Variability in Crop Response to Spatiotemporal Variation in Climate in China, 1980–2014

Abstract

As the population increases and climate extremes become more frequent, the pressure on food supply increases. A better understanding of the influence of climate variations on crop yield in China would be of great benefit to global food security. In this study, gridded, daily meteorological data and county-level annual yield data were used to quantify the climate sensitivity of corn, rice, and spring wheat yields, and identify the spatiotemporal variation relationship between climate and yields from 1980 to 2014. The results showed that rice and corn were more sensitive to climate variations than spring wheat, both spatially and temporally. Photosynthetic active radiation (PAR) was found to be beneficial to rice in northeast China and the Yangtze River basin, as well as corn in the south and spring wheat in Xinjiang, but not to rice in the south of the Yangtze River and spring wheat in the southeast coast. The temperature centroid shift was the main driving factor causing the movement of the centroid of the three crops. For every 1 km shift of the temperature centroid, the corn and rice yield centroids moved 0.97 km and 0.34 km, respectively. These findings improve our understanding of the impacts of climate variations on agricultural yields in different regions of China.

1. Introduction

Agriculture is the industry most affected by climate change [1,2]. Understanding the impact of the climate on crop yields is fundamental to agriculture’s adaptation to climate change and ensuring food security [3,4]. In general, the critical climate factors for crop yields include temperature [5,6], precipitation [7], photosynthetic radiation [8], and drought conditions [9], or climate co-variability [10,11,12]. Process-based crop simulation models [13,14,15,16] and statistical models [17,18], including machine learning methods [19], are the most common methods used in this type of research.

A number of studies have shown a significant relationship between climate change and yields at both global [20,21,22] and regional scales [23,24,25,26]. However, many knowledge gaps between the effects of interannual climate fluctuations and yields need to be explored [27]. In particular, the increase in extreme weather events in recent years [28,29,30] has directly affected climate variability [31], and has led to substantial grain yield reductions. For example, changes in rainfall patterns, increased interactions between temperature and rainfall, reduced water resources, and increased frequency and intensity of extreme events have made it more difficult to ensure agricultural resilience in a changing climate [32].

The relationship between climate and yield varies from region to region, especially in a vast territory, such as China, which feeds 22 percent of the world’s population on only 7 percent of the world’s arable land [33]. Variations in land and climate conditions, unbalanced economic development, different crop species, irrigation, and water management all affect farmland systems and productivity patterns. To gain a better understanding of the spatial heterogeneity of food production capacity in response to climate change, it is crucial to conduct a finer-scale analysis of the spatial pattern of changes in crop yield variability across the whole country.

In addition, a deeper understanding of the spatial variability of these changes will enable us to study the spatial evolution of climate–yield relations and improve the ability to predict yield distribution patterns [34]. Although there has been some exploration of the climate–yield relationship using temporal and spatial models [35,36,37,38], methods that can be used to analyze the spatiotemporal variation relationship between climate and yields are still scarce.

In this study, we focus on three scientific questions: (1) How have crop yields and climate variability changed during the past three decades? (2) How much does climatic variability contribute to crop yield variability in each county? (3) How much does climate change affect the spatial pattern of crop yield, and what methods can be used to measure it?

2. Materials and Methods

2.1. Crop Yield Census and Climate Data

The county-level yearly crop yields for corn, rice, and wheat from 1980 to 2014, along with the harvest area, were obtained from the Ministry of Agriculture and Rural Affairs of China. Then, outliers of biophysically unattainable records were eliminated. Additionally, both missing data for periods of more than 15 years and continuous missing data for periods of more than 5 years were discarded. Missing values were interpolated using the 5-year moving average method. Then, the yields of the remaining 1879 counties for corn, 1625 counties for rice, and 1673 counties for wheat were area-weighted and aggregated to 0.5° grids. The daily meteorological data at a resolution of 0.5 grid were obtained from the National Meteorological Center of China (http://www.nmic.cn/ (accessed on 1 January 2021)). These data were based on the temperature (T) and precipitation (P) records of high-density stations (2472 national meteorological observation stations), on which spatial interpolation was conducted using the thin-plate spline (TPS) method. The daily photosynthetically active radiation (PAR) dataset, including 724 routine stations across China from 1961 to 2014, was obtained from the China Science Data Bank (dataset DOI: 10.11922/sciencedb.400). The annual mean, maximum, minimum, and growing period (from May to August) mean of T, P, and PAR were calculated. The time series of yields for the three crops and climate variables from 1980 to 2014 were linearly detrended based on the least square method.

