Multimodel ensembles are powerful tools for evaluating agricultural production. Multimodel simulation results provided by the Global Gridded Crop Model Intercomparison (GGCMI) facilitate the evaluation of the grain production situation in China. With census crop yield data, the performance of nine global gridded crop models (GGCMs) in China was evaluated, and the yield gaps of four crops (maize, rice, soybean, and wheat) were estimated. The results showed that GGCMs better simulated maize yields than those of other crops in the northeast, north, northwest, east, and center. GEPIC (CLM-CROP) performed best in simulating maize (wheat) yield in the north, east, and northwest (southwest and south), due to reasonable parameter (cultivar and phenology parameters) settings. Because the rice phenology parameters were calibrated against phenological observation networks and a simple nitrogen limitation index was introduced, ORCHIDEE-CROP performed well in rice yield simulation and soybean yield simulation (center and southwest). Among four crops, wheat has the largest yield gap (7.3–14.1%), in which the poor soil of northwest (14.1%) exposes wheat to relatively high nutritional stress. Thus, in northwest China, optimizing nitrogen management in wheat production can effectively mitigate the negative impact of climate change on crop production.

  • GGCMs performed better in simulating maize yield in most parts of China than those of other crops.

  • Cultivar and phenology parameters settings affected GGCMs' performance in simulating crop yield in China.

  • The yield gap of wheat was the highest under no nitrogen stress compared with other crops in China, especially in the northwest.

  • The northwest of China showed great potential for increasing crops yield through optimizing nitrogen management.

In 2050, the global human population is projected to reach over 9 billion – an addition of two billion people to current estimates (Godfray et al. 2010). Food demand has increased with the growing and increasingly wealthy population (Bajzelj et al. 2014). It has been projected that at least 50% more agricultural production would be needed to meet the demand for food by 2050. Therefore, increasing crop production would be one of the main tasks facing mankind in the coming decades. In general, there are two broad options for increasing crop production: (1) expanding the area of cropland at the expense of other ecosystems or (2) increasing the yields (per unit area) of our existing croplands. However, as much of the remaining cultivatable land is located under tropical rainforests with high social, economic, and ecological value, improving the yield on existing agricultural lands is a high priority (Licker et al. 2010).

China, the world's most populous country of 1.3 billion with a rapidly growing economy, faces a major challenge in achieving food security (Fan et al. 2012). In China, rapid urbanization has caused extensive grassland or farmland to be converted into construction land (Wang 2017). To ensure that there is enough farmland to produce food, the government of China announced a conservation reserve policy called the ‘Redline of Arable Line’ (RAL) in 2006. It aimed to reserve approximately 120 million ha (1,800 million mu) of arable land for food production (Wen 2011). Consequently, holding the RAL and increasing the yield of the existing arable land is a primary task. At present, the yields of existing croplands are far from reaching their full potential in China (Tao et al. 2015; Sun et al. 2018; Pu et al. 2019). Thus, to increase crop yields in the future, we must first quantify the ‘yield gap’.

The yield gap is usually defined as the difference between the potential yield and the actual field yield. To date, researchers have used various approaches, including field experiments, farmer surveys, and crop model simulations to quantify the increases in crop yields achieved by improving agricultural management practices on different spatial scales (Licker et al. 2010; Johnston et al. 2011; Van Ittersum et al. 2013). For example, the data of on-farm trials showed that the yield of irrigated maize in northeast China and North China could be increased by 4.8–8.1 and 3.8–6.1 t·ha−1 by optimizing nitrogen fertilizer and soil quality management, respectively (Qiao et al. 2021). Data of on-farm rice experiments collected between 2000 and 2013 suggested that the yield gap of rice was 0.6 t·ha−1 in the main rice production areas of China (Xu et al. 2016). Compared with field experiments and survey-based methods of quantifying yield gaps, the application of crop model simulations can certainly avoid all biotic and abiotic stresses (Van Ittersum et al. 2013). Using 80% of the potential yield as an exploitable level, the yield gap of winter wheat in the north China plain (NCP) simulated by the EPIC (Environment Policy Integrated Climate) model was 1.0 t·ha−1 (Lu & Fan 2013). The spatiotemporal distribution of the yield gap of winter wheat in NCP was simulated by the APSIM-Wheat model, and the results showed that the yield gap decreased year by year and the yield of 32.3% wheat areas stagnated (Li et al. 2014). Pu et al. (2019) used the Global Agro-ecological Zones (GAEZ) model to simulate the spatial and temporal changes of the maize yield gap in northeast China, showing that the actual yield of 17 out of 40 cities was close to the potential yield. Although relatively robust estimates of the yield gap can be obtained by crop model simulations in a cropping system or region under the dominant weather and soil conditions. There is great uncertainty regarding the simulation results of a single model, reflecting the internal assumptions of the model and crop parameter calibration (Tebaldi & Knutti 2007). In this study, we used a methodology of the multimodel ensemble to quantify the yield gap, which could reduce the uncertainty of a single model and maintain the advantages of model simulation at temporal and spatial scales (Asseng 2013).

The Global Gridded Crop Model Intercomparison Project (GGCMI) is a comparative study of global gridded crop models (GGCMs) under the framework of ISIMIP (Elliott et al. 2015), which provides crop yield simulation results as a support for quantifying yield gaps in this study. GGCMs have been widely utilized to detect the responses of crops to climate change and understand risks and opportunities with respect to food production and food security (Rosenzweig et al. 2014). However, GGCMs have a huge diversity of structures and internal assumptions, leading to uncertainty in the results. Although previous studies have compared yield simulation results with reference data provided by the FAO at different spatial scales (Müller et al. 2017), uncertainties in reference data used at the subnational scale will undoubtedly affect the accuracy of subnational model assessments. Therefore, the purposes of this study were: (1) evaluating GGCMs’ performance in simulating maize, rice, soybean, and wheat yields in different regions of China by using an observation-based yield dataset, (2) analyzing the differences between models’ performances, and (3) estimating four major crop yield gaps in different regions of China by using the multimodel ensemble mean.

Study area

According to the crop growing seasons in China's agricultural phenological map, China is divided into seven regions (Figure 1) (Chen et al. 2015). Maize is planted in most parts of China, and rain-fed maize is the main crop in the northeast, southwest, and south. Wheat is mainly planted in the north and northwest, and irrigated wheat is the main wheat. Soybean in China is mainly rain-fed and planted in the northeast. Rice is mainly planted in the east, center, southwest, and south and is mainly irrigated.

Figure 1

Distribution of four major crops (maize, rice, soybean, and wheat) in planting areas and the irrigated area proportion in China (Chen et al. 2015). The percentages and the length of individual bars in the map indicate the irrigated areas.

Figure 1

Distribution of four major crops (maize, rice, soybean, and wheat) in planting areas and the irrigated area proportion in China (Chen et al. 2015). The percentages and the length of individual bars in the map indicate the irrigated areas.

Close modal

Models participating and experimental setup

Fourteen model groups have contributed to the GGCMI, following the protocol for the GGCMI (Elliott et al. 2015). For this, GGCMI defined three distinct types of model configurations in Phase 1. First, each group is asked to develop their own ‘default’ configuration based on the management and technology assumptions and inputs they typically use for simulations in the historical period. Second, each group must also prepare a ‘harmonized’ configuration (‘Fullharm’) using input data, parameters, and definitions provided by the GGCMI coordinators. Finally, each model that considers nitrogen is also to be run in a configuration without nitrogen stress, ‘Harmnon’, to allow for a direct comparison with models that do not explicitly consider the nitrogen cycle (Elliott et al. 2015). In this study, we selected the simulation results of GGCMs from the ‘Fullharm’ scenario to compare the performance of GGCMs with the same climate and farmland management datasets as inputs.

