Sectoral Impact Model Intercomparison Project (ISI-MIP). The ISI-MIP provides global daily meteorological data at a spatial resolution of 0.5°× 0.5°from 1971 to 2099 based on five general circulation models (GCMs), including GFDL-ESM2M. This set of data was bias corrected and there are no data missing (https://www.isimip. org/gettingstarted/input-data-bias-correction/) and therefore has been widely recognized and applied (Yue et al. 2018;Yue, Zhang, and Shang 2019;Jiang, Yue, and Gao 2019).

EPIC model-based winter wheat drought risk assessment method
The process of assessing the drought risk of winter wheat under the RCPs scenario using the EPIC model is as follows. First, a model parameter sensitivity analysis using the SCE-UA algorithm was performed, the sensitivity parameters were calibrated and verified, and the winter wheat EPIC parameter combination on the HHH Plain was obtained (section 2.3.1). Then, the drought risk of winter wheat on the HHH Plain under RCP4.5 and 8.5 scenarios was assessed by evaluating the drought hazard intensity and physical vulnerability of winter wheat using the well calibrated EPIC model (section 2.3.2).

EPIC model calibration and validation
Each combination of crop parameters in the EPIC model represents a specific crop variety (Williams et al. 1989). Therefore, it is necessary to calibrate and verify the parameters, that is, through parameter sensitivity analysis, parameter adjustment and parameter validation, and then the EPIC model parameter combination of winter wheat on the HHH Plain is obtained.
First, a parameter sensitivity analysis is performed. Based on previous studies (Wang et al. 2013), this study selected ±20% of the default values of 41 parameters that affect the crop growth environment as the upper and lower limits of the parameter values to fully express the sensitivity of different parameters to the impact of crop growth, used the global sensitivity analysis method Extended Fourier Amplitude Sensitivity Test (EFAST) to generate 6237 sets of winter wheat parameter Global Agro-Ecological Zones, GAEZ (Fischer et al. 2012) combinations. In this study, the 2003-2004 winter wheat management data of Yucheng Experimental Station were taken as an example (Fan et al. 2014), and winter wheat yield corresponding to different parameter combinations was simulated in 2004. Then, parameter sensitivity analysis was carried out. Next, parameter calibration was performed on the selected 8 sensitive parameters. We use the Shuffled Complex Evolution method developed at the University of Arizona (SCE-UA) to calibrate the sensitive parameters of winter wheat (Yue et al. 2015). The process is as follows: according to the crop parameters selected by the sensitivity analysis, 150% of the default parameter value is set as the upper limit of the parameter range, and 50% of the default parameter value is set as the lower limit of the parameter range. The principle of calibration is to minimize the total absolute error between the annual simulated yield and the actual yield of each typical station in every agricultural area, and then the crop parameter combination is obtained. Among them, the selection of typical stations in the agricultural area is based on the principle of closest to the center; that is, the selected typical stations are as close as possible to the center of the area so that they can fully represent the geographic information of the agricultural subregion. Finally, the model performance is evaluated. This paper uses the mean absolute error, mean relative error, and root mean square error (RMSE) to evaluate the accuracy of the calibration results of EPIC model parameters. The specific evaluation variable is as follows: (1) Mean absolute error, the difference between the predicted value and the observed value, which is the average absolute deviation of the predicted value from the observed value (Equation 1): (2) Mean relative error, the average ratio of the absolute error to the observed value (Equation 2): (3) Root mean square error, the standard deviation of the sample mean (Equation 3): where x i is the simulated value and y i is the observed value, n is the sample number.

