A Spatialization Method for Grain Yield Statistical Data: A Study on Winter Wheat of Shandong Province, China

Core Ideas Dividing the study area into subregions and classifying them improved winter wheat classification accuracy.Single‐phase Normalized Difference Vegetation Indices acquired on 6 March, 23 April, 25 May, and 29 June were the best variables for building the wheat yield spatialization model.The extraction accuracy of winter wheat area greatly impacted the spatialization of yield.The proposed model can provide a technical reference for producing high‐resolution crop yield distribution maps. Grain yield data based on administrative divisions (counties, cities, etc.) for statistics lack spatial information, which can be effectively solved by grain yield spatialization. This paper proposes a spatialization method for grain yield based on the Moderate Resolution Imaging Spectroradiometer (MODIS) Normalized Difference Vegetation Index (NDVI) time series data. The method was tested by taking winter wheat (Triticum aestivum L.) in Shandong Province in China as an example. First, the classification and regression tree (CART) algorithm was trained to extract the winter wheat planting pixels in 2016. The average NDVIs of the different growing stages (returning green, jointing, heading, and milk ripening) were calculated from the MODIS NDVI time series data. The relationship between winter wheat yield and NDVI variables (including single‐phase NDVI and the average NDVI of different growing stages) was analyzed by univariate and multiple linear regressions. The NDVI variable with the highest correlation to winter wheat yield and the minimum root mean square error of the fitting equation were chosen as input to build the spatialization model. The results show that the classification accuracy of winter wheat estimated with the confusion matrix was 82.51% and that the average precision of planting acreage compared with county‐level statistical data was 87.64%. The average relative error of yield spatialization at the county level was 22.71%. The method developed in this paper is easy to operate and popularize, and it can provide a technical reference for producing high‐resolution crop yield distribution maps of long time series through spatialization.