Assessing the impact of climate change in the wheat–maize cropping system across the Huang–Huai–Hai Plain under future climate scenarios

The increase in greenhouse gas emissions due to anthropogenic activities is no doubt an undisputable contributor to global climate change. The International Panel on Climate Change (IPCC) issued a warning in 2018 about anthropogenic warming, indicating an increase of 1.5 °C in global temperature between the years 2032–2052, without major cuts in carbon dioxide (CO₂) emissions (IPCC 2018). Agriculture and food systems are likely to be most affected by climate change and variability because of their high dependence and sensitivity to climatic conditions (Motha & Baier 2005). There are also concerns that climate change may lead to a loss in water and consequently will negatively affect crop yields of staple cereals such as wheat, rice, and maize to varying degrees depending on regions and latitudes. The IPCC special report ‘Climate Change and Land’ (IPCC 2019) has reported adverse impact of observed climate change on wheat and maize yields in many lower-latitude regions in recent decades, unlike higher latitudes where yields have generally increased. Maize and wheat projection statistics at the global level confirm this trend, showing a yield increase for wheat and maize at high latitude and a decrease at low latitude by the end of the century (Rosenzweig et al. 2014).

Agriculture in China, particularly Northeast China, has been one of the most sensitive areas to climate change in the country during the last five decades (Lv et al. 2015). It is projected that climate change will have a significant impact on China's agriculture, as with each degree-Celsius increase in global mean temperature, the loss in yield is estimated to be about 8.0% for maize and 2.6% for wheat for the period 2071–2100 in comparison with the 1981–2010 baseline. Considering the global projections, it is important to analyze the impact of climate change on wheat and maize yield in the Huang–Huai–Hai (3H) Plain, which is one-sixth (23.3 million ha) of China's total cultivated area (134.86 million ha).

Climatic parameters, such as temperature and precipitation, have a direct impact on crop yield. The rising temperatures can also affect crop production negatively due to shortened growing seasons and decreased photosynthetic accumulation in plants (Lobell & Field 2007). Another effect of heat stress is the increased transpiration resulting in higher plant water demand. Similarly, changes in rainfall patterns also affect yield, particularly in rain-fed cropping areas. The comprehensive effect of climate change on plant productivity understandably depends on the interaction of all these factors. Therefore, to ensure suitable site-specific adaptation measures in the 3H Plain, it is critical to quantitatively estimate the impact of changes in all climatic variables on agricultural production and to map the spatiotemporal variability of crop yield to empower the resource managers with the information to ensure best management strategies.

Several studies in China have conducted statistical downscaling and employed process-based crop models to study the impact of climate change on crop yield and change in irrigation demand (Tang et al. 2018; Liu et al. 2021). The relationship between climate variability and crop yield has been confirmed in many studies using statistical approaches. To simulate crop yield using process-based models, climate inputs are required at a finer spatial resolution, and therefore, global or regional scale projections are not adequate. Therefore, crop models in combination with downscaled future climate data from global circulation models (GCMs) are an important tool to assess the crop response to various climate scenarios, irrigation regimes, and agronomic management conditions.

Even though a lot of research has been done in China related to statistical downscaling and the impact of change in climatic variables on meteorological patterns, the impact of climatic changes has been least studied in crop rotation systems, particularly wheat and maize. Therefore, there is a need to quantify and assess the impact of climate change on yield to further expand the knowledge for climate adaptation potential in agricultural applications. The goal of this study was to understand the impact of climate change in the wheat–maize cropping system at 90 study locations in the 3H Plain. The specific objectives of the study were to (1) produce downscaled climate data for the selected study sites; (2) explore the change in temperature and precipitation during winter wheat and summer maize crop growth periods, and (3) understand the impact of climatic changes on the wheat and maize yield in the 3H Plain. The impact of climate change results has been produced for medium (RCP4.5) and high (RCP8.5) emission scenarios (2016–2030, 2031–2070, and 2071–2099). To simulate the crop yields, the AquaCrop Model was used for historical and future timelines.