Drought risk assessment of winter wheat under climate chang
This study adopted the definition of risk proposed by the IPCC (2015), that is, risk is the probability of harmful consequences, which is the product of hazard, vulnerability of disaster subject and exposure (Carrão, Naumann, and Barbosa 2016). In present study, the following equation is used.
where dR is the winter wheat drought risk, dHI is the drought hazard intensity, pV is the winter wheat physical vulnerability, and E is the wheat exposure (E is 1 where winter wheat is planted, and E is 0 where no winter wheat is planted). According to the value of dR, the HHH Plain is divided into high drought risk areas (dR > 0.7), medium drought risk areas (0.3 ≤ dR≤0.7), and low drought risk areas (dR < 0.3). dHI is calculated from the meteorological data of 2010-2099 under the RCP4.5 and RCP8.5 scenarios by simulating the water stress of winter wheat in each 5 ′ ×5 ′ grid unit based on the EPIC model (Yue et al. 2018). According to the dHI value, the HHH Plain is divided into high drought hazard areas (dHI > 0.7), medium drought hazard areas (0.3 ≤ dHI≤0.7), and low drought hazard areas (dHI < 0.3).
pV is the physical vulnerability of winter wheat in the HHH Plain. The adverse effects of meteorological factors on wheat yield were used to characterize the physical vulnerability of wheat (Papathoma-Köhle et al. 2011;Uzielli et al. 2008). Firstly, we assumed that the normal annual yield not affected by drought is the 95% maximum annual yield during the base period (1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005). Secondly, the annual yield of winter wheat was simulated for each 5 ′ ×5 ′ grid unit based on the EPIC model under RCP scenarios. Finally, pV was obtained by calculating the physical yield loss rate of the future relative to the base period. The calculation is as follows: where YI jP is the physical yield loss rate of site j, Y jbase is the yield of the base period of site j, and Y jPRCP is the yield under different RCP scenarios of site j. The value is normalized as follows: where pV is the physical vulnerability and max(YI PH ) is the maximum yield loss rate in the HHH Plain.
According to the value of pV, the HHH Plain is divided into high physical vulnerability areas (pV > 0.7), medium physical vulnerability areas (0.3 ≤ pV≤0.7), and low physical vulnerability areas (pV < 0.3).

EPIC model calibration and validation
The result of parameter sensitivity analysis is shown in Figure 2 (See Appendix 1 for the connotation of relevant parameters). Compared with the first-order sensitivity index, the total sensitivity index can reflect the overall impact of the same degree of changes of different parameters, including the relationships among parameters (Vazquez-Cruz et al. 2014;Wang et al. 2005). Total sensitivity index was used to choose the sensitive parameters, Considering the sensitivity index in this case study and previous studies using the EPIC model in the HHH Plain (Wang et al. 2005;Fan et al. 2014), the following eight parameters were selected as the sensitive parameters of winter wheat production in HHH Plain to calibrate the EPIC model, which were the biomass-energy ratio (WA), harvest index (HI), maximum potential leaf area index (DMLA), fraction of growing season when leaf area declines (DLAI), first point on optimal leaf area development curve (DLAP1), second point on optimal leaf area development curve (DLAP2), leaf area index decline rate parameter (RLAD), and potential heat unit (PHU). The above 8 sensitive parameters were calibrated using SCE-UA algorithms, and the winter wheat parameters in each planting area were obtained (Table 2). In general, the EPIC model verification result was acceptable by evaluating the mean absolute error, mean relative error, and root mean square error of winter wheat yield from 2001 to 2009 at typical stations in the agricultural subregions of the HHH Plain (Figures 3 and 4, Table 3). However, it showed that the mean absolute and relative errors of each wheat planting area are negative, which indicates that the simulated yield of most areas is lower than the actual yield, which may be due to the neglect of farmland management measures, such as pesticides in the model simulation process, which led to the simulated yield.

Winter wheat drought hazard
The drought hazard index (dHI) of winter wheat in the HHH Plain under the RCP4.5 and RCP8.5 scenarios in the short-term (2010-2039), mid-term (2040-2069), and long-term (2070-2099) scenarios are shown in Figures 5-7.  In all periods of the two scenarios, the dHI of winter wheat was higher than 0.5 in more than 20% of the regions (Figure 5 and 6). From 2010-2099, the average dHI value under the RCP4.5 and RCP8.5 scenarios were 0.355 and 0.347, respectively, showing a spatial distribution pattern of high in the south and low in the north. Areas with high drought hazard (dHI > 0.7, area accounting for 10.6%) are mainly distributed in the Huanghuai Plain agricultural region and partly distributed in the Jiluyu low-lying plain agricultural region; and areas with low drought hazard (dHI < 0.3, area accounting for approximately 47.4%) are mainly distributed in the Yanshan Taihang Mountain Piedmont Plain agricultural region and Shandong Hilly agriculture and forestry region ( Figure 5-7).
