Risk assessment of possible impacts of climate change and irrigation on wheat yield and quality with a modi ﬁ ed CERES-Wheat model

The effects of climate change on yield and quality in different climate regions have high uncertainty. Risk assessment is an effective measure to assess the seriousness of the projected impacts for decision-makers. A modi ﬁ ed quality model was used to simulate integrated impacts of climate change, environment, and management on wheat yield and quality. Then, the Canadian Earth System Model version 5 (CanESM5) was used to forecast the daily meteorological data, and the Statistical Downscaling Model (SDSM V5.2) was used for downscaling. The modi ﬁ ed CERES-Wheat was combined with the forecasted meteorological data to simulate the future wheat yield and grain protein concentration (GPC). The risk to wheat yield and quality in three climatic regions in Northwest China under two climate change scenarios of the CanESM5 was assessed. The average temperature increased by 0.22 – 3.34 (cid:1) C, and precipitation increased by 10 – 60 mm from 2018 to 2100. Elevated temperature and precipitation had positive effects on the yields. The risk to yield in most regions with climate change decreased by 3.8 – 25.1%. The risk to GPC in all regions with climate change decreased by 7.3 – 27.2%. Irrigation decreased the risk to yield greatly but had different effects in the three climatic regions. The risk to yield with irrigation decreased by 37.7 – 52.1%. In contrast to previous studies, in this study, the risk to GPC with irrigation substantially increased by 25.8 – 28.9% in humid regions and 3.9 – 8.8% in subhumid regions and decreased by 37.7 – 52.1% in semiarid regions. The irrigation should be discreetly applied for different climatic regions to combat climate change.


INTRODUCTION
In monsoon climates where the annual precipitation fluctuates greatly, the quality and yield of grain are seasonally variable because rainfall is unreliable and there is a significant risk of heat waves during the grain-filling phase (Pleijel et al. ; Ahmad et al. ). It is anticipated that under climate change, the growing-season rainfall in many arable cropping regions will change (Dubey et al. ), and there will be a greater incidence of extreme climatic events (IPCC ). The precipitation change in different climatic regions will not be the same because of climate change, but more extreme precipitation events will likely occur (Xu & Liu ). At the same time, climate change can result in entirely different effects on grain protein concentration (GPC) (Garofalo et al. ). GPC is a function of genes, environment, and management. GPC has important effects on nutrition, flour yield, and processing quality (Meng et al. ). A GPC of approximately 8% is suitable for pastry and cookies, and a GPC of 13-14% is suitable for leavened bread and pasta. Steamed bread and white, salted noodles (Chinese style noodles) should have a medium protein content (approximately 10%) (Addo et al. ). When the environment or management is less than optimal for wheat, the resulting GPC may not match the intended end use of the wheat (Albert et al. ).
The artificial climate chamber, open top chamber, and free-air carbon dioxide enrichment methods are the most widely used methods to study the effect of climate change on crop growth (Allen et al. ). However, these artificial methods are different from the natural environment or are very expensive. The combination of general circulation models (GCMs) and crop models assesses the effects of climate change on cropping systems and is more comprehensive and convenient, but it depends on the progress of the theory (Gilardelli et al. ). Crop models have often been used to evaluate the effect of climate change on yield ( Jabeen et al. ), but their effect on quality has been rarely reported (Nuttall et al. ). CERES-Wheat can interpret and simulate nitrogen movement and transfer along the soil-plant-atmosphere continuum, but it cannot output parameters related to quality such as GPC. Many crop models only simulate GPC by a harvest index or the constant nitrogen-to-protein conversion factor (Fp), 5.70 (Zhang et al. ). It is necessary to add a quality module for crop models to simulate grain quality with different stresses.
GCMs are used for climate prediction (Gaffin et al. ) but cannot be directly used for climatic studies because of their coarse spatial resolution (Rajabi & Shabanlou ). The Canadian Earth System Model version 5 (CanESM5) (Swart et al. ), based on the Coupled Model Intercomparison Project Phase 5 (CMIP5), and reanalysis from the National Center for Environmental Prediction (NCEP) have been verified in China (Qing et al. ). The simulation of precipitation and temperature had relatively better results (Song et al. ). Downscaling is necessary to extract location-scale information from the climatic data downloaded from GCMs (Meenu et al. ).
Statistical downscaling methodologies and dynamical downscaling approaches are common downscaling methods (Schmidli et al. ). Statistical downscaling methodologies have several advantages, such as low cost, easy operation, and rapid calculation (Timbal et al. ). The statistical downscaling model (SDSM V5.2) is a statistical downscaling methodology that has been widely used in Risk is a combination of two factors: the probability that an adverse event will occur and the consequences of the adverse event. Risk assessment encompasses an analysis phase and an implementation phase, risk treatment (Omenn ). By comparing the outputs of a multisimulation with a critical threshold, it is possible to evaluate the risk related to future climate conditions. Climate change has important effects on grain quality, but adaptation to climate change has long been neglected in terms of quality.
The most commonly used methods to alleviate the impact of climate change are changing the sowing date, using newly cultivated varieties, and implementing reasonable irrigation (Li et al. ; Worku et al. ). Irrigation is the most common and simple management approach. Irrigation can improve the nitrogen use efficiency of crops, but it has an adverse effect on winter wheat protein accumulation The wheat yield and quality response to climate change can also vary depending on the location and climate change scenario. Simulation modeling provides an opportunity to understand the broad-scale feedback between climate change and agro-production systems at a regional level. In comparison to other models, the modified CERES-Wheat model could better respond GPC to stresses, and the modified CERES-Wheat model was able to simulate GPC with abiotic stress from climate change more precisely. The aims of this study are to assess the risk of climate change to wheat yield and quality and address the risk from irrigation.