In the Discussion section, in order to identify the movement of the centroids of artificial land and cultivated land, the remote sensing monitoring data on land use in China in 1980, 1990, 2000, 2005, 2010, and 2015 was provided by the Resource and Environment Data Cloud platform of the Chinese Academy of Sciences (http://www.resdc.cn/ (accessed on 1 January 2021)) was used.

2.2. Methods

In this research, linear regression trend analysis was used to estimate the yield trends of each grid cell for each crop. t-tests were conducted to determine the significance of the changing trend at the 10% level. Then, 5-year, 10-year, and 15-year time windows were used to examine the robustness of the results.

The interannual variability of crop yields could be explained by climate variability. The year-to-year variability of the crop yield was measured based on the standard deviation (STD) of annual crop yield anomalies with the linear trend removed. Additionally, the coefficient of variation (CV, i.e., STD normalized by the mean) was also used in this study to assess the stability of the temporal and spatial heterogeneity of the yield and climate over the past 35 years. In addition, a range of statistical models showing the detrended anomalies of temperature (T), precipitation (P), and photosynthetically active radiation (PAR) during a crop’s growing season in relation to the detrended anomalies of yields for each grid were explored. A multiple linear regression model for each crop and grid cell was also constructed using observed weather variables with the census yield.

To further understand the spatial variation and regression relationship between crop yield and climate variables, the centroid model and standard deviation ellipse (SDE) (Welty Lefever, 1926) were used to analyze the centroid shift in direction, range, and coefficient variation (CV) among all crop yield and climate factors. The centroids of average temperature (T), precipitation (P), and photosynthetically active radiation (PAR) during the growing period, and the annual centroids of the yields of three major crops (corn, rice, and wheat) in China from 1980 to 2014 were estimated.

The centroid location of variable v for year t consists of longitude (X_v,t) and latitude (Y_v,t), and the form is expressed as follows:

$$X_{v,t}=\frac{{\sum }_{i=1}^n{V}_{i,t}\times X_{i,t}}{V_{i,t}};Y_{v,t}=\frac{{\sum }_{i=1}^n{V}_{i,t}\times Y_{i,t}}{V_{i,t}}$$

(1)

where X_i,t and Y_i,t are the longitude and latitude of the geographical centroid of grid i in year t, respectively; V_i,t is the crop yields or climate variables V for year t in grid i.

To investigate how much the interannual moving of climatic centroids explains the interannual shift in crop yield centroids, an approach based on the first-difference time series of all centroids (i.e., ΔX_v,t and ΔY_v,t) was used. Then, a stepwise regression of the crop yield centroid onto the centroids of the growing season’s means T, P, and PAR time series was developed (Equation (2)). Due to the limited number of data samples, the bootstrap method was used to reconstruct 500 sample data points and estimate the uncertainty associated with the derived regression coefficients.

The construction of multiple linear regressions of the centroids using annual mean T, P, and PAR was conducted, but the growing season’s climate variables were found to be more correlated with the crop yield centroid. The p-value and coefficient of determination (R²) were used for validating the model’s statistical significance and explanatory power.

$$\Delta C_{crop,t}={\beta }_{t,0}+{\beta }_{t,1}\Delta T_{t,x,y}+{\beta }_{t,2}\Delta P_{t,x,y}+{\beta }_{t,3}\Delta R_{t,x,y}+{\epsilon }_t$$

(2)

where ΔC_crop,t is the first difference in crop yield centroid in year t; ΔT_t,x,y, ΔP_t,x,y, and ΔR_t,x,y represent the first difference in the growing season’s average temperature, precipitation, and photosynthetically active radiation at location x (i.e., longitude) and y (i.e., latitude) in year t, respectively; Βt,0 is the centroid interannual variability caused by urbanization and agricultural management; β_t,1, β_t,2, and β_t,3 are the model coefficients needing to be fit; and ε_t is an error term.

3. Results

3.1. Yield Variability

The areas with the highest average rice yield were mainly distributed in northeastern China and the Yangtze River basin in the past 35 years (Figure 1a). High corn-producing areas were mainly distributed in the northeast, northwest, and north of China and Shandong province (Figure 1b). North China and Shandong province were the main areas with high wheat yields (Figure 1c). In general, the areas with a higher average yield were also the areas with high yield stability, as indicated by the low values obtained for the standard deviation and coefficient of variation. However, parts of Xinjiang showed both a high average yield and a high year-to-year yield fluctuation for corn and wheat (Figure 2). The trend analysis showed that the middle latitude (i.e., northeast China and Xinjiang province) was the main area showing an increase in the production of rice and corn. In parts of northeastern China, wheat had a significant decreasing trend (Figure 1). The areas with corn yield growth were widely distributed and mainly in the northeast, northwest, north, southwest, and central areas of China.