The area of high drought hazard under the RCP4.5 scenario decreased and then increased over the three-time periods (Figure 5), while the area of high drought hazard under the RCP8.5 scenario showed a continuous decrease ( Figure 6). Compared with the short-term, the proportion of high drought hazard areas in the long-term increased by 0.5% and decreased by 1.1% under the RCP4.5 and RCP8.5 scenarios, respectively (Figure 7). In addition, the area of low drought hazard both increased and then decreased over time under climate change ( Figures 5 and 6). Compared with the short-term, the low drought hazard proportion in the long-term decreased by 1.7% and increased by 3.9% under the RCP4.5 and RCP8.5 scenarios, respectively (Figure 7).
The above results indicated that, compared with the short-term, in the long-term, the area of high drought hazard under the RCP4.5 scenario will increase, while the area of low drought hazard will decrease; the area of high drought hazard under the RCP8.5 scenario will decrease, but the area of low drought hazard will increase. That is, the overall drought hazard of the RCP4.5 scenario will be higher than that of the RCP8.5 scenario on the HHH Plain and continued to show a spatial pattern of high drought hazard in the south and low drought hazard in the north. However, the overall drought hazard increases with time under the RCP4.5 scenario, and RCP8.5 shows a decreasing trend.
The pV of winter wheat was higher than 0.5 in more than 18% of the areas in the short-term of RCP4.5 and in all terms of RCP8.5 (Figure 8 and 9). During 2010-2099, the average pV under the   RCP4.5 and RCP8.5 scenarios were 0.275 and 0.348, respectively, showing a spatial distribution pattern of high in the south and low in the north. Areas with high physical vulnerability (pV > 0.7, area accounting for approximately 6.3%) are mainly distributed in the Huanghuai Plain agricultural region; areas with low physical vulnerability (pV < 0.3, area accounting for approximately 55.3%) are mainly distributed in the Yanshan-Taihang Mountain piedmont plain agricultural region and Shandong Hilly agricultural and forestry region (Figures 8 and 9). The high physical vulnerability areas under the RCP4.5 scenario showed a continuous decreasing trend, and there was a sharp decline from the short-term to the mid-term, with the area proportion decreasing from 7.8% in the short-term to 1.1% in the mid-term (Figure 8). The high physical vulnerability areas under the RCP8.5 scenario decreased and then increased over time (Figure 9). Compared with the short-term, the proportion of high physical vulnerability areas decreased by 6.8% and increased by 2.9% under the RCP4.5 and RCP8.5 scenarios, respectively ( Figure 10). In addition, the low physical vulnerability areas under the RCP4.5 scenario increase and then decrease (Figure 8), while the area of low physical vulnerability under the RCP8.5 scenario showed a continuous decreasing trend. There will be a sharp decline in the mid-to long-term, with the area proportion dropping from 58.3% in the mid-term to 29.6% in the long-term ( Figure 9); compared with the short-term, the proportion of low physical vulnerability areas will be increased by 23.4% and decreased by 30.3% under the RCP4.5 and RCP8.5 scenarios, respectively ( Figure 10).
The above results showed that compared with the short-term, in the long-term, the high physical vulnerability area under the RCP4.5 scenario will decrease, while the low physical vulnerability area will increase; under the RCP8.5 scenario, the high physical vulnerability areas will increase, while the low physical vulnerability areas will decrease in the long-term compared with the short-term. That is, the performance of winter wheat physical vulnerability on the HHH Plain shows that the RCP8.5 scenario is higher than RCP4.5, and its overall spatial pattern is high in the south and low in the north. However, the overall physical vulnerability will decrease under the RCP4.5 scenario but increase under the RCP8.5 scenario over time.