Experimental location and treatments
The field experiment, conducted from October 2014 to June 2017, was located in the Guanzhong irrigation area, and where the average annual precipitation is 580 mm, the average precipitation in the wheat growing season, the average annual temperature, and the annual sunshine duration are 203 mm, 13 C, and 2,196 h, respectively. The wheat variety 'Xiaoyan 22' (Triticum aestivum L.) is widely grown in northwestern and northern China. The four irrigation treatments (I0, rainfed; I1, 60 mm; I2, 120 mm; and I3, 180 mm) involved the whole plots. The four fertilization treatments (N0, 0 kg N ha À1 ; N1, 105 kg ha À1 ; N2, 210 kg N ha À1 ; and N3, 315 kg N ha À1 ) were the split-plot treatments.  where ΔKN pot is the potential kernel nitrogen accumulation, g N kernel À1 d À1 ; T max , T min , and T mean are daily maximum, minimum, and mean temperatures ( C), respectively.
where α is a coefficient of advection, dimensionless; SR is the daily total solar radiation, MJ m À2 d À1 ; ALBEDO is the crop albedo, dimensionless; E EQ is equilibrium evapotranspiration, MJ m À2 d À1 .
Modified CERES-Wheat output the yield, grain nitrogen yield at maturity (GNAM), water stress (S W ), and nitrogen stress (S N ). The calculation of Fp (Equation (4) where Fp is the nitrogen-to-protein conversion factor, dimensionless; S W is the water stress factor, dimensionless; and S N is the nitrogen stress factor, dimensionless.

Weather data
Daily maximum and minimum temperature ( C), precipitation (mm), and solar radiation (MJ m -2 d À1 ) are the essential weather data for CERES-Wheat ( Jones et al. a, b). The weather data included historical weather data and climate scenario data.

Historical weather data
Historical weather data   These data included 26 atmospheric predictor variables (e.g., mean sea level pressure, 850 hPa zonal velocity, total precipitation, and mean temperature at 2 m) for 1961 À 2017 and GCM projections for the SSP2-4.5 and 5-8.5 from 2018 to 2100.

Calibration and validation
Calibration was completed running GLUE (He et al. ).
Related data were from a treatment of sufficient irrigation and sufficient fertilization (I3N3) in 2014.
Model validation was completed by comparing measured and simulated soil water content, yield, and GPC for 2015- 2017. Three evaluation indicators were used to test the accuracy. The percent deviation (d) was smaller, and the simulated values were better. The smaller the relative root mean square error (RRMSE) (Equation (6)) was, the smaller the difference in the measured and simulated values (0-10%, excellent; 10-20%, good; 20-30%, fair and >30%, poor). In addition, r 2 was used to assess the downscaled climatic projection, and the best index was 1.
where s i is the simulated value, o i is the observed value, and n is the sample size.