Figure 2d shows the temporal change in crop yields and the climatic coefficient of spatial variation (the standard deviation of a given year divided by the mean value). The high coefficient of spatial variation varied greatly from place to place. For the whole study period, the degree of spatial variation ranked from high to low was P > wheat > corn > rice > T > PAR. The distribution of rainfall in China was extremely uneven, but PAR was relatively homogeneous. Wheat and rice covered large producing areas, while the corn planting area was relatively scattered. There was an abrupt increase in the coefficient of spatial variation for wheat during the period 2004–2010.

After a rapid decline (annual growth rate of −5%) from 1998 to 2003, the total wheat yield entered the stage of recovery growth in 2004. Accompanied by the adjustment of planting structure, the spatial distribution of wheat yield became increasingly homogeneous during that period.

3.2. Climate and Yield Variability

We found the average value of all grids (0.5° × 0.5°) for crop yields, temperature (T), precipitation (P), and photosynthetically active radiation (PAR) for the growing season of each year from 1980 to 2014, and then constructed a linear regression model to explain the impact of climate variability on crop yield variability at the national scale, with R² = 0.71 for corn, R² = 0.78 for rice, and R² = 0.57 for wheat, p < 0.01.

In total, 2725 linear regression models for corn, 2312 for rice, and 2865 for wheat were constructed. Not all crop growing regions showed statistically significant effects of climate variations on crop yield variability, with R² = 0.3 for corn, R² = 0.3 for rice, and R² = 0.22 for wheat. The spatial distribution of R² is shown in Appendix A Figure A1. However, the vast majority of crop harvesting regions did experience the influence of climate variability on crop yields, with ~73% of corn-harvesting regions, ~74% of rice-harvesting regions, and ~46% of wheat-harvesting regions showing variability.

The effect of interannual climate fluctuations on yield varied by crop type and region. Rice and corn were more sensitive to climate variability than wheat. Consistent with previous studies, temperature [39,40,41] was the most important factor affecting yield, while precipitation had a greater influence on the fluctuation range of yield.

During the past 35 years, the increase in temperature (T) has been beneficial to rice production in the northeast, southwest, and southern areas of China, but has harmed rice production in the Yangtze River basin and Hainan province (Figure 3a). For corn, the increase in temperature was more conducive to yields in the southern area of the Yangtze River and had a negative effect in the northeast and northwest regions (Figure 3d). In terms of wheat, the rising temperature was detrimental to most of the main areas of wheat production in northern China (Figure 3g). The warming-induced wheat yield rise in southern China was not statistically significant and was possibly due to the compounding effects of non-climatic factors, such as farmland management.

The increase in precipitation (P) was not conducive to the increase in rice production in northeast China, the middle and lower reaches of the Yangtze River, and the southern coastal areas, but it positively affected north China, southwest China, and parts of Xinjiang province (Figure 3b). The contribution of precipitation to corn yield gain was similar to that of rice in most areas (Figure 3e). In contrast, the increase in precipitation in most parts of China exerted a negative effect on wheat, except for in arid areas in northwest China, Xinjiang province, and parts of Chongqing municipality (Figure 3h).

The increase in photosynthetically active radiation (PAR) had a significant effect on the increase in rice yield in northeast China and the Yangtze River basin, but showed a weak negative effect in Xinjiang and the southern area of the Yangtze River (Figure 3c). As for corn, PAR had an obvious stimulatory effect in Xinjiang and southern China, especially in southwest China (Figure 3f). The increase in PAR mainly promoted the growth of wheat yield in Xinjiang and the northeast and southwest regions, but was not beneficial in the lower reaches of the Yangtze River and the southeast coastal areas.

3.3. Spatial Effect of Climate on Yield Variation

The coefficient of spatial variation reflects the dispersion degree of variable distributions in space. The regression models showed that the spatial variation in temperature played a critical role in reshaping the spatial pattern distribution of yield. P played the least important role, and PAR had the opposite effect. At the 99% confidence level, the spatial heterogeneity of the climate explained 38% of the spatial variation in the rice yield and 51% of the variation in the corn yield. However, the spatial variation in the climate had no significant effect on wheat yield.