As some groups could not supply simulation results for all crops or model configurations (see Table 1), we only used simulation results from nine models (CLM-CROP, EPIC-BOKU, EPIC-IIASA, EPIC-TAMU, GEPIC, ORCHIDEE-CROP, PAPSIM, PDSSAT, and PEGASUS; Table 2). These GGCMs were driven by the historical weather datasets WFDEI, AgMERRA, WATCH (WFD), GRASP, AgCFSR, and Princeton GF (Elliott et al. 2015). As these weather datasets are based on station data and/or reanalysis data, we assumed that different weather datasets have little impact on the simulation results. This study did not take into account the contribution of uncertainties in historic weather datasets to crop model skills. Therefore, we only used simulation results of the models driven by the weather dataset WFDEI. The WFDEI dataset spans the time from 1979 to 2010 and provides daily data on the most important meteorological driver variables, and groups applied their interpolation to subdaily values if needed (Müller et al. 2017).

Table 1

Available simulation results of GGCMs

GGCMMaize
Rice
Soybean
Wheat
DaFHDFHDFHDFH
CGMS-WOFOST b – – √ – – √ – – √ – – 
CLM-CROP √ √ √ √ √ √ √ √ √ √ √ √ 
EPIC-BOKU √ √ √ √ √ √ √ √ √ √ √ √ 
EPIC-IIASA – √ √ – – – – √ √ – √ √ 
EPIC-TAMU – √ √ – – – – – – – √ √ 
GEPIC √ √ √ √ √ √ √ √ √ √ √ √ 
LPJ-GUESS √ – √ √ – √ √ – √ √ – √ 
LPJML √ – √ √ – √ √ – √ √ – √ 
ORCHIDEE-CROP √ √ – √ √ √ √ √ – √ √ √ 
PAPSIM √ √ √ – – – √ √ √ √ √ √ 
PDSSAT √ √ √ √ √ √ √ √ √ √ √ √ 
PEGASUS √ √ √ – – – √ √ √ √ √ √ 
PRYSBI2 √ – – √ – – √ – – √ – – 
GGCMMaize
Rice
Soybean
Wheat
DaFHDFHDFHDFH
CGMS-WOFOST b – – √ – – √ – – √ – – 
CLM-CROP √ √ √ √ √ √ √ √ √ √ √ √ 
EPIC-BOKU √ √ √ √ √ √ √ √ √ √ √ √ 
EPIC-IIASA – √ √ – – – – √ √ – √ √ 
EPIC-TAMU – √ √ – – – – – – – √ √ 
GEPIC √ √ √ √ √ √ √ √ √ √ √ √ 
LPJ-GUESS √ – √ √ – √ √ – √ √ – √ 
LPJML √ – √ √ – √ √ – √ √ – √ 
ORCHIDEE-CROP √ √ – √ √ √ √ √ – √ √ √ 
PAPSIM √ √ √ – – – √ √ √ √ √ √ 
PDSSAT √ √ √ √ √ √ √ √ √ √ √ √ 
PEGASUS √ √ √ – – – √ √ √ √ √ √ 
PRYSBI2 √ – – √ – – √ – – √ – – 

aD, F, and H are the ‘Default, Fullharm, and Harmnon’ scenarios, respectively.

b‘√’ represents available data, and ‘–’ represents unavailable data.

Table 2

Summary of base cases of the models

GGCMModel typeWFDEIManagement scenariosIrrigation methodResolution
CLM-CROP Ecosystem 1980–2012 D/F/Ha Firr/Noirr 0.5°×0.5° 
EPIC-BOKU Site-based 1979–2010 D/F/H Firr/Noirr 0.5°×0.5° 
EPIC-IIASA Site-based 1979–2010 F/H Firr/Noirr 0.5°×0.5° 
EPIC-TAMU Site-based 1979–2009 F/H Firr/Noirr 0.5°×0.5° 
GEPIC Site-based 1979–2009 D/F/H Firr/Noirr 0.5°×0.5° 
ORCHIDEE-CROP Ecosystem 1979–2009 D/F/H Firr/Noirr 0.5°×0.5° 
PAPSIM Site-based 1979–2009 D/F/H Firr/Noirr 0.5°×0.5° 
PDSSAT Site-based 1979–2009 D/F/H Firr/Noirr 0.5°×0.5° 
PEGASUS Ecosystem 1979–2010 D/F/H Firr/Noirr 0.5°×0.5° 
GGCMModel typeWFDEIManagement scenariosIrrigation methodResolution
CLM-CROP Ecosystem 1980–2012 D/F/Ha Firr/Noirr 0.5°×0.5° 
EPIC-BOKU Site-based 1979–2010 D/F/H Firr/Noirr 0.5°×0.5° 
EPIC-IIASA Site-based 1979–2010 F/H Firr/Noirr 0.5°×0.5° 
EPIC-TAMU Site-based 1979–2009 F/H Firr/Noirr 0.5°×0.5° 
GEPIC Site-based 1979–2009 D/F/H Firr/Noirr 0.5°×0.5° 
ORCHIDEE-CROP Ecosystem 1979–2009 D/F/H Firr/Noirr 0.5°×0.5° 
PAPSIM Site-based 1979–2009 D/F/H Firr/Noirr 0.5°×0.5° 
PDSSAT Site-based 1979–2009 D/F/H Firr/Noirr 0.5°×0.5° 
PEGASUS Ecosystem 1979–2010 D/F/H Firr/Noirr 0.5°×0.5° 

aD, F, and H represent the default, Fullharm, and Harmnon scenarios, respectively.

Reference data

We used reported province-scale yield data (, t·ha−1) for the evaluation of the GGCMs.
(1)
where is the reported province-level unit yield, is the reported province-level total yield, and is the province-level crop acreage.

The data were from the ‘National Agricultural Science Data Sharing Center’ (http://crop.agridata.cn/ch_intro.asp) and were available at the province scale from 1981 to 2005. These data only represented the average crop yields at the provincial scale and consisted of the yield of rain-fed crops and irrigated crops.

Mask data

Each modeling group supplied data for each crop for all land grid cells with separate simulations for purely rain-fed conditions (‘Noirr’) and conditions with full irrigation (‘Firr’) (see Table 2). The resolution was 0.5°×0.5°. To evaluate the models’ performance in simulating crop yield, the mask data, the harvested area of irrigated and rained crops, were used to process simulation results. The mask data used in this study were from MIRCA2000 (https://www.uni-frankfurt.de/45218031/data_download), which provided separate data on crop-specific irrigated and rain-fed harvested areas per grid cell (0.5° × 0.5°) (Portmann 2010).

Data processing

Calculation of regional simulated crop yields

As reference data were available at the provincial scale, we aggregated simulation results to the provincial scale using an area-weighted average as described in the following equation:
(2)
where i is the index of any grid cell assigned to the spatial unit in question for growing season t, and n is the number of grid cells in that spatial unit. Airr,i,t is the irrigated harvested area (ha) in grid cell i, and Anoirr,i,t is the rain-fed harvested area (ha) in grid cell i. Yirr,i,t is the simulated yield (t·ha−1) under fully irrigated conditions in grid cell i, and Ynoirr,i,t is the simulated yield (t·ha−1) under rain-fed conditions in grid cell i.

As many of the trends in yield are driven by intensification and altered management (Ray et al. 2012), the trends should be removed from the simulation and reference data for facilitating the evaluation of GGCMs’ simulations. For this, the anomalies were computed by subtracting a moving average of a 5-year window (t − 2 to t + 2).

Estimation of regional crop yield gaps

In the GGCMI, the input data of the GGCMs in the ‘Fullharm’ scenario were consistent with those in the ‘Harmnon’ scenario except for nitrogen fertilizer input data. ‘Harmnon’ was defined, under the scenario of zero (or near-zero) stress from nitrogen, as ‘potential yield’ for the purpose of defining the yield gap and related analyses (Elliott et al. 2015). Therefore, we defined the difference between the simulated yield under the Harmnon scenario (Yh, t·ha−1) and the Fullharm scenario (Yf, t·ha−1) as the yield gap (Yg, t·ha−1) without nitrogen stress. The simulated yield gap of a single GGCM was calculated by using Equations (3) and (4). The results of Equation (3) are the absolute value of the yield gap (Yg,i, t·ha−1). Yg,p,i in Equation (4) is the relative value of the yield gap (%). Then, we used the multimodel ensemble mean to estimate the crop yield gap, as shown in Equation (5).
(3)
(4)
(5)
where Yg,p,e is the crop yield gap of the multimodel ensemble mean, n is the number of GGCMs, and i is a member of the GGCMs.