Scenarios
Due to the long and narrow topography in Shaanxi Province, different cropping systems are used in different regions. To ensure the results were explanatory and uniform, the management parameters were set as follows: fertilization was 210 kg N ha À1 before sowing; the sowing dates were 1-15 October when the daily average temperature was above 20 C and precipitation was less than 2 mm; the cultivation method was rotary tillage; the plant population density was 240 seeds m À2 ; and the harvesting time occurred when winter wheat had matured. The scenarios included irrigation (60 mm during wheat wintering and 60 mm during the elongation stage) and rainfed (no irrigation). Soil data from 12 locations were used for modified CERES-Wheat and collected from the China Soil Database (http://vdb3.soil.csdb. cn/) ( Table 1). The change in CO 2 concentration was not considered. A greater uncertainty in simulating wheat yields under climate change was due to the response mechanism of rising CO 2 concentrations than to the variations among the downscaled GCMs (Asseng et al. ).

Risk assessment
To estimate the risk to wheat yield and quality, the yield and GPC were compared with a critical threshold, which is calculated as the 60-year mean yield and GPC for the historical period . The risk to wheat yield and GPC shortfall was then defined as the relative frequency of yield and GPC below the threshold, representing the likelihood of future yield and GPC being lower than the historical mean yield and GPC. The risk to yield (Equation (7)) and GPC (Equation (8)) were estimated as: (8) where R and R 0 are the risk to yield and GPC, respectively, %; M and M 0 are the determination factors of yield and

Calibration and validation
Modified CERES-Wheat The percent deviation values between the simulated and observed phenological period, yield, grain nitrogen concentration, and evapotranspiration (ET) were all less than 10%, implying that the model simulated the phenology, wheat growth, nitrogen transformation, and water transformation well ( Table 2). The genetic coefficients were relatively stable and could be used in the same or similar 12 sites.

Statistical downscaling model
The daily maximum temperature, daily minimum temperature, monthly total precipitation, and monthly total radiation downscaling were selected as the predictor variables. The SDSM was calibrated from 1961 to 1990 using these predictors. To validate the reliability of the calibrated SDSM, the simulated result from 1991 to 2017 was compared with the observed data from 1991 to 2017. The coefficient of determination (r 2 ) for these linear relationships varied from 0.64 to 0.93 (Table 3). A very significant power correlation between the predictand and predictor variables (p < 0.001 level) was evident.

Analysis of historical and future climate
The terrain of Shannxi is long and narrow, and the latitude range is very large. The annual average temperature (1957-2017) is 2.8-11.6 C during the wheat growing season. The temperature to the south is higher and gradually decreases from south to north (Figure 3 The precipitation during the growth period gradually decreased from south to north. The available precipitation in the semiarid region during the growth period was scarcer than that in the other regions (Figure 3(b)). Precipitation in all regions increased under SSP2-4.5. However, for some locations, such as Wuqi, precipitation decreased very little.
The precipitation situation was similar to SSP5-8.5. The average precipitation during the growth period increased

Winter wheat yield and risk
Winter wheat yield with latitude increasingly showed a tendency of increasing first and then decreasing. The average historical yield at the 12 locations ranged from 3,849 to 9,434 kg ha À1 and had extreme variation under rainfed conditions. The maximum yields were 9,434 and 9,025 kg ha À1 in Fuping and Shangluo in the humid region, respectively.