Spatial pattern
The average annual winter wheat drought risk (Figures 11 and 12) was derived from the drought hazard and physical vulnerability calculated using the EPIC model with the input of the meteorological, soil, and other geographic environmental data in the HHH Plain under the RCP4.5 and RCP8.5 scenarios in 2010-2099 without considering irrigation measures.
The overall drought risk on the HHH Plain showed a spatial pattern of high in the south and low in the north. The drought risk from high to low is Huanghuai Plain agricultural region, Jiluyu low- lying plain agricultural region, Shandong Hilly agricultural and forestry region, Yanshan-Taihang Mountain piedmont plain agricultural region (Figure 11). The average drought risk was 0.211 (RCP4.5) and 0.207 (RCP8.5) from 2010 to 2099. Areas with medium and high drought risk (dR > 0.3, area accounting for approximately 23.4%) were mainly distributed in the Huanghuai Plain agricultural region and partly located in the northern Jiluyu low-lying plain agricultural Region; low drought risk areas (dR < 0.3, accounting for approximately 76.6%) are mainly distributed in the Yanshan Taihang Mountain piedmont plain agricultural region and Shandong Hilly agricultural and forestry region (Figures 11 and 12).

Temporal evolution
The high drought risk areas showed a continuous increasing trend under the RCP4.5 and 8.5 scenarios (Figures 13 and 14). Compared with the short-term, the proportion of high drought risk areas increased by 0.63% and 1.18% under the RCP4.5 and RCP8.5 scenarios, respectively, in the longterm. However, the low drought risk areas under the two scenarios show an increase first and then a decrease over time (Figure 13c, Figure 14c). In the long-term, the low drought risk areas increased by 1.69% and decreased by 3.28% in RCP4.5 and RCP8.5 compared with the shortterm (Figure 13c, Figure 14c).  Compared with the short-term, the high drought risk areas will increase under the RCP4.5 scenario, as well as the low drought risk areas; in addition, the high drought risk area under the RCP8.5 scenario will also increase, while the low drought risk will decrease in the long-term. That is, the drought risk of winter wheat on the HHH plain under the RCP4.5 scenario is higher than that under RCP8.5, and its overall spatial pattern is high in the south and low in the north. However, the regional drought risk difference increases under the RCP4.5 scenario, while the overall drought risk increases under the RCP8.5 scenario over time.

Discussion
This paper analyzed the drought hazard of winter wheat on the HHH Plain from 2010 to 2099. The results showed that the dHI in the RCP4.5 scenario will be higher than that in RCP8.5 (Figure 7), with a high spatial pattern in the south and a low spatial pattern in the north (Figures 5 and 6), which is similar to the spatial pattern of historical periods. For example, during the winter wheat growing season from 1996 to 2012, the comprehensive drought index in the southern part of the HHH Plain was higher than that in the north and the southern part showed a continuous drought trend Wang et al. 2018).
In addition, Liu, Chen, and Pan (2020a) also found that under the RCP4.5 or RCP8.5 scenario, the drought hazard in the southern HHH Plain from 2010 to 2099 will be higher than that in the northern HHH Plain, which may be related to the higher frequency of extreme drought in the south than the north. Yao et al. (2019) showed that the frequency, duration, and severity of droughts in the southern HHH Plain will be both higher than those in the northern HHH Plain in the shortmid-long term under the RCP4.5 and RCP8.5 scenario. The reason for the high dHI values in the south may be that the increase in the highest temperature and solar radiation in the south of the HHH Plain will be higher than that in the north, while the precipitation trend will be opposite in the future, which results in a higher drought hazard in the southern region (Xiao et al. 2020). Huang et al. (2018) found that the 12-month SPEI showed that the drought hazard under the RCP4.5 scenario on the HHH Plain will be higher than that under the RCP8.5 scenario in the short-to mid-term. The explanation is that the maximum temperature (RCP4.5-0.77°C, RCP8.5-0.54°C) and solar radiation (RCP4.5-2.19%, RCP8.5-1.7%) of RCP4.5 will increase more compared with RCP8.5, although a small increase in precipitation (RCP4.5-13.6%, RCP8.5-15.8%) in the short-term and maximum temperature and slight precipitation increases will occur with greater increases in solar radiation in the mid-long term. These changes will cause a higher drought hazard under the RCP4.5 scenario of the HHH Plain in the future (Qu et al. 2019). These analyses showed that with the effects of temperature, precipitation, solar radiation and extreme drought events, the drought hazard in the south of the HHH Plain will be higher than that in the northern region in the future and will also be higher under the RCP4.5 scenario than under the RCP8.5 scenario.