Metrics index

In this study, the correlation coefficient (R), normalized root-mean-square error (RMSE), percent bias (PBIAS), and mean bias (Equations (6)–(9)) were used to evaluate the GGCM performance in simulating the spatiotemporal variations in crop yields (Müller et al. 2017; Al Samouly et al. 2018):
(6)
(7)
(8)
(9)
where yieldsim,i is the simulated yield of the GGCMs, yieldref,i is the referenced yield, n is the number of growing seasons in the sample, and i is a certain growing season in the sample. The yield used to calculate the metric indexes was the result of detrending. The standard deviation (SD) was calculated for different metric indexes (P) by using the following equation:
(10)

It is difficult to attribute the sequences of growing periods to the calendar year in simulated data where model groups also interpreted the reported standards differently, which leads to the relatively vague matching of simulated and reference time series. Thus, we tested whether the time-series correlation could be substantially improved by shifting the time series by 1 year (Müller et al. 2017). This shift was applied only if the correlation coefficient was improved by at least 0.3.

Evaluation of crop yields

The correlation coefficient (R), normalized RMSE and percent bias (PBIAS) between the simulation yield of nine models and reference yield of the four crops (1981–2010) were calculated in seven regions (Figure 2; Supplementary Figure S1, 2). The R values showed that the yield variability can be reproduced well for maize in the northeast (Figure 2(b), R = 0.3–0.4) and northwest (Figure 2(a), R = 0.45–0.55) regions. The models presented relatively high R values for soybean (Figure 2(d), R = 0.25–0.6) in the north and rice (Figure 2(a), R = 0.25–0.5) in the center, while the R (<0.4) values have little difference among the four crops in the east, southwest, and south. The RMSE results showed that the models have relatively low RMSEs for soybean and wheat in the northeast (Supplementary Figure S1b, 0.47–0.56 and 0.51–0.54) and northwest (Supplementary Fig. S1a, 0.39–0.54 and 0.48–0.58), while in the north (0.51–0.64), east (0.5–0.62), and south (0.4–0.6) small differences were found in RMSE values in the four crops. In addition, the models have low RMSEs for maize (Supplementary Figure S1c, 0.24–0.47) and wheat (Figure S1f, 0.41–0.68) in the center and southwest, respectively. The results showed a relatively small PBIAS for maize in all the regions except the south. Additionally, the PBIAS values were relatively low for rice in the northeast (−36 to −9%), north (−55 to 21%), east (−61 to 2%), center (−65 to 5%), and south (−3 to 15%) and for wheat in the north (−0.6 to 53%), east (−6 to 58%), and southwest (5 to 63%). Generally, in most parts of northern China (i.e., northeast, north, and northwest) and certain regions of southern China (i.e., east and center), the models had better performance in simulating maize yield than that of the other crops.

Figure 2

Correlation coefficient (R) between the yield simulated by GGCMs and the reported yield (1981–2010) for four crops in the different regions of China (a–g). The boxes represent the 25th − 75th percentiles, whiskers (the horizontal lines linked to boxes by the vertical line) represent the 1.5 interquartile range (IQR), and thick dashed inboxes show the median R values for all models. Symbols in different colors denote the nine models; the five-pointed star denotes the multimodel ensemble mean.

Figure 2

Correlation coefficient (R) between the yield simulated by GGCMs and the reported yield (1981–2010) for four crops in the different regions of China (a–g). The boxes represent the 25th − 75th percentiles, whiskers (the horizontal lines linked to boxes by the vertical line) represent the 1.5 interquartile range (IQR), and thick dashed inboxes show the median R values for all models. Symbols in different colors denote the nine models; the five-pointed star denotes the multimodel ensemble mean.

Close modal

Different models showed variably among the four crops. The R (Figure 2), GEPIC (R = 0.45), EPIC-IIASA (R = 0.41), and EPIC-TAMU (R = 0.4) simulated a higher R for maize than that of other crops in the northeast, and EPIC-TAMU and EPIC-BOKU also performed well for maize in the northwest (R = 0.51 and 0.60), east (R = 0.37 and 0.49), and southwest (R = 0.42 and 0.51). In addition, in the northwest, EPIC-IIASA (R = 0.45), PDSSAT (R = 0.57), ORCHIDEE-CROP (R = 0.47), PEGASUS (R = 0.45), and the ensemble (R = 0.61) simulated maize yield with the highest R among the four crops. For the four crops, GEPIC (R = 0.52) and EPIC-BOKU (R = 0.57) simulated the rice yield better than that of other crops in the center, whereas EPIC-TAMU (R = 0.56), PDSSAT (R = 0.45), and PAPSIM (R = 0.39) showed a better performance in simulating wheat yield than other models. Moreover, GEPIC (R = 0.62), EPIC-BOKU (R = 0.4), ORCHIDEE-CROP (R = 0.49), and PEGASUS (R = 0.37) also showed a superior simulation for wheat than for other crops in the south, and GEPIC (R = 0.63), EPIC-IIASA (R = 0.59), EPIC-BOKU (R = 0.59), PDSSAT (R = 0.4), and the ensemble (R = 0.5) performed well for soybean in the north.

For the normalized RMSE (Supplementary Figure S1), in the center, many models presented low RMSEs for maize, such as PDSSAT (RMSE = 0.38), PAPSIM (RMSE = 0.24), GEPIC (RMSE = 0.19), EPIC-TAMU (RMSE = 0.24), EPIC-IIASA (RMSE = 0.41), PEGASUS (RMSE = 0.43), and ensemble (RMSE = 0.3). Additionally, PEGASUS showed a relatively low RMSE for maize in the north (RMSE = 0.35), northwest (RMSE = 0.4), and south (RMSE = 0.36). ORCHIDEE-CROP and EPIC-BOKU performed well for wheat simulations with low RMSEs in the north (0.29 and 0.5), northwest (0.29 and 0.5), and south (0.36 and 0.47), while CLM-CROP presented low RMSEs for rice in the northeast (0.47), north (0.39), and northwest (0.26).

For the percent bias (PBIAS) (Supplementary Figure S2), PDSSAT had a lower PBIAS for rice than other crops in most parts of China, such as the northwest (−11%), east (5%), center (−3%), southwest (2%), and south (15%), while GEPIC showed a lower PBIAS for maize than other crops in the northeast (−16%), north (−1%), northwest (6%), east (−5%), and southwest (−1%). EPIC-TAMU, EPIC-IIASA, and the ensemble also had low PBIAS values for maize simulations in the northeast region. In addition, the PBIAS of wheat was lower than that of other crops for the simulations of EPIC-TAMU in the north (−9%), northwest (−10%), east (−6%), and southwest (−4%) and of CLM-CROP in the north (−31%), northwest (26%), east (−23%), center (−14%), southwest (5%), and south (−10%). For the four crops, ORCHIDEE-CROP (PEGASUS) showed a smaller PBIAS for soybean than that of other crops in the northwest (4%) and south (−12%), northeast (9%), and east (−11%)) and for rice in the northeast (−9%).

Generally, GEPIC, EPIC-IIASA, and EPIC-TAMU provided a better simulation for maize than for other crops in the northeast. Compared with other crops in the northwest, PEGASUS had better performance for maize. In addition, only one model (i.e., EPIC-BOKU) presented a better performance for wheat than for the other crops in the south. However, for rice and soybean, none of the models performed well in all three (i.e., R, RMSE, and PBIAS) or any two indexes (i.e., R and RMSE; R and PBIAS; and RMSE and PBIAS).