Grain protein content and risk
The historical GPC decreased from south to north first and then increased (Figure 7(a)). That of the humid and semiarid regions increased to 11.69 and 11.88% under SSP2-4.5, respectively. However, the GPC in the subhumid region decreased to 10.94% (Figure 7(b)). With the continuously increasing radiative forcing, the GPC of the humid and semiarid regions unexpectedly decreased to 11.49 and 11.27%     In this study, we first used modified CERES-Wheat to simulate GPC using the nitrogen-to-protein conversion factor. Then, CanESM5 was selected to simulate daily meteorological data, and SDSM was used to downscale the data. Modified CERES-Wheat was combined with the simulated meteorological data to simulate the future wheat yield and quality in Shaanxi. The temperature and precipitation increased under SSP2-4.5 and SSP5-8.5. It is generally acknowledged that a temperature increase is beneficial to GPC, whereas increases in precipitation and irrigation are beneficial to yield but harmful to GPC (Horstmann ). How yield and GPC change still needs further study. An irrigation scenario was also used as part of the strategy for responding to climate change. In the CERES-Wheat and other crop models, there was a correlation between photosynthetic intensity and CO 2 ( Jones et al. a, b). Therefore, the yield was too high, and the GPC was too low, which differed from the actual situation. The mechanism by which crop growth responds to increases in CO 2 is uncertain (Asseng et al. ). Thus, increasing CO 2 was not considered as a factor in this study, which may make the results uncertain.

Evaluation of the simulation of GPC and yield
An accurate simulation of phenology ensured wheat-filling duration, which is the foundation of the accumulation of protein and dry matter, and the simulated yield was underestimated without nitrogen fertilization (N0). Yao et al.
() revealed that CERES-Wheat undervalues the capacity to resist stress.
The simulation of GPC was very good with severe water and nitrogen stress because GPC is the ratio of grain yield to grain nitrogen yield. The accumulation of nitrogen and dry matter during the grain-filling stage is more affected by temperature than by water and nitrogen stress (Ritchie & Otter ). However, the simulated results of the GPC with slight nitrogen stress (N1) were overestimated. Asseng et al. () found that the accumulation rate of nitrogen was overestimated when the average temperature was less than 10 C.

Effects of climate change and irrigation on yield and risk
Precipitation and temperature increased with climate change, but the yields of the different climatic regions tation is insufficient, then drought stress will be stronger (Panozzo et al. ).
In comparison to the other regions, the humid region had greater water and heat resources. Temperature and water were not the most important limiting factors for this region. The temperature and precipitation increased, but the yield only slightly decreased or increased. The reason may be that the yield in the humid region was at a relatively higher level and had largely reached the potential yield, and the excess precipitation decreased radiation.
The study of Wang et al. (2012) showed similar results: reference crop evapotranspiration and crop-water requirements significantly decreased under climate change.
Irrigation was still good for yield but only played a limited role. The temperature in the subhumid region increased and precipitation decreased under SSP2-4.5, while the winter wheat yields increased. Zhang and Yan (2003) recognized that although precipitation decreased, the appropriate increase in temperature promoted the growth of winter wheat. Radiation and temperature had higher suitability for yield in the subhumid region than in the other regions, and these factors increased under climate change (Jing & Fu-ning ). Decreasing precipitation became the main limiting factor for wheat yield, and water could not match the increasing temperature. Much radiation was wasted. Thus, the increasing yield was mainly due to irrigation.

Effects of climate change and irrigation on GPC
Temperature and precipitation increased with increasing latitude, but the GPC decreased first and then increased.
High temperatures can affect the formation of carbohydrates and proteins, influencing both the GPC and yield.
The optimum temperature for grain dry matter accumulation is 20.7 ± 1.4 C during the grain-filling period (Porter & Gawith ), and the optimum temperature for grain nitrogen accumulation is 35 C (Jones et al. a, b). At temperatures >30 C, the rate of starch accumulation decreases, resulting in a higher GPC (Jenner a).
The humid region was relatively more suitable for the accumulation of protein, and the subhumid region was more suitable for the accumulation of dry matter. The semiarid region had temperature and precipitation that were lower than optimal, and elevated temperature and precipitation were beneficial to the accumulation of protein and dry matter in this region.
The GPC of the humid and semiarid regions increased slightly under SSP2-4.5 and SSP5-8.5, but the reasons for the increase were not the same. A relatively higher tempera- (3) Irrigation decreased the risk to yield greatly in all regions but had totally different effects in the three climatic regions. The risk to yield with irrigation decreased 37.7-52.1% in different climates. The risk to GPC with irrigation increased greatly by 25.8-28.9% in the humid region and by 3.9-8.8% in the subhumid region and decreased by 37.7-52.1% in the semiarid region.

DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.