Automatic Mapping of Maize Cultivated Area Based on A Novel Maize Mapping Index

Maize is an important grain crop worldwide, and accurate maize mapping is essential for agricultural production and food security. However, most previous studies on maize mapping rely on training samples, which still face challenges regarding the absence of field samples. In this study, a novel Maize Mapping Index (MMI) was proposed for the automatic identification of maize cultivated area based on Sentinel-2 Multispectral Instrument (MSI) images. The MMI was developed at California, Iowa, Arkansas and Texas sites in the United States (U.S.) and further validated in Shanxi Province, China. The proposed MMI considers differences in crop phenology by using the features within an adaptive temporal window, as indicated by the normalized difference red edge (NDRE). The results indicated that the reflectivity of both the red-edge 1 and shortwave infrared 1 of maize were less than those of the other crops at the peak time of NDRE. Additionally, the spectral angle at the red-edge 2 band of maize was considerably greater than that of the other crops. On the basis of these findings, the MMI was constructed to distinguish maize from nonmaize within the cropland area. The MMI for maize was greater than that for other crops. Compared with the Otsu and Rosin methods, the Kapur threshold method achieved a better performance in the determination of the MMI threshold. The maize maps generated by the MMI and Kapur method across various study sites achieved overall accuracies and Kappa coefficients greater than 84.36% and 0.69, respectively, and no training samples were required. The coefficients of determination (R2) between the detected and statistical maize areas in Shanxi Province were 0.84 and 0.77 at the city and county levels, respectively. These findings provide insights for maize mapping at a large scale where field samples are absent.