The centroid is a physical index that can be used to indicate the spatial distribution and intensity of the climate and crop yields. Its trajectory reflects a comprehensive process of crop yield response to the climate spatial variation. The larger the range of the centroid’s movement is, the worse the spatial stability of the element will be (Figure 4). In the past 35 years, the movement range of the centroids from high to low has been P > wheat > corn > rice > PAR > T. This result is consistent with the above analysis of spatial variation coefficients (Figure 2d). The centroid movements of corn, P, and T are more significant in the northeast-southwest direction. Additionally, the movement of wheat and PAR took place in the east-west direction. From 1980 to 2014, the centroids of rice, wheat, and P moved to the north by 82.72 km, 83.39 km, and 25.27 km, respectively. The centroids of corn and T moved to the southwest by 33.47 km and 19.49 km, respectively. Additionally, the centroid of PAR moved to the west slightly by 15.9 km. The details of the interannual movement of each centroid are shown in Appendix A Figure A2 and Figure A3.

Furthermore, we calculated the trajectory of the centroids of cultivated land and artificial land from 1980 to 2015. As shown in Appendix A Figure A3, the centroid of artificial land moved obviously from east to west, while the centroid of cultivated land moved from south to north.

Pearson correlation analysis demonstrated that T showed a positive correlation with the year-to-year variation in the centroids of corn (r = 0.31) and rice (r = 0.44). P exhibited a negative correlation with the variations in the rice centroid (r = −0.43), and PAR showed a negative correlation with the variations in the wheat centroid (r = −0.31). In addition, it was found that corn showed a significant positive correlation with the rice centroid (r = 0.64) movement, while P showed a significant negative correlation with T (r = −0.75) and PAR (r = −0.79). All values were above the 95% confidence level.

The construction of a linear regression model showed that only the centroid variation models of corn (R² = 0.24) and rice (R² = 0.28) in the latitude direction passed the f-test at the 95% confidence level, while these were at the 90% confidence level for wheat (R² = 0.16) after excluding temperature effects. This implied that the spatial variation in the climate had a weak effect on the spatial interannual variation in wheat, and the spatial variation in temperature had no significant effect on the spatial distribution of crop yields. Sensitivity analysis indicated that, for every 1 km shift in the temperature centroid, the corn and rice yield centroids moved by 0.97 km and 0.34 km, respectively. Precipitation had a slightly negative effect on the rice yield centroid variations. For every 1 km shift in the precipitation centroid, the yield centroids of rice and wheat moved by 0.14 km and 0.15 km in the opposite direction, respectively. Rice is mainly planted in the middle and lower reaches of the Yangtze River basin, which is prone to flood and waterlogging disasters because of its flat terrain and abundant summer rainfall. PAR also played a negative role in all the crops’ centroid movements in the latitude direction. T and PAR were highly positively correlated (r = 0.68), but they showed the opposite effect on the movement of yield centroids in this model. This is closely related to the distribution of surface and atmospheric environments in China (i.e., DEM, atmospheric humidity, and cloud distribution) [42]. In general, the direction changes of T and PAR were consistent, but in China, P was negatively correlated with the change in the centroid of T. P was accompanied by the joint influence of clouds and water vapor, leading to a reduction in PAR. Thus, the climate co-variability weakened the yield increase brought by warming.

4. Discussion

4.1. Selection of Spatial and Temporal Scale

The selection of a spatial-temporal scale is a key aspect of geography research [43]. Crop and climate observation data, however, are typically measured with national or subnational surveys that obscure geographic variability across nations, regions, provinces, and localities.

Compared to a separate linear model established for each 0.5° grid (similar to the county-level scale), the effects of climate variability on yield variability may be overestimated at the national scale. Scale is the spatial or temporal dimension of an object or process under study and can be described in terms of resolution and scope, which marks the level of knowledge of the details of the object under study. Even when using the same set of data, different spatial scales will give different calculation results. This demonstrates that access to fine-scale data is a necessity in the study of the effects of climate change on the spatial production of food.

The explanatory power of climate variation on yield has increased over the years. For example, the precipitation in the middle and lower reaches of the Yangtze River basin changes dramatically over the growing seasons, often leading to flooding and waterlogging disasters. In addition, the selection of the growing season zone is an important aspect of the temporal scale [44]. There are also some improvements that can be inserted here, such as selecting different crop phenological periods in different regions as the growing period [45]. In the future, the different effects of regions and varieties on growing degree days (GDDs) should be considered, and regression models based on agroclimatic indicators should also be constructed.