The data used in this study included meteorological data (1978–2018) and GCM data (2016–2100). The meteorological data used in this study were the China Meteorological Forcing Dataset (CMFD) developed by the National Tibetan Plateau Data Center (TPDC), Beijing, China. This data had a spatial and temporal resolution of 0.1° and 3 h, respectively, which were updated and evaluated regularly by the TPDC (He et al. 2020). The dataset was released in 2019 and made through a combination of remote sensing products, a reanalysis dataset, and in situ observation data at weather stations. The daily mean, maximum, and minimum temperatures and precipitation data for the period of 1979–2018 were used to generate future climate predictions. The data from the TPDC climate forcing product were extracted using China Meteorological Administration (CMA) ground weather observation station locations issued in its list available at the CMA website (http://data.cma.cn/). The TPDC data have shown a high correlation with the data available from the ground observation stations of the CMA and can be used for a wide range of applications at a finer resolution.

The second-generation Canadian Earth System Model (CanESM2) consists of the physical coupled atmosphere-ocean model CanCM4 coupled to a terrestrial carbon model (CTEM), and an ocean carbon model (CMOC) was used for the period 2016–2100. CanESM2 predictor's data were developed by the Canadian Centre for Climate Modeling and Analysis (CCCma) of Environment and Climate Change Canada representing the IPCC Fifth Assessment Report (AR5). The predictor data were available in the 128×64 grid cells covering the global domain according to the T42 Gaussian grid at the uniform horizontal resolution of 2.8125° along the longitude and the latitude (available at http://climate-scenarios.canada.ca). The National Centers for Environmental Prediction (NCEP) reanalysis climatic data were used to identify appropriate atmospheric variables and calibrate and validate the model predictions in historical time, and CanESM2 climatic data were used to generate RCP4.5 and RCP8.5 scenario datasets.

Statistical Downscaling Model (SDSM) 4.2, open-source software, which is a decision support tool to assess the local climate change impacts using a robust statistical downscaling technique, was used in this study. The SDSM supports the creation of multiple, low-cost, and site-specific scenarios from daily weather variables, such as maximum and minimum temperatures, precipitation, and humidity under current and future regional climate forcing. The SDSM also produces a range of statistical parameters such as variances, frequencies of extremes, and spell lengths. The statistical software provides the support for predictor (weather station historical data) pre-screening, calibration, diagnostic testing, statistical analysis, and graphing of the climate data. The SDSM has been applied in various geographical settings across the globe to produce high-resolution climate change future scenarios including various parts of China. In this study, the SDSM was used to downscale the climate data of each station by establishing the statistical relationship between predictors and predictands using multi-linear regression and stochastic bias correction techniques.

The data processing was divided into five steps, namely, predictor selection, the calibration of daily weather predictands using selected GCM predictors, the generation of model predictions for the historical period (1979–2018), the validation of the model predictions with observed data, and the generation of the future downscaled weather series based on the best-identified predictors. Before any data processing, quality control of observed data for all weather variables was done for each study location and ensured that there were no data gaps.

First, the empirical relationship between the CMFD predictands (e.g. temperature and minimum and maximum temperatures) and the NCEP predictors (e.g. specific humidity, precipitation, geopotential height, and mean sea level pressure) was identified, which varied over space and time. NCEP predictors showed a high correlation with the observed historical weather data; therefore, they were used for calibration and validation of the SDSM output results.

Suitable NECP predictors (from Table 1) were first selected based on the strength of their empirical relationship with predictands (temperature and precipitation). Correlation matrix, partial correlation, and p-value statistics were used to define the strength of the relationship between predictor and predictand, which varied over space and time. The probable predictors displaying significant correlation at 95% confidence level with the observed data were selected. Furthermore, the exploration of predictor and predictand relationship was also carried out using scatter plot analysis. In this study, weather data for 90 locations were screened for three weather variables (precipitation and maximum and minimum temperatures). Therefore, 26 large-scale predictors were analyzed more than 7,020 times (90 stations × 26 large-scale predictors × 3 predictands).

The second step was the calibration of the SDSM using the selected predictors for a specific predictand at each of the study locations (Figure 1). For each predictand, the model was calibrated under an unconditional process for maximum and minimum temperatures and a conditional process for precipitation on a monthly scale. Unconditional processes follow as a direct link between large-scale predictands and local-scale predictors, while conditional processes depend on an intermediate variable between predictands and predictors. Therefore, the temperature was modeled as unconditional and precipitation as a conditional process.