Evaluation of crop yields in different regions

For maize, GGCMs performed well in simulating yield temporal variability in the northwest, in which EPIC-BOKU and the ensemble showed a large R (∼0.60 and ∼0.61) (Supplementary Figure S3a,b). EPIC-BOKU also performed well in the north (R = 0.50) and southwest (R = 0.50), but this model tended to overestimate the maize yield in these two regions (3.0 and 2.7 t·ha−1, Figure 3(a) and 3(b)) due to the use of high-yielding cultivars in EPIC-BOKU (Folberth et al. 2016). Because EPIC-based models consider the influence of water stress on the harvest index, the interannual variation in maize yield is greatly influenced by the degree of water stress (Müller et al. 2019), which resulted in EPIC-based model simulations reproducing the temporal variability of maize yields well. For example, GEPIC had relatively high R values (0.40, 0.55, and 0.69) and relatively low mean bias (−0.05, 0.26, and −0.23 t·ha−1) in the north, east, and northwest. In the southwest, GEPIC showed a relatively small mean bias (−0.01 t·ha−1) and low R (0.14). Figure 3(b) reveals that most models show a low mean bias in the northeast (<0.7 t·ha−1; i.e., PAPSIM, PDSSAT, EPIC-TAMU, EPIC-IIASA, EPIC-BOKU, and CLM-CROP). Compared with the single model in the center and northeast, ensemble also has the lowest mean bias (−0.3 and 0.14 t·ha−1). Therefore, the performance of the multimodel ensemble was better than that of the single model in northeast and central China for the long-term average level of maize yield simulations.

Figure 3

Mean bias values between the simulated yields by GGCMs and the reported yields for four crops ((a,b): maize; (c,d): rice; (e,f): soybean; (g,h): wheat) in seven regions of China. Heatmap of simulated maize yields by GGCMs compared to the reported yields (1981–2010) in seven regions of China. The boxes represent the 25th − 75th percentiles, whiskers (the horizontal lines linked to boxes by the vertical line) represent the 1.5 interquartile range (IQR), and thick dashes are the median mean bias values for all models.

Figure 3

Mean bias values between the simulated yields by GGCMs and the reported yields for four crops ((a,b): maize; (c,d): rice; (e,f): soybean; (g,h): wheat) in seven regions of China. Heatmap of simulated maize yields by GGCMs compared to the reported yields (1981–2010) in seven regions of China. The boxes represent the 25th − 75th percentiles, whiskers (the horizontal lines linked to boxes by the vertical line) represent the 1.5 interquartile range (IQR), and thick dashes are the median mean bias values for all models.

Close modal

For rice, the GGCMs showed good performance in simulating yield variability in the southwest, among which CLM-CROP had the largest R (∼0.45) (Supplementary Figure S3c,d). However, CLM-CROP (CLM4.5) introduced the plant functional type of tropical rice rather than temperate rice (Badger & Dirmeyer 2015), which caused the model to underestimate the rice yield in most regions of China (mean bias of >4.0 t·ha−1) (Figure 3(c)). GEPIC (EPIC-BOKU) had a relatively high R (0.39 and 0.31) in the northwest and south (center ∼0.57 and east ∼0.25) and underestimated rice yield in both regions with a mean bias of 1.8 t·ha−1 (center ∼3.6 t·ha−1 and east ∼3.9 t·ha−1). ORCHIDEE-CROP performed better than the other models in simulating the rice yield in the northeast, north, northwest, and east, with a mean bias ranging from 0.19 to 1.28 t·ha−1. The rice phenology parameters of ORCHIDEE-CROP were calibrated against phenological observation networks in China (Wang et al. 2017; Müller et al. 2019), which allowed this model to perform well in simulating rice yield. In addition, the irrigation effects on crop yields were not accounted for in ORCHIDEE-CROP (Wu et al. 2016), so the yield simulated by this model can reflect the impact of climate change on the annual fluctuations in crop yields. For example, in the northeast, where rain-fed rice accounts for a large area (20%), ORCHIDEE-CROP showed a higher R (0.44) than other regions. PDSSAT showed a relatively low mean bias (<0.5 t·ha−1) and R value for the central, southwestern, and southwestern regions. Figure 3(c) (Supplementary Figure S3c) illustrates that the mean bias (R) of ensemble (five models) was less than 0 t·ha−1 (0.3) for all regions, which indicates that GGCMs have poor performance in simulating rice yields in China.

For soybean, GGCMs showed good performance in simulating the yield variability in the north and northwest, among which GEPIC (0.63 and 0.53), EPIC-IIASA (0.59 and 0.49), and EPIC-BOKU (0.59 and 0.44) had relatively large R values (Supplementary Fig. S3e,f). However, GEPIC, EPIC-IIASA, and EPIC-BOKU tended to overestimate the soybean yields in all regions (Figure 3(e)). EPIC-based models take water stress (the most severe stress) as a factor to correct potential biomass, which ignores the influence of other stress factors (e.g., nutrients, salinity, and aeration) on biomass (Folberth et al. 2016), resulting in high simulated yields of soybean. PDSSAT (PAPSIM) had a high R in the east (0.37) and south (0.45) (northeast ∼0.46 and center ∼0.30), while it overestimated soybean yield with mean biases of 1.86 and 2.19 t·ha−1 (0.90 and 2.21 t·ha−1). ORCHIDEE-CROP performed well in the southwest with a relatively low mean bias (0.46 t·ha−1) and high R (0.45). Additionally, ORCHIDEE-CROP showed a low mean bias (0.05–0.36 t·ha−1) in the northeast, northwest, central, and southern regions. Due to the lack of an explicit parameterization of nitrogen processes in ORCHIDEE-CROP, this model is currently unable to account for dynamic nitrogen stress within the crop growing season (Wu et al. 2016). For this, ORCHIDEE-CROP introduces a very simple parameter for the effects of nitrogen fertilization on plant productivity, which probably leads to good agreement between the observed and modeled crop yields. In the north and east, PEAGASUS simulated soybean yield well with a relatively low mean bias (0.56 and 0.18 t·ha−1). PEGASUS assumes that there is no nitrogen stress when simulating soybean yield (Deryng et al. 2011), which resulted in the performance of PEGASUS in simulating soybean yield in the Fullharm scenario mainly depending on the input soybean calendar dataset (planting and harvest dates). Figure 3(f) reveals that most models and the ensemble have low mean biases in the northeast (<0.5 t·ha−1; i.e., PDSSAT, ORCHIDEE-CROP, EPIC-IIASA, EPIC-BOKU, and PEGASUS). However, the ensemble overestimated the soybean yield in the northeast with a mean bias of >0.5 t·ha−1, which indicates that the ensemble performed poorly in simulating the long-term mean soybean yields in different regions of China.

For wheat, the R results showed that GGCMs performed well in the northwest, with EPIC-IIASA having a relatively large R (0.47) (Supplementary Fig. S3 g,h). GEPIC simulated wheat yield variability well in the east (R = 0.52) and south (R = 0.62). However, EPIC-IIASA and GEPIC introduce a high denitrification soil water threshold to reduce soil nitrogen loss, thus reducing nitrogen stress and leading to overestimation of wheat yield (Figure 3(g)). Compared with EPIC-IIASA, GEPIC, and EPIC-BOKU, wheat showed higher nitrogen stress in EPIC-TAMU (Folberth et al. 2016), which resulted in lower wheat yields simulated by EPIC-TAMU. The results in Figure 3(g) show that the wheat yield simulated by EPIC-TAMU was close to the reported dataset in the north (−0.38 t·ha−1), northwest (−0.28 t·ha−1), east (−0.22 t·ha−1), and southwest (−0.10 t·ha−1). This indicates that EPIC-TAMU is suitable for simulating the long-term average wheat yield in most parts of China. CLM-CROP had low (0.14 t·ha−1) and high R (0.34) values in the southwest but showed a low mean bias and R in the center (−0.38 t·ha−1) and south (−0.21 t·ha−1). Only in the northeast did the ensemble perform better than other single models in simulating wheat yield variability.

Figure 4

Yield gaps of major food crops in different regions obtained with the GGCMs, including the mean and standard deviation. (a–d) represent maize, rice, soybean, and wheat, respectively.

Figure 4

Yield gaps of major food crops in different regions obtained with the GGCMs, including the mean and standard deviation. (a–d) represent maize, rice, soybean, and wheat, respectively.