Thus, large spatial and temporal scales may conceal more detailed physical interactions between climate and yield and can lead to an overestimation of the effect of climate on yields. More multi-scale analyses are needed in order to verify this scale effect.

4.2. Explanation for Shift of Crop Yield Centroids

Against the background of global warming and urbanization, China’s main grain-producing areas and crop production centers are transferred. China’s granaries are quietly moving from south to north. In addition to climate change, these differences are also caused by socio-economic factors, such as the adjustment of land use structures, the abandonment of farmland due to migrant workers, and the encroachment of urban areas into farmland.

The centroid regression model was used to explore the correlation between climate variations and yield centroid shifts. Assuming that there is a kind of gravitational force between the centroids, the movement of the climate centroid may lead to a shift in the crop yield of the centroids. Thus, the centroid distance between climate and crop yield should be constant without other factors. The distance between rice and T, P, and PAR changed most steadily, indicating that rice was greatly affected by climate factors and less affected by other factors. However, wheat was the crop that was most affected by non-climatic factors, especially in 2011. The distance between the centroids of each crop and the centroid of precipitation fluctuated greatly. This showed that the spatial variation in precipitation was more dramatic, but its influence on the crop production pattern was weak. These findings are consistent with those of a previous centroid regression model [46].

The impact of climate change on agriculture is profound, and differences in the level of management, as well as the effects of crop species on yield, cannot be ignored [47]. However, the impact of urbanization on arable land distribution and soil quality remains unclear [48]. The lack of knowledge on this topic will inevitably affect the accuracy of grain yield estimation on a large regional scale. In this study, the centroid model was employed to explore centroid movement trajectories and the correlation between climate and the yields of three crops. It was found that the stability of temperature was the highest, but the temperature variations in the growing seasons had the most significant effect on the reshaped yield pattern distribution. Other scholars have come to similar conclusions [46,48]. In addition, the pattern changes of urban land and cultivated land were found to be directly related to the yield distribution. Since China’s economic reform and opening up, the rapid onset of urbanization has led to a large amount of arable land being turned into sites for urban construction [49,50]. China’s economic development and urbanization process started in the eastern coastal areas and gradually moved to the central and western regions. The changed direction of the cultivated land’s centroid from the south to the north was consistent with that of the yield of the three grain crops studied in this research, especially that of rice. It was found that the northward movement of rice-producing areas was affected by climate spatial distribution changes and restricted by the change in urbanization and farmland patterns. However, the difference was that the change in the centroid of yield was an active adaptation to the climate, while the change caused by the urbanization of cultivated land was due to passive stress. The spatial change in the grain production capacity was not only affected by climate change and the land space squeeze in urbanization, but also affects the environment in turn. This can lead to reductions in carbon storage, decreases in water and soil quality, and decreases in biodiversity [51,52,53].

4.3. Limitations and Future Study

Non-climatic factors were not considered in this study. The use of agricultural film, fertilizers, and pesticides has increased food production, but at an environmental cost [54].

The characteristics of spatiotemporal heterogeneity were not considered enough in this study. The climatic yield correlation between regions has heterogeneity, but also a geographical correlation, such as spatial autocorrelation. Although our research is based on a scientific statistical analysis framework, the interaction between the cells was not considered to be sufficient. For further studies, geographical detectors (Wang et al., 2010), geographically weighted regression [55], and other geostatistical methods should be considered in exploring the internal influence mechanisms.

In addition to the independent factors, the effects of climate coverability factors and joint risks on yields should also be considered. Extreme events can be determined based on a combination of climate factors, such as droughts, floods, and heat waves, which can also have a significant impact on crop growth and production [56]. For example, agricultural drought can be understood through the relationship between temperature and precipitation, but its internal mechanisms are difficult to directly observe and analyze [57]. To answer this question, more complex statistical models, including hierarchical linear models, causal models, and machine learning, are needed.

5. Conclusions

In this study, it was found that the relationship between climate variation and yield varies across regions and crop species. Temperature is the most important long-term factor affecting crop yield. The main factor affecting the interannual fluctuation of crop yield is precipitation. It was also found that wheat had the least response to climate variation among rice and corn. Based on these results, the spatial distribution and variation characteristics of temperature should be considered when planning and adjusting agricultural planting structures. Meanwhile, precipitation factors, irrigation water sources, and other types of infrastructure should be taken into account to ensure stable grain yields.