In this study, the downscaling process was carried out for all 90 study locations (Figure 1), and 20 ensembles were produced for each variable and location. The mean of 20 ensembles has been taken by writing the code in R for further use in the study (Supplementary material S1). Ideally, the model is calibrated using part of the observed available data, withholding the remainder of the data for independent model validation. In this study, the observed data were available for the period 1979–2018 (40 years). Half of the data (1979–1998) were used for calibration and the remainder was held for independent model validation.

For the historical period (2008–2016), the growth of winter wheat and summer maize was simulated based on the soil data, field management, and crop phenology data obtained from the published literature for the Xiaotangshan County in Beijing (Jin et al. 2014); Luancheng County in Hebei (Umair et al. 2017); Xinxiang County in Henan (Tang et al. 2018); Taian County in Shandong (Bian et al. 2016) and Guangde County in Anhui (Chen et al. 2016). The literature also provided sufficient information regarding sowing and harvesting dates, irrigation timing, amounts, and other management measures needed for calibration and validation of the AquaCrop Model (Shirazi et al. 2021) (Table 2). The calibration process also involved adjusting the non-conservative parameters, which included initial canopy cover CC_o (%), canopy growth coefficient (CGC), coefficients for triggering water stress affecting leaf expansion, and canopy senescence until a close match between observed and simulated yield and biomass was obtained. The performance was evaluated by comparing the simulated grain yield results and the observed grain yield data obtained from the literature under full irrigation for both winter wheat and summer maize during the 2011–2016 period. The AquaCrop model was then used to simulate the potential yield of winter wheat and summer maize using downscaled data for future scenarios for the period 2016–2099.

The projected changes in the weather variables (precipitation and maximum and minimum temperatures) and potential yield for both future climate scenarios (RCP4.5 and RCP8.5) were analyzed for three timelines: 2030s (2016–2040), 2050s (2041–2070), and 2080s (2071–2099). Under both RCP4.5 and RCP8.5 scenarios, the increase in precipitation and mean, maximum, and minimum temperatures was projected relative to the base period (1981–2016). The period from early October to mid-June is considered as the winter wheat growth period (WGP) and mid-June to the end of September as the summer maize growth period (MGP) as this cycle is widely practiced in wheat and maize rotation areas across the 3H Plain, but in the lower regions of the 3H Plain, the maize harvest is around mid-September. In this study, we considered wheat and maize crop cycles based on actual agronomic conditions that varied between different provinces in the 3H Plain. The ordinary Kriging method was used to analyze the changes by generating spatial distribution maps for all variables in ArcGIS-V10.1.

For R², the results closer to 1 indicate a perfect fit. For RMSE, a result close to zero indicates a better fit of the model. MAE measures the absolute value of the difference between the simulated value and the observed value. The values can range from 0 to infinite. Lower values represent a better agreement between predicted and observed values. Index of Agreement (IOA) measures how well the model produced the estimates, the value of 1 indicates a perfect match between observed and predicted and 0 indicates no agreement at all.

The model results of the bias-corrected weather data for the validation period (1999–2018) were evaluated based on the R², RMSE, MAE, percent bias, and IA. For the mean monthly sum of the precipitation, R² (0.925–0.993), RMSE (0.237–1.007), MAE (0.237–1.007), percent bias (±3.7%), and IA (0.954–0.996) were in the acceptable range at 0.05 significance level, which indicated that the model had the feature of high fitting precision.

Crop yields were simulated for both winter wheat and summer maize using agronomic practices mentioned in Table 2 (section 2.5), which are widely used across the 3H Plain. For all locations of winter wheat, the percent deviation of grain yield estimates from the observed data was between 0.42 and −11.0% for both the calibration and validation stages. Similarly, for summer maize, the percent deviation of grain yield estimates from the observed data varied between −1.44 and −8.90% for both calibration and validation stages, respectively. The RMSE between the observed and simulated grain yield was 0.12–1.01 t ha⁻¹ for winter wheat, while it was 0.50–0.53 t ha⁻¹ for summer maize. Overall, the calibration results show a reasonably close match between the observed and those simulated by the model for both winter wheat and summer maize.

To better understand the functional relationship between independent variables (precipitation and maximum and minimum temperatures) and dependent variable (yield), we conducted multiple regression analysis to know the strength of the relationship (Table 3) and the projected change in yield with the change in climatic factors. The analysis was carried out to better understand the relevant contribution of each of the independent variables to the total variance for yield. The results in Table 3 show the strength of relationship between the wheat and maize yield and independent climatic variables and their significance under both RCP4.5 and RCP8.5. The relationship of winter wheat was significant (p ≤ 0.05) with precipitation and maximum and minimum temperatures under both RCPs for all time periods. For summer maize, the negative relationship of yield is prominently significant (p ≤ 0.05) with minimum and maximum temperatures under both RCP scenarios.