Close modal

Evaluation of yield gaps in different regions

The yield gap of the four crops was the average value of the nine models. All the models had the lowest yield gap for soybean among all crops (Figure 5). Generally, the biological nitrogen fixation ability of soybean can reduce nitrogen stress in this crop, resulting in the low potential for soybean yield increases without N stress. Most models show a higher yield gap for wheat than for maize in China (PDSSAT, PAPSIM, ORCHIDEE-CROP, EPIC-TAMU, and EPIC-BOKU), which indicates that the yield potential of wheat in China is higher than that of maize under optimized nitrogen management. Notably, PEGASUS tended to simulate a higher yield gap for maize than for wheat (Figure 4). After using the fertilization application rates provided by the GGCMI in PEGASUS, the values of the nitrogen stress factor (fN) in maize and wheat were similar (mean value of ∼0.7) (Deryng et al. 2011). However, the wheat yield simulated by PEGASUS was much higher than the simulated maize yield in the Fullharm scenario (Figure 6), which led to the yield gap of wheat being significantly higher than that of maize after no nitrogen stress (fN = 1). For rice, the yield gap simulated by most models was less than that of maize and wheat (PDSSAT, PAPSIM, EPIC-TAMU, EPIC-IIASA, EPIC-BOKU, CLM-CROP, and PEGASUS), revealing that there is low yield potential for rice in China under optimized nitrogen management.

Figure 5

Yield gaps of four crops (maize, rice, soybean, and wheat) simulated by different models. The numbers in parentheses in the figure represent the values of the highest column.

Figure 5

Yield gaps of four crops (maize, rice, soybean, and wheat) simulated by different models. The numbers in parentheses in the figure represent the values of the highest column.

Close modal
Figure 6

Mean biases of simulated maize and wheat yields by PDSSAT and PEGASUS under the ‘default’ and ‘Fullharm’ scenarios in different regions of China from 1981 to 2005. Black is the ‘default’ scenario, and red is the ‘Fullharm’ scenario. Please refer to the online version of this paper to see this figure in color: http://dx.doi.org/10.2166/nh.2022.087.

Figure 6

Mean biases of simulated maize and wheat yields by PDSSAT and PEGASUS under the ‘default’ and ‘Fullharm’ scenarios in different regions of China from 1981 to 2005. Black is the ‘default’ scenario, and red is the ‘Fullharm’ scenario. Please refer to the online version of this paper to see this figure in color: http://dx.doi.org/10.2166/nh.2022.087.

Close modal

For the four crops, there were great differences among GGCMs in the simulated yield gaps. For maize, the SD values of the yield gap were relatively high in the northeast (∼2.11 t·ha−1), north (∼1.86 t·ha−1), and northwest (∼2.77 t·ha−1) regions due to the high yield gap (>4.0 t·ha−1) simulated by PEGASUS (Figure 4(a)). For PEGASUS, the algorithm of the effect of the nitrogen stress factor (fN) on crop yield easily amplifies crop biomass without nitrogen stress (Deryng et al. 2011). PEGASUS also showed a relatively high yield gap for wheat, which led to the high SD values of the yield gap in the northeast (1.28 t·ha−1), northwest (1.31 t·ha−1), east (1.11 t·ha−1), and south (1.01 t·ha−1) (Figure 4(d)). In addition, PAPSIM simulated a relatively large yield gap (>2.0 t·ha−1) for wheat in certain regions (i.e., northeast, north, northwest, east, and center), which may be because there was no fertilizer application in the late growth stage of wheat accounted for in PAPSIM in the Fullharm scenario. For PAPSIM, it is assumed that half of the fertilizer is applied at planting, and half is applied 40 days later, which will cause wheat to suffer from high nitrogen stress in the later growth stage (Müller et al. 2019). Therefore, PAPSIM showed a high yield gap for wheat under no nitrogen stress. For rice and soybean, low SD values of yield gaps were found in the north (0.04 and 0.015 t·ha−1), east (0.04 and 0.013 t·ha−1), and center (0.07 and 0.008 t·ha−1), while the simulated yield gaps among the models were quite different in other regions. For example, GEPIC showed a relatively high yield gap in the southwest and south due to considering the effect of water erosion on the soil nitrogen content (Folberth et al. 2016). There is heavy rainfall in the southwest and south, so a large amount of soil nitrogen was lost in the process of GEPIC simulation, increasing the nitrogen stress of crops. Notably, the yield gap of rice simulated by EPIC-BOKU also showed high SD values, but the values of the yield gap it showed were not reasonable (<0 t·ha−1). A possible explanation for this is that EPIC-BOKU's nitrogen application rate in the ‘Harmnon’ scenario was less than that in the ‘Fullharm’ scenario.

Removing the unreasonable yield gaps simulated by PEGASUS and EPIC-BOKU, the calculation of the yield gaps of four crops by the multimodel (other models) ensemble mean method is presented in Figure 7. The largest yield gap in China was observed for wheat (7.3–14.1%), followed by rice (0.1–5.9%), maize (0.3–4.5%), and soybean (0.2–0.8%). For wheat, the poor soil in the northwest causes crops to suffer from relatively high nutritional stress, which led to the largest yield gap (14.1%). The east, center, and southwest also have a higher yield gap (>10%) of wheat than other regions under no nitrogen stress. In the east, center, and southwest, the area of rain-fed wheat accounts for a large proportion of the total area of wheat in each region. The water stress degree of rain-fed wheat is greater than that of irrigated wheat, which results in a higher nitrogen stress degree for rain-fed wheat than irrigated wheat. For maize, northwest China had the largest yield gap (4.5%), while northeast China and southwest China, with relatively large proportions of rain-fed rice area, had relatively large rice yield gaps (5.9 and 4.7%).

Figure 7

Yield gaps of four crops in different regions calculated by the multimodel ensemble mean method after removing the unreasonable yield gaps simulated by PEGASUS and EPIC-BOKU.

Figure 7

Yield gaps of four crops in different regions calculated by the multimodel ensemble mean method after removing the unreasonable yield gaps simulated by PEGASUS and EPIC-BOKU.

Close modal

Differences in model performance

The parameter settings of the models affect their performance. GEPIC performed well in simulating maize yields in most regions (north, east, and northwest), possibly due to the introduction of low-yielding maize cultivars (Folberth et al. 2016). Liu et al. (2013) also showed that there was good agreement between the simulated maize yield by GEPIC and the reference dataset for China. Compared with GEPIC, EPIC-IIASA (EPIC-BOKU and EPIC-TAMU) introduced a high-yielding maize cultivar, resulting in the overestimation of maize yield. Similarly, EPIC-IIASA overestimated maize yield by 20% for the United States due to the setting of high-yielding variety parameters (Folberth et al. 2016). In this study, EPIC-TAMU had good performance in simulating wheat yields in China. In China, there is high-intensity irrigation and fertilization during the wheat growing season, which leads to the major loss of soil nitrogen, accounting for 15–55% of the applied amount (Zhu & Chen 2002). EPIC-TAMU introduced a low denitrification soil water threshold to enhance soil denitrification (Folberth et al. 2016); thus, EPIC-TAMU performed well for wheat simulations in regions with severe nitrogen loss. In addition, the performance of the models was also related to whether the crop phenology parameters had been corrected. We found that ORCHIDEE-CROP performed well in rice yield simulation, possibly because the rice phenology parameters of ORCHIDEE-CROP were calibrated against phenological observation networks in China (Müller et al. 2019). Research has shown that the optimized ORCHIDEE-CROP can reproduce the complex regional variations in rice yields in China (Wang et al. 2017).

The harmonization of growing seasons from the GGCMI also affected the performance of the models. GGCMs only simulate the rice yield in a single growing season in the GGCMI, while rice has multiple cropping seasons. Thus, the complexity of the multiple cropping seasons makes it difficult to match the simulated growing season with the reference growing season (Iizumi & Ramankutty 2015), which led to poor simulation for rice. In addition, the uncertainty of the crop calendar dataset used in the GGCMI resulted in a large bias in the simulated results, especially for the models that are sensitive to the planting and harvest dates. For instance, PDSSAT and PAPSIM overestimated the maize and wheat yields in most parts of China. The cultivar phenology parameters of PDSSAT and PAPSIM were calibrated by the planting and harvest dates (Müller et al. 2019). Therefore, for these two models, if the crop calendar data are not calibrated, it may result in a large deviation between the simulated results and the actual yields.