The results of the physical vulnerability of winter wheat on the HHH Plain from 2010 to 2099 showed that the pV presents a spatial pattern of high in the south and low in the north (Figures 8  and 9). The pV of winter wheat under the RCP8.5 scenario will be higher than that under RCP4.5 Figure 14. Wheat drought risk maps for the short-term (a), medium-term (b), long-term (c) and area ratio (d) under the RCP8.5 scenario.
( Figure 10). Similarly, Hu, Mo, and Lin (2015) indicated that the winter wheat yield loss is high in the south on the HHH Plain in 2050 under the RCP4.5 or RCP8.5 scenario. In addition, Lv et al. (2013) found that under full irrigation conditions, the vulnerability of winter wheat on the HHH Plain will be high in the south in the short-mid-long term under the A2, A1 and B1 scenarios. Sun et al. (2005) showed that under the B2 scenario, the pV of irrigated winter wheat on the HHH Plain will be higher in the southern region than in the north in the future, which may be related to the most essential impact of temperature on winter wheat yield in this region. For example, the impact of the average temperature, sunshine duration, and accumulated precipitation on winter wheat yield was −3.75%°C −1 , 0.05%(10 h) −1 and −0.01%(10 mm) −1 , respectively, in this region from 1981 to 2018 (Liu, Zhang, and Ge 2020b). The increase of temperature in the southern region will be greater than those in the north in the future (Xiao et al. 2020), which will bring a higher pV in the south. In the medium-long term, the RCP4.5 scenario will have smaller temperature increases than the RCP8.5 scenario (Qu et al. 2019), so pV is lower under the RCP4.5 scenario. The above analysis shows that under the dominant influence of warming, the pV in the south of the HHH Plain will continue to be higher than that in the north, and it will also be higher under the RCP8.5 scenario than under the RCP4.5 scenario.
The results showed that the dR of winter wheat on the HHH Plain will be high in the south and low in the north (Figure 11), which is consistent with the spatial pattern of the drought hazard ( Figures 5 and 6) and the pV (Figures 8-10). And the dR under the RCP4.5 scenario will be higher than that under RCP8.5 (Figure 12), which may be caused by the higher drought hazard under the RCP4.5 scenario (Figures 5 and 6). Based on the previous analysis, the increase of temperature dominates the change in the pV of winter wheat in the region; that is, the increase of temperature in the future will directly reduce the yield of winter wheat. Further analysis of the drought hazard in the region reveals that although there is more precipitation in the south, its temperature increases are also higher (Xiao et al. 2020), which intensifies the drought hazard of the south. Ultimately, the yield loss of winter wheat in the south will be more serious than that in the north. In other words, increasing temperature will indirectly reduce the yield of winter wheat by increasing drought hazards. Study has shown that every 1°C increase in temperature will cause a global wheat yield decrease of 4.1%-6.4% in the future (Liu et al. 2016). The above analysis shows that under the influence of climate change, increase of temperature is the essential factor threatening the yield of winter wheat on the HHH Plain and even the world.