The summary of the entire regression model can be seen in Table 4, which shows that a significant amount of variation in yield can be explained by the change in our independent climatic variables.

The regression coefficients were produced to understand the effect of change in independent variables on yield under RCP4.5 and RCP8.5 (Table 5). Finally, to better understand the perspective, we used the climate estimates (precipitation and maximum and minimum temperatures) under RCP4.5 and RCP8.5, and analyzed the effects of changes on yield separately and combined for all time periods (Table 6). For winter wheat, irrigation is widely applied and has a significant effect on crop yield; therefore, we included irrigation as part of the analysis.

For winter wheat, an increase of every 1% in precipitation is contributing to increase the yield by 0.68%, while the increase in maximum temperature gives a decrease of 0.72% of yield under RCP4.5. The increase in 1% minimum temperature is increasing the wheat yield by 0.39% under RCP4.5, while the increase is merely 0.13% under RCP8.5, which indicates that the inhibitory effect of increased maximum temperature is more pronounced as compared to the minimum temperature. With every 1% increase in irrigation, the increase in wheat yield is 0.30% under RCP4.5 and 0.20% under RCP8.5. In terms of the comprehensive effect of 1% change in climatic variables and irrigation is beneficial to the wheat yield by an increase of 0.65% under RCP4.5 and 0.54% under RCP8.5.

We used estimates of climate change to obtain the effects on the projected yield under RCP4.5 and RCP8.5 (Table 6) for both winter wheat and summer maize. For winter wheat, an increase in the maximum temperature gives a decrease in yield, while an increase in precipitation and minimum temperature increases the yield under RCP4.5 for the whole study area. A similar effect has been observed under RCP8.5, but this increase is relatively less as compared to RCP4.5. The comprehensive effect of an increase in all climatic variables will benefit wheat yield with an increase under both RCP scenarios. The positive effect of the increase in climatic variables will increase gradually from the 2030s to 2080s, and the study area experiences a substantial increase in precipitation counteracting the negative effect of rising temperatures during the winter WGP.

For summer maize, the increase in precipitation and maximum and minimum temperatures is projected to decrease yields under RCP4.5. The effect of increased precipitation under RCP8.5 becomes beneficial to increase the yields, but the increased maximum and minimum temperatures are projected to decrease the yields. The comprehensive effect of all variables is projected to decrease maize yields under both RCP scenarios. The highest decrease in maize yields was projected in the 2080s under both RCP scenarios.

It is noted that there are large uncertainties in projections generated by various models that stem from various sources, such as downscaling models, methods and bias correction, emission scenarios, and time period considered in the study. Since agricultural applications require higher precision to avoid the accumulation of error in the simulated yield from crop models, these uncertainties make use of these downscaled products challenging in agricultural applications. Greater consensus between predictors and climate models is expected to improve the forecast results. Therefore, this study dealt with the uncertainty by evaluating the model performance using various evaluation methods and repeated the process several times to improve the model performance and reduce the bias.

Downscaled climate change predictions in this study show a systematic increase in mean temperature and precipitation during the growth period of both winter wheat and summer maize under all emission scenarios. The substantial increase in all climatic variables is projected during the summer MGP as compared to the winter WGP. According to Ding et al. (2007), under varied emission scenarios, China's average annual temperature projected increase is 1.5–2.1 °C by 2020, 2.3–3.3 °C by 2050, and 3.9–6.0 °C by 2100, and an increase of 10–12% in precipitation relative to the base period 1961–1990. It has also been analyzed that daily minimum temperature is projected to increase more rapidly than daily maximum temperature, leading to an increase in daily mean temperature. This will result in a greater likelihood of extreme events and can have detrimental effects on crop grain yield. Geng et al. (2019) projected an increase in precipitation in the 2060s during the winter WGP by 17.31 and 22.22% under RCP4.5 and RCP8.5 in the North China Plain (NCP) region, while our results show an increase of 1.33–4.17% under RCP4.5 and 3.13–8.61% under RCP8.5 for the period 2016–2099 for the winter wheat growing season. The reason can be due to variation in the selection of different study sites. In general, various statistical models indicate that northern China will have more precipitation in the future and significant warming during the 21st century (Ding et al. 2007). An increase in precipitation is thought to be beneficial for wheat and maize growth, reducing the irrigation burden, but an increase in temperature can reduce the yield. Since the mid-1990s, increasing climate warming in the NCP has increased the probability of drought by an average of 10.1% (Hu et al. 2014). The warming trend will have a significant effect on drought occurrence in the summer MGP in the NCP region (Hu et al. 2014). Such changes can cause considerable changes in irrigation practices and grain yield for both of these crops in the NCP region.