Effects of agronomic measures on the yield gaps of major food crops in China

Optimizing nitrogen management to increase yield is one of the main ways to mitigate the impact of climate change. Of the four major crops, wheat has a higher yield gap (7.3–14.1%) under no nitrogen stress in all regions, among which northwest China has the highest wheat yield gap (14.1%). Similarly, previous research also showed that the wheat yield in northwest China increased the most without water and nutrient limitations (Lu & Fan 2013; Gou et al. 2017). A possible explanation for this might be that the soil nutrients in the northwest are more deficient, and the contents of available nitrogen are lower than those in other regions. Studies have shown that high-intensity irrigation and the application of chemical fertilizer are usually carried out during the whole growth period of wheat to increase yield in the northwest (Zhu & Chen 2002). These practices reduced the nitrogen-use efficiency of wheat and led to a large nitrogen surplus. A large part of this surplus nitrogen would be lost by volatilization, leaching, and runoff. Moreover, as the main wheat-producing area in China, the yield gap of wheat in the north is 10.1% (0.7 t·ha−1). This result is similar to (1.0 t·ha−1) 18% of wheat yield gap reported by earlier studies on wheat yield gaps in NCP (Lu & Fan 2013). Thus, the results suggest that some potential exists to increase wheat production in China, particularly in the main growing areas of wheat (north and northwest). The most important one to exploit existing yield potentials is to improve the use efficiency of irrigation and fertilizers by highly effective farming practices, such as timely applications of irrigation and fertilizers based on crop requirements and soil conditions (Chlingaryan et al. 2018).

The yield gap of maize in all regions was less than 5% without nitrogen stress. This result agreed with the findings of previous research, in which the maize yield could be increased by 5% over that with current farming practices through proper N management alone (Cui et al. 2008). In this study, we found that the yield gap of maize in the northwest was the highest (4.5%) among all regions. However, Meng et al. (2013) showed that the maize yield gap in the northwest was up to 7.6 t·ha−1 (104%) without being restricted by nutrients, toxicity, pests, and weeds. Water is the main factor restricting the growth of maize in the northwest, where most farmers do not or cannot irrigate at the appropriate time (Meng et al. 2013). If irrigation is applied at this critical stage, the maize yield can be improved by 19% (Chen et al. 2008). Additionally, soil quality is also the main factor restricting the increase of maize yield. Qiao et al. (2021) showed that the soil quality component of the efficiency yield gap accounted for a considerable proportion (9.6–25.8% for Northeast and 6.0–17.7% for NCP) of the maize yield gap. The results of this study did not reflect the effect of soil quality on the yield gap. The rapid benefits of maize yield brought by the promotion of hybrid species and the application of nitrogen fertilizer have been exhausted (Gale et al. 2014). Thus, in order to reduce the maize yield gap, the effective ways are to improve irrigation efficiency and soil quality.

In this study, the range of the rice yield gap was 0.1–5.9% in China. This means that there is the potential for an increase in rice yield through nitrogen management. Zhao et al. (2012) showed that the net effect of applying nitrogen fertilizer on rice yield is small. Data of 2,218 on-farm rice experiments collected between 2000 and 2013 also showed that the rice had a relatively low yield gap (0.6 t·ha1) in the main rice production areas of China (Xu et al. 2016). Generally, the crop yield could reach 75–80% of the potential yield threshold at which crop yield typically stagnates at regional to national scales due to diminishing returns from further investment in yield-enhancing technologies and inputs (Deng et al. 2019). As in California, where the agricultural inputs are high, irrigated rice production stagnated at 76% of the potential yield (Espe et al. 2016). At present, China's average rice yield represents 69% of the potential yield, which has also approached this threshold (Deng et al. 2019). Thus, the basic yield of rice is high, which results in a lower yield-increasing potential of rice.

The nitrogen absorbed by soybean through biological nitrogen fixation accounted for 55% of the total nitrogen accumulated by aboveground biomass (Ciampitti & Fernando 2018). In the later stage of the growing season, nitrogen absorbed by biological nitrogen fixation can meet the demands of soybean, reducing nitrogen stress (Córdova et al. 2019). As found in this study, the soybean yield increased less under nitrogen management, with a yield gap of 0.2–0.8%. However, accurate identification and correction of abiotic (atmospheric and soil factors) or biotic stresses (pest problems) could increase soybean yields from 25 to 66% (Board & Kahlon 2011). In addition, breeding new varieties through genetic improvement could significantly increase soybean yields. Previous studies have indicated that an increase in soybean yield by 50% was attributed to genetic improvement (Board & Kahlon 2011). Thus, the improvement of nitrogen management has little effect on increasing soybean yield. More consideration should be given to strengthening the research and development of new varieties.

This study evaluated GGCM performance in simulating maize, rice, soybean, and wheat yields in different regions of China by using an observation-based yield dataset and estimated the four major crop yield gaps of different regions of China based on the simulation results under the ‘Fullharm’ and ‘Harmnon’ scenarios by using the multimodel ensemble mean. The conclusions are as follows:

  1. GGCMs have a better performance in simulating maize yield in most parts of northern China (i.e., northeast, north, and northwest) and certain regions of southern China (i.e., east and center) than those of the other crops. Moreover, of all four crops, the rice yield in the south was best simulated by the models. GEPIC, EPIC-IIASA, and EPIC-TAMU (PEGASUS) showed a better simulation for maize than for other crops in the northeast (northwest). In addition, only one model (i.e., EPIC-BOKU) exhibited a better performance for wheat among the four crops in the south. However, for rice and soybean, none of the models performed well based on any two or all three indexes.

  2. For maize, GEPIC performed well in the north, east, and northwest, while ORCHIDEE-CROP performed well for simulating the rice yield in the northeast. ORCHIDEE-CROP had good performance for soybean yield simulations in the center and southwest, and CLM-CROP well simulated the wheat yields in the southwest and south. The performance of the models was closely related to the setting of crop cultivar parameters and whether the crop phenology parameters had been calibrated. Moreover, the uncertainty of the crop calendar dataset used in the GGCMI was also one of the main factors affecting the simulation results of the models.

  3. The yield gaps of the four crops under no nitrogen stress in China were 7.3–14.1% for wheat, >0.1–5.9% for rice, >0.3–4.5% for maize, and 0.2%–0.8% for soybean. For wheat, the northwest had the largest yield gap (14.1%), and the east (11.3%), center (12.9%), southwest (10.1%), and north (10.5%) also had high wheat yield gaps under no nitrogen stress.

We acknowledge support and assistance with data provision from the Agricultural Intercomparison and Improvement Project (AgMIP). We acknowledge the global gridded crop modeling groups that participated in the GGCMI program. This research was funded by the National Key R&D Program of China (Grant No. 2017YFA0603702) and the National Natural Science Foundation of China (Grant Nos 41701023 and 51809105).

All relevant data are available from https://data.isimip.org/.