It is worth to point out that the HHH Plain is the main winter wheat production region in China (Chang et al. 2020), while China, as the world's largest wheat production country, accounted for 13.46% of the world's total wheat production from 1960 to 2016 (FAOSTAT 2017;Zhang et al. 2015b). Therefore, winter wheat production in this region is very important worldwide. However, winter wheat production in this region faces a serious threat of drought risk Yue et al. 2015); in addition, wheat drought risk on the HHH Plain will further sharpen in the future under climate change (Chen, Zhang, and Tao 2018;Li et al. 2015b), and it has the highest drought risk in China and the world (Yue et al. 2018;Zhang et al. 2015a). Therefore, as the main winter wheat-producing region with high drought risk, the HHH Plain needs to proactively respond to the threat of winter wheat drought risk. This paper suggests that to reduce the drought risk of winter wheat directly or indirectly caused by temperature increase, governments should make joint efforts to reduce carbon emissions to reduce the global warming rate. On a regional scale, due to the high drought risk in the south of the HHH Plain, it is necessary to adjust the land use structure to change the spatial distribution of winter wheat planting in this region, as well as to reduce its drought exposure (Yue, Zhang, and Shang 2019) to avoid drought risk directly. In addition, from the perspective of winter wheat households, as temperature increase is the main factor affecting the vulnerability of winter wheat in this region, farmers should actively adapt to climate change, such as cultivating, promoting and using drought-and heat-resistant winter wheat varieties, and adopting reasonable irrigation and fertilization managements to adapt to climate change for reducing the drought risk, which results in severe yield reductions, to ensure global food security.
In the present study, the EPIC model plays a key role in assessing winter wheat drought risk. Therefore, the performance of EPIC model determines the reliability of our research results. We evaluated mean absolute error, mean relative error, and root mean square error (RMSE) ( Table  3) and provided a scatter plot with the 1:1 line (Figure 4) to validate the EPIC model, and the results proved reliable. Some other index such as normalized RMSE (NRMSE), can be a possibility to improve the model verification performance. Therefore, more indicators could be considered in future research to validate the reliability of model output.

Conclusion
Assessing the drought risk of agricultural sectors under various climate scenarios are of great significance for formulate reasonable strategies to adapt to climate change in advance to achieve sustainable agriculture.
We proposed a model regarding drought risk as a function of drought hazard, winter wheat physical vulnerability, and its exposure. The key innovation to this approach is a mechanism crop model, the EPIC model, was adopted to support risk assessment by simulating the winter wheat growth process and output the corresponding yield and water stress under climate scenarios. Our research findings demonstrated the ability of the EPIC model to assess crop drought risks caused by climate change upon a case study on the HHH Plain in China under RCP4.5 and RCP8.5.
It indicates that the winter wheat drought risk on the HHH Plain will exacerbate with changing climate. The average dR under the RCP4.5 scenario will be slightly higher than that under RCP8.5, i.e. 0.211 and 0.207, respectively. The high drought risk areas will increase by 0.63% and 1.18% under the RCP4.5 and RCP8.5 scenarios, respectively, by the end of this century. Meaning that the area of high drought risk in the HHH Plain is predicted to be increase over time.
Spatially, the average drought risk of winter wheat is high in the south and low in the north. It is arranged from high to low as follows: Huanghuai Plain agricultural region, Jiluyu low-lying plain agricultural region, Shandong Hilly agricultural and forestry region, and Yanshan Taihang Mountain piedmont plain agricultural region. From this point of view, we suggest that the strategy of future winter wheat drought risk prevention in different regions of North China Plain should formatted according to local conditions. We found that it is not precipitation, but increasing temperature will leads to increasing winter wheat drought risk in HHH Plain. Therefore, we argue that when assessing the risk of crop drought, more attention should be paid to the adverse effects of warming, especially extremely high temperature.
We suggest that, for effectively managing future drought risk of winter wheat and achieving the goal of ensuring food security, the central governments should strive to achieve the expected temperature rise targets, local governments should actively avoid high drought risk areas for planting winter wheat, and farmers should make full use of drought and heat resistance winter wheat varieties supplemented by farmland management measures such as irrigation and fertilization.

Disclosure statement
No potential conflict of interest was reported by the author(s).