The statistical models for China estimate that without climate adaptation, CO₂ fertilization, and genetic improvements in the crop cultivars, each degree-Celsius increase in global mean temperature will result in the loss of yield of 2.6 ± 3.1% for wheat and −8.0 ± 6.1% for maize (Lobell & Field 2007). During the 1981–2016 period, in absolute terms, the contribution of climatic factors to winter wheat yield was estimated to be around 0.76–1.92%, while the contribution of technological advancements has been estimated to have a contributed 27.19–60.43% increase in agricultural yield (Geng et al. 2019), which indicates the effective adaptation strategies already in place. In the case of summer maize, Xiao et al. (2020) suggest that due to climate changes, maize yield will decrease in China by 2.3–2.4%, but water-use efficiency will increase by 11.8–26.6% if no adaptive strategy is put in place. In most cases, warm temperature increases the crop phonological development resulting in early senescence and reduced grain yield and biomass. The direct negative temperature impact on yield could be additionally affected via indirect temperature impacts. For instance, increasing temperature will increase atmospheric water demand, which could lead to additional water stress from increased water pressure deficits, subsequently reducing soil moisture and decreasing yield. It is also suggested that increased temperature impacts the evapotranspiration rate, leading to a faster loss in soil moisture and an increased need for irrigation. So, the areas that are likely to get wetter during summer in the MGP are also likely to experience temperature-driven drying.

The elevated temperatures are expected to accelerate the crop growth leading to shorter crop cycles (Salman et al. 2021). On the provincial scale, changes in climatic factors can vary from North to South in the NCP region which can have a varying effect on wheat yield. Geng et al. (2019) suggested that a 1% increase in mean temperature during the WGP can lead to a loss of 0.109% of winter wheat yield per unit area keeping other factors constant, and a 1% increase in precipitation will increase the wheat yield by 0.186% to a point where it becomes harmful (Geng et al. 2019). However, the cumulative effect of temperature and precipitation on winter wheat will be positive under RCP4.5 and RCP8.5 by 3.22 and 4.13%, respectively, for the period 2021–2050. The climatic factors will positively contribute to the winter wheat yield in the Jing–Jin–Ji region, Henan and Shandong. Our results also show similar spatiotemporal changes and much of the increase in wheat yield is concentrated in the central and south 3H region.

In another study, Tang et al. (2018) used the Decision Support System for Agrotechnology Transfer (DSSAT) model in the 3H Plain and estimated that the potential yield, crop water requirement (ET_c), and effective precipitation during winter wheat growing seasons might increase in the future under RCP4.5, while irrigation water requirements would decrease. The wheat and maize crop productivity has also been estimated in a study based on the potential crop evapotranspiration and water availability in the NCP. Under full irrigation when water demand is met, the wheat yield ranges from 6.9 to 10.2 t ha⁻¹ and increases with latitude. Under rain-fed conditions, the wheat yield is much lower in all areas ranging from 0.6 to 5.2 t ha⁻¹ and decreases in low-latitude areas due to decline in annual rainfall. Under full irrigation, the maize yield ranges from 9.2 to 13.6 t ha⁻¹, and under rain-fed conditions, it ranges from 6.1 to 13 t ha⁻¹ with the lowest yield in the middle of the NCP due to the lowest amount of rainfall (Wang et al. 2008). This is consistent with the results of the current study where winter wheat yield increase is estimated largely in the central and lower 3H region, while summer maize yield is projected to decline in areas with reduced rainfall, mainly in north parts of the region.