Al Samouly
A.
,
Luong
C. N.
,
Li
Z.
,
Smith
S.
,
Baetz
B.
&
Ghaith
M.
2018
Performance of multi-model ensembles for the simulation of temperature variability over Ontario, Canada
.
Environ. Earth Sci.
77
(
13
),
524
.
https://doi.org/10.1007/s12665-018-7701-2
.
Asseng
S.
2013
Uncertainty in simulating wheat yields under climate change
.
Nat. Clim. Change
3
,
827
832
.
https://doi.org/10.1038/nclimate1916
.
Badger
A. M.
&
Dirmeyer
P. A.
2015
Climate response to Amazon forest replacement by heterogeneous crop cover
.
Hydrol. Earth Syst. Sci.
12
(
1
),
879
910
.
https://doi.org/10.5194/hessd-12-879-2015
.
Bajzelj
B.
,
Richards
K. S.
,
Allwood
J. M.
,
Smith
P.
,
Dennis
J. S.
,
Curmi
E.
&
Gilligan
C. A.
2014
Importance of food-demand management for climate mitigation
.
Nat. Clim. Change
4
(
10
),
924
929
.
https://doi.org/10.1038/NCLIMATE2353
.
Board
J. E.
&
Kahlon
C. S.
2011
Soybean yield formation: what controls it and how it can be improved
. In:
Soybean Physiology and Biochemistry
.
InTech
.
https://doi.org/10.5772/17596
.
Chen
G. P.
,
Yang
G. H.
,
Zhao
M.
,
Wang
L. C.
,
Wang
Y. D.
,
Xue
J. Q.
,
Gao
J. L.
,
Li
D. H.
,
Dong
S. T.
,
Li
C. H.
,
Song
H. X.
&
Zhao
J. R.
2008
Studies on maize small area superhigh yield trails and cultivation technique
.
J. Maize Sci.
16
,
1
4
.
https://doi.org/10.3724/SP.J.1005.2008.01083s
.
Chen
C.
,
Pang
Y. M.
,
Pan
X. B.
&
Zhang
L. Z.
2015
Impacts of climate change on cotton yield in China from 1961 to 2010 based on provincial data
.
J. Meteorolog. Res.
29
(
3
),
515
524
.
https://doi.org/10.1007/S13351-014-4082-7
.
Chlingaryan
A.
,
Sukkarieh
S.
&
Whelan
B.
2018
Machine learning approaches for crop yield prediction and nitrogen status estimation in precision agriculture: a review
.
Comput. Electron. Agric.
151
,
61
69
.
https://doi.org/10.1016/j.compag.2018.05.012
.
Ciampitti
I. A.
&
Fernando
S.
2018
New insights into soybean biological nitrogen fixations
.
Agron. J.
110
(
4
),
1185
1196
.
https://doi.org/10.2134/agronj2017.06.0348
.
Córdova
S. C.
,
Castellano
M. J.
,
Dietzel
R.
,
Licht
M. A.
,
Togliatti
K.
,
Martinez-Feria
R.
&
Archontoulis
S. V.
2019
Soybean nitrogen fixation dynamics in Iowa, USA
.
Field Crops Res.
236
,
165
176
.
https://doi.org/10.1016/j.fcr.2019.03.018
.
Cui
Z. L.
,
Chen
X. P.
,
Miao
Y. X.
,
Zhang
F. S.
,
Sun
Q. P.
,
Schroder
J.
,
Zhang
H. L.
,
Li
J. L.
,
Shi
L. W.
,
Xu
J. F.
,
Ye
Y. L.
&
Liu
C. S.
2008
On-farm evaluation of the improved soil N–based nitrogen management for summer maize in North China Plain
.
Agron. J.
100
,
517
525
.
https://doi.org/10.2134/agronj2007.0194
.
Deng
N. Y.
,
Grassini
P.
,
Yang
H. S.
,
Huang
J. L.
,
Cassman
K. G.
&
Peng
S. B.
2019
Closing yield gaps for rice self-sufficiency in China
.
Nat. Commun.
10
(
1
),
1
9
.
https://doi.org/10.1038/s41467-019-09447-9
.
Deryng
D.
,
Sacks
W. J.
,
Barford
C. C.
&
Ramankutty
N.
2011
Simulating the effects of climate and agricultural management practices on global crop yield
.
Global Biogeochem. Cycles
25
.
https://doi.org/10.1029/2009GB003765
.
Elliott
J.
,
Müller
C.
,
Deryng
D.
,
Chryssanthacopoulos
J.
,
Boote
K. J.
,
Büchner
M.
,
Foste
I.
,
Glotter
M.
,
Heinke
J.
,
Iizumi
T.
,
Izaurralde
R. C.
,
Mueller
N. D.
,
Ray
D. K.
,
Rosenzweig
C.
,
Ruane
A. C.
&
Sheffield
J.
2015
The global gridded crop model intercomparison: data and modeling protocols for phase 1 (v1. 0)
.
Geosci. Model Dev.
8
,
261
277
.
https://doi.org/10.5194/gmd-8-261-2015
.
Espe
M. B.
,
Cassman
K. G.
,
Yang
H.
,
Guilpart
N.
,
Grassini
P.
,
Van Wart
J.
,
Anders
M.
,
Beighley
D.
,
Harrell
D.
,
Linscombe
S.
,
McKenzie
K.
,
Mutters
R.
,
Wilson
L. T.
&
Linquist
B. A.
2016
Yield gap analysis of us rice production systems shows opportunities for improvement
.
Field Crops Res.
196
,
276
283
.
https://doi.org/10.1016/j.fcr.2016.07.011
.
Fan
M. S.
,
Shen
J. B.
,
Yuan
L. X.
&
Jiang
R. F.
2012
Improving crop productivity and resource use efficiency to ensure food security and environmental quality in China
.
J. Exp. Bot.
63
(
1
),
13
24
.
https://doi.org/10.1093/jxb/err248
.
Folberth
C.
,
Elliott
J.
,
Müller
C.
,
Balkovic
J.
,
Chryssanthacopoulos
J.
,
Izaurralde
R. C.
,
Jones
C. D.
,
Khabarov
N.
,
Liu
W. F.
,
Reddy
A.
,
Schmid
E.
,
Skalský
R.
,
Yang
H.
,
Arneth
A.
,
Ciais
P.
,
Deryng
D.
,
Lawrence
P. J.
,
Olin
S.
,
Pugh
T. A. M.
,
Ruane
A. C.
&
Wang
X. H.
2016
Uncertainties in global crop model frameworks: effects of cultivar distribution, crop management and soil handling on crop yield estimates
.
Biogeosci. Discuss.
1
30
.
https://doi.org/10.5194/bg-2016-527
.
Gale
F.
,
Jewison
M.
&
Hansen
J.
2014
Prospects for China's Corn Yield Growth and Imports. United States
.
Department of Agriculture Economic Research Service
,
Washington, DC
.
Godfray
H. C. J.
,
Beddington
J. R.
,
Crute
I. R.
,
Haddad
L.
,
Lawrence
D.
,
Muir
J. F.
,
Pretty
J.
,
Robinson
S.
,
Thomas
S. M.
&
Toulmin
C.
2010
Food security: the challenge of feeding 9 billion people
.
Science
327
,
812
818
.
https://doi.org/10.1126/science.1185383
.
Gou
F.
,
Yin
W.
,
Hong
Y.
,
Van Der Werf
W.
,
Chai
Q.
,
Heerink
N.
&
Van Ittersum
M. K.
2017
On yield gaps and yield gains in intercropping: opportunities for increasing grain production in northwest China
.
Agric. Syst.
151
,
96
105
.
https://doi.org/10.1016/j.agsy.2016.11.009
.
Iizumi
T.
&
Ramankutty
N.
2015
How do weather and climate influence cropping area and intensity?
Global Food Secur
.
http://dx.doi.org/10.1016/j.gfs.2014.11.003
.
Johnston
M.
,
Licker
R.
,
Foley
J.
,
Holloway
T.
,
Mueller
N. D.
,
Barford
C.
&
Kucharik
C.
2011
Closing the gap: global potential for increasing biofuel production through agricultural intensification
.
Environ. Res. Lett.
6
(
3
),
034028
.
https://doi.org/10.1088/1748-9326/6/3/034028
.
Li
K. N.
,
Yang
X. G.
,
Liu
Z. J.
,
Zhang
T. Y.
,
Lu
S.
&
Liu
Y.
2014
Low yield gap of winter wheat in the North China Plain
.
Eur. J. Agron.
59
(
59
),
1
12
.
https://doi.org/10.1016/j.eja.2014.04.007
.
Licker
R.
,
Johnston
M.
,
Foley
J. A.
,
Barford
C.
,
Kucharik
C. J.
,
Monfreda
C.
&
Ramankutty
