Spatio-Temporal Patterns And Influence Factors Of Winter Wheat Production In China

Food security is a critical strategical issue to maintaining domestic economic development,social stability,and national independence in the world.At present,wheat is one of China’s three major food crops and has the world’s most extensive planted acreage and the highest total production.From 2004-2015,China’s wheat production increased every year and ensured the food security of China.Moreover,winter wheat represents China’s 89.6 percent of total wheat production and 81.8 percent of the wheat-planted area.Hence,revealing the spatio-temporal evolution of winter wheat production and quantifying the driving factors is an urgent research topic in wheat production.We collected data from statistical yearbooks and built a spatio-temporal dataset with province-level,city-level,and county-level wheat production data.Then,we developed a novel spatio-temporal interpolation method based on the backpropagation(BP)neural network with spatiotemporal heterogeneity for interpolating missing data in the dataset.With the dataset,the methods integrating machine learning,spatial analysis,and multi-scale techniques were applied to reveal the spatio-temporal patterns and driving factors of China’s winter wheat production.The research results will provide a scientific reference for making national or regional crop production policies and plans.Incomplete yield series data constrain their application and result in inconvenience in information mining.This study improves the existing spatio-temporal interpolation method and succeeds in interpolating wheat yield data recorded by 889 counties in China’s winter wheat areas from 2004 to 2015.In the new method,pre-interpolation is first implemented to improve interpolation’s completion rate and eliminate the effect of absent neighboring values on the positions to be interpolated.Then,based on spatio-temporal heterogeneity,data reconstruction is performed in spatial and temporal dimensions.Finally,the reconstructed results are combined with the BP neural network model for spatio-temporal integration.The proposed method shows the superiority in interpolation precision and the completion rate compared to the traditional interpolation methods.It can predict all missing data,and the precision accuracy is increased by 43 percent from the traditional temporal interpolation method and 27 percent from the traditional spatial interpolation method.Moreover,the artificial network model is better than the multivariate linear regression model(MLR)or support vector machine model(SVM)in the proposed method.The results of this method will provide a useful reference for interpolating yield data on other cultivars.With the well constructed spatio-temporal dataset,the characteristic of spatio-temporal China winter wheat patterns from 2004 to 2015 is revealed by multi-scale spatio-temporal analysis.Spatial autocorrelation analysis shows that both indicators,yield and proportion of wheat planted,have strong spatially correlated characteristics in the region.The Huanghuai Plain and Fenwei Valley are intensively planted areas,and the proportion of wheat planted decreases to the north and south from these two regions.Most counties’ total production is increasing while the yield difference per acre among counties is decreasing.The Huanghuaihai Plain is the hotspot of the high-yield area,and the wheat yield decreases radically from this center.The centroid trajectory of wheat production suggests that the area sown dominates the region’s directional distribution of total wheat production.The centroid analysis,based on total yield and planted acreage,captures winter wheat production’s movement.From 2004 to 2015,the winter production centroid continues moving east.The movement speeds and directions obtained from county-level data,city-level data,and province-level data are almost identical.However,the centroid’s moving speed changed significantly in 2008.The centroid’s moving speed of the total production is 7.75km per year before 2008,while only 2km per year after 2008.Furthermore,the centroid movement shows different patterns in four sub winter wheat area.The centroid in the North sub-region shifted obviously,while only changed slightly in the Huanhuai subregion.From the above,wheat production in the Huanghuai subregion has a significant industrial agglomeration.This subregion stimulates the surrounding wheat production and efficiency,which is the main reason for the increment of China’s wheat production.Moreover,the wheat-planted area in the low-yield region decreases,and it will be a risk in the national food supply.All these results should be referenced when making national or regional wheat production plans.Cluster analysis is used to delineate subregions with similar spatio-temporal patterns and shows the evolution of wheat production.We perform use clustering and spatio-temporal clustering respectively on the county-level time series data of wheat yield and annual variation from 2004 to 2015.The spatial autocorrelation index is used to assess the spatial continuity of clustering results,and geo-detector is for evaluating the heterogeneity of clustering results on levels and trends of yield per acre.It is found that the best clustering results are obtained by spatio-temporal clustering on annual variation series.It has 0.66 for the Moran’I,and the explanatory power of the yield level and yield trend are 0.634 and 0.275,respectively.Secondly,according to the optimal clustering results,the main winter wheat region can be divided into seven wheat yield subregions labeled with high/medium/low yield level and high/medium/low yield increment trends.The high-yield region is the Huanghuaihai Plain,and the average yield in the area is close to 6000kg/ha.The area along the Yellow River in Shanxi,Hebei,and Shandong provinces has the highest yield increment,and the average yield has increased by more than 1100kg/ha during the past 12 years in this area.The patterns of yield change and planted acreage change indicate an apparent agglomeration in China’s winter wheat production.More fields are used to grow winter wheat in the area with a high yield level and high or medium yield increment.The wheat-planted area has increased by 16.53 percent and 14.38 percent in these two subregions,respectively.In the future,the results of this zoning could be used for planning planting distribution strategies and agricultural management program.Based on the Random forest model,an integrated study of the combined effects of natural conditions and agricultural resource inputs on winter wheat yield in the Huanghuaihai region is conducted.The study aims to discover the impact of element indicators on wheat yield and the spatial distribution of feature contribution.Firstly,a random forest model is built to explore the relationship between wheat yield and meteorological,soil,and agricultural resource inputs.The best combination of factor indicators in the model is irrigation ratio,soil pH,soil carbon content,fertilizer input,agricultural machinery input,labor input,pesticide input,and precipitation during the growing period.Moreover,the irrigation ratios and soil pH are the two most critical factors and can improve modeling accuracy by more than 8%,twice as high as the other elements.Partial dependency plots are used to describe how each factor affects the model’s predictions.The results showed that the marginal effect all factors have on the predicted yield.The input-output efficiency of all resource input elements decreases as inputs increase.The irrigation ratio is the most critical factor.Finally,we map each factor’s contribution and reveal the spatial differences in the factors’ contributions.The maps show that the irrigation ratio increases yield in most areas.The overall agricultural mechanism inputs in the Huanghuaihai Plain increase steadily;however,it is higher in the north than in the south.Moreover,the contributions of fertilizer inputs and pesticide inputs are randomly distributed in space,while natural factors show evident spatial continuity.This study indicates that agricultural resource input is the primary driver of the high wheat yield of the Huanghuaihai Plain.Meanwhile,irrigation,fertilizer,and pesticide inputs began to decrease in some counties gradually.These results will be useful in agricultural resources management and help ensure the food security of China.