The seasonality of the climatic changes also points toward the change in the length of crop cycles for both wheat and maize. The reduction of the wheat crop cycle from 242 days in the base period to 239 and 235 days under RCP4.5 and RCP8.5, respectively, indicates the effect of 1.71–1.28 °C warming. The summer maize crop length has declined substantially from 110 days in the base period to 104 and 100 days under RCP4.5 and RCP8.5, respectively. Liu et al. (2021) has also shown that the increase in accumulated thermal temperature will reduce the vegetative and reproductive growing period of wheat in the 3H Plain. Sharafati et al. (2022) have projected the impact of climate change on the variability of crop cycle length, crop yield, and water productivity in Iran. The results suggest an increase in wheat yield (14–54%), a decrease in the crop cycle length (1–12%), and an increase in water productivity (9–96%) in the future (2021–2080) as compared to the baseline period (1985–2016) (Sharafati et al. 2022).

In addition to temperature impacts, combined with the summer and autumn rainfall in the 3H region in the year 2021, the seasonal rainfall in the maize harvest period was too frequent and strong, resulting in floods and seriously affecting the maize production and also winter wheat sowing. According to the results of this study, the increase in precipitation may bring flood disaster risk while alleviating the contradiction of water shortage in agricultural production. The change in water budget during the winter WGP may not be very significant, but, during the summer MGP, it is expected to improve from 109 mm in the 2030s to 126 mm in the 2080s under RCP4.5 and 107 mm in the 2030s to 163 mm in the 2080s under RCP8.5 (Shirazi et al. 2022). The study of climate change impacts offers a good choice to change the cropping system for reducing the planting area of water-consuming crop (e.g. winter wheat) in those over-pumping areas to balance groundwater use and crop yield. The timely devised adaptive and mitigation strategies are the key to minimize the negative impacts of disasters on wheat and maize crops.

The study employed a crop simulation model combined with future climate data downscaled at 90 study locations across the 3H Plain to simulate changes in temperature and precipitation and their impact on wheat and maize yield and the cropping season within the region. The simulation results using the medium- and high-emission scenarios reveal that wheat and maize yield is sensitive to changes in temperature and precipitation. The projected climatic data indicate an increase in the average temperature throughout all RCP scenarios across the 3H Plain for both the winter WGP (1.17–1.21 °C under RCP4.5 and 1.17–1.28 °C under RCP8.5) and the summer MGP (1.29–1.92 °C under RCP4.5 and 1.84–2.08 °C under RCP8.5). The projected precipitation for both medium- and high-emission scenarios has been estimated to increase in the 2030s, 2050s, and 2080s during the WGP and the MGP. The comprehensive effect of an increase in all climatic variables will benefit wheat yield with an increase up to 1.74% (0.25 t ha⁻¹) and 3.00% (0.6 t ha⁻¹) under RCP4.5 and RCP8.5, respectively. The positive effect of the increase in climatic variables will increase gradually from the 2030s to the 2080s. It is likely that a substantial increase in precipitation will counteract the negative effect of rising temperatures during the winter WGP, particularly in Shandong and parts of Henan in comparison to other provinces. The northern areas of the 3H plain will have crop water requirements similar to present conditions, leading to slight variations in wheat yield. For summer maize, the comprehensive effect of all variables is projected to decrease the yield up to 4.44% (0.5 t ha⁻¹) and 5.40% (0.6 t ha⁻¹) under RCP4.5 and RCP8.5, respectively, by the end of this century. The loss of maize yield is much more pronounced in the north 3H region due to an increase in mean temperatures, despite any substantial change in precipitation. The highest decrease in maize yields was projected in the 2080s under both RCP scenarios. The effect of increased thermal temperature for both winter wheat and summer maize is projected to reduce the length of growth cycles in both vegetative and reproductive growing periods. The projected changes suggest that the enhanced efforts for adaptation and mitigation strategies such as adjustments in planting dates, cultivar improvements, enhanced soil fertility, better irrigation, and management practices are needed to minimize the climate change impacts on reference crops. This study only covered the use of single GCM, and future work could be extended to the use of multi-models or multi-model ensemble to generate future daily weather data. In addition, we only used a single crop model. The use of multiple crop models and their response to ensembled climate data can also reveal interesting findings.

This research was supported by the Joint Foundation between the National Science Foundation of China (NSFC) and the Consultative Group for International Agricultural Research (CGIAR) (No. C31661143011) and the Agricultural Science and Technology Innovation Program of CAAS ‘Research on national food security strategy of China in the new era’ (CAAS-ZDRW202012).

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

The authors declare there is no conflict.