N.
2010
Mind the gap: how do climate and agricultural management explain the ‘yield gap’ of croplands around the world?
Global Ecol. Biogeogr.
19
,
769
782
.
https://doi.org/10.1111/j.1466-8238.2010.00563.x
.
Liu
S.
,
Yang
J. Y.
,
Zhang
X. Y.
,
Drury
C. F.
,
Reynolds
W. D.
&
Hoogenboom
G.
2013
Modelling crop yield, soil water content and soil temperature for a soybean–maize rotation under conventional and conservation tillage systems in Northeast China
.
Agric. Water Manage.
123
,
32
44
.
https://doi.org/10.1016/j.agwat.2013.03.001
.
Lu
C. H.
&
Fan
L.
2013
Winter wheat yield potentials and yield gaps in the North China Plain
.
Field Crops Res.
143
,
98
105
.
https://doi.org/10.1016/j.fcr.2012.09.015
.
Meng
Q. F.
,
Hou
P.
,
Wu
L.
,
Chen
X. P.
,
Cui
Z. L.
&
Zhang
F. S.
2013
Understanding production potentials and yield gaps in intensive maize production in China
.
Field Crops Res.
143
,
91
97
.
https://doi.org/10.1016/j.fcr.2012.09.023
.
Müller
C.
,
Elliott
J.
,
Chryssanthacopoulos
J.
,
Arneth
A.
,
Balkovic
J.
,
Ciais
P.
,
Deryng
D.
,
Folberth
C.
,
Glotter
M.
,
Hoek
S.
,
Iizumi
T.
,
Izaurralde
R. C.
,
Jones
C.
,
Khabarov
N.
,
Lawrence
P.
,
Liu
W. F.
,
Olin
S.
,
Pugh
T. A. M.
,
Ray
D. K.
,
Reddy
A.
,
Rosenzweig
C.
,
Ruane
A. C.
,
Sakurai
G.
,
Schmid
E.
,
Skalsky
R.
,
Song
C. X.
,
Wang
X. H.
,
De Wit
A.
&
Yang
H.
2017
Global gridded crop model evaluation: benchmarking, skills, deficiencies and implications
.
Geosci. Model Dev. Discuss.
1
39
.
https://doi.org/10.5194/gmd-10-1403-2017
.
Müller
C.
,
Elliott
J.
,
Kelly
D.
,
Arneth
A.
&
Yang
H.
2019
The global gridded crop model intercomparison phase 1 simulation dataset
.
Sci. Data
6
,
50
.
https://doi.org/10.1038/s41597-019-0023-8
.
Pu
L. M.
,
Zhang
S. W.
,
Yang
J. C.
,
Chang
L. P.
&
Bai
S. T.
2019
Spatio-temporal dynamics of maize potential yield and yield gaps in Northeast China from 1990 to 2015
.
Int. J. Environ. Res. Public Health
16
(
7
),
1211
.
https://doi.org/10.3390/ijerph16071211
.
Qiao
L.
,
Silva
J. V.
,
Fan
M. S.
,
Mehmood
I.
,
Fan
J. L.
,
Li
R.
&
Ittersum
M. K.
2021
Assessing the contribution of nitrogen fertilizer and soil quality to yield gaps: a study for irrigated and rainfed maize in China
.
Field Crops Res.
2021
(
273
),
108304
.
https://doi.org/10.1016/j.fcr.2021.108304
.
Ray
D. K.
,
Ramankutty
N.
,
Mueller
N. D.
,
West
P. C.
&
Foley
J. A.
2012
Recent patterns of crop yield growth and stagnation
.
Nat. Commun.
3
(
1
),
1
7
.
https://doi.org/10.1038/ncomms2296
.
Rosenzweig
C.
,
Elliott
J.
,
Deryng
D.
,
Ruane
A. C.
,
Müller
C.
,
Arneth
A.
,
Boote
K. J.
,
Folberth
C.
,
Glotter
M.
,
Khabarov
N.
,
Neumann
K.
,
Piontek
F.
,
Pugh
T. A. M.
,
Schmid
E.
,
Stehfest
E.
,
Yang
H.
&
Jones
J. W.
2014
Assessing agricultural risks of climate change in the 21st century in a global gridded crop model intercomparison
.
Proc. Natl. Acad. Sci. U S A
111
,
3268
3273
.
https://doi.org/10.1073/pnas.1222463110
.
Sun
S.
,
Yang
X. G.
,
Lin
X. M.
,
Sassenrath
G. F.
&
Li
K. H.
2018
Winter wheat yield gaps and patterns in China
.
Agron. J.
110
(
1
),
319
330
.
https://doi.org/10.2134/agronj2017.07.0417
.
Tao
F. L.
,
Zhang
S.
,
Zhang
Z.
&
Rötterc
R. P.
2015
Temporal and spatial changes of maize yield potentials and yield gaps in the past three decades in China
.
Agric. Ecosyst. Environ.
208
(
208
),
12
20
.
https://doi.org/10.1016/j.agee.2015.04.020
.
Tebaldi
C.
&
Knutti
R.
2007
The use of the multi-model ensemble in probabilistic climate projections
.
Philos. Trans. A Math. Phys. Eng.
365
(
1857
),
2053
2075
.
https://doi.org/10.1098/rsta.2007.2076
.
Van Ittersum
M. K.
,
Cassman
K. G.
,
Grassini
P.
,
Wolf
J.
,
Tittonell
P.
&
Hochman
Z.
2013
Yield gap analysis with local to global relevance – a review
.
Field Crops Res.
143
(
1
),
4
17
.
https://doi.org/10.1016/j.fcr.2012.09.009
.
Wang
X.
2017
Impacts of Climate Change and Agricultural Managements on Major Global Cereal Crops
.
Université Pierre et Marie Curie-Paris VI
.
Wang
X. H.
,
Ciais
P.
,
Li
L.
,
Ruget
F.
,
Vuichard
N.
,
Viovy
N.
,
Zhou
F.
,
Chang
J. F.
,
Wu
X. C.
,
Zhao
H. F.
&
Piao
S. L.
2017
Management outweighs climate change on affecting length of rice growing period for early rice and single rice in China during 1991–2012
.
Agric. Forest Meteorol.
233
,
1
11
.
https://doi.org/10.1016/j.agrformet.2016.10.016
.
Wen
J. B.
2011
Report on the work of the government
. In
Proceedings of the Delivered at the Fourth Session of the Eleventh National People's Congress
,
March 5
,
Beijing
.
Wu
X.
,
Vuichard
N.
,
Ciais
P.
,
Viovy
N.
,
de Noblet-Ducoudré
N.
,
Wang
X.
,
Magliulo
V.
,
Wattenbach
M.
,
Vitale
L.
,
Di Tommasi
P.
,
Moors
E. J.
,
Jans
W.
,
Elbers
J.
,
Ceschia
E.
,
Tallec
T.
,
Bernhofer
C.
,
Grünwald
T.
,
Moureaux
C.
,
Manise
T.
,
Ligne
A.
,
Cellier
P.
,
Loubet
B.
,
Larmanou
E.
&
Ripoche
D.
2016
ORCHIDEE-CROP (v0), a new process based agro-land surface model: model description and evaluation over Europe
.
Geosci. Model Dev. Discuss.
8
,
4653
4696
.
https://doi.org/10.5194/gmdd-8-4653-2015
.
Xu
X. P.
,
He
P.
,
Zhao
S. C.
,
Qiu
S. J.
,
Johnston
A. M.
&
Zhou
W.
2016
Quantification of yield gap and nutrient use efficiency of irrigated rice in China
.
Field Crops Res.
2016
(
186
),
58
65
.
https://doi.org/10.1016/j.fcr.2015.11.011
.
Zhao
X.
,
Zhou
Y.
,
Wang
S. Q.
,
Xing
G. X.
,
Shi
W. M.
,
Xu
R. K.
&
Zhu
Z. L.
2012
Nitrogen balance in a highly fertilized rice–wheat double-cropping system in Southern China
.
Soil Sci. Soc. Am. J.
76
(
3
),
1068
.
http://dx.doi.org/10.2136/sssaj2011.0236
.
Zhu
Z. L.
&
Chen
D. L.
2002
Nitrogen fertilizer use in China – contributions to food production, impacts on the environment and best management strategies
.
Nutr. Cycling Agroecosyst.
63
(
2–3
),
117
127
.
https://doi.org/10.1023/A:1021107026067
.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY-NC-ND 4.0), which permits copying and redistribution for non-commercial purposes with no derivatives, provided the original work is properly cited (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Supplementary data