Development of a new index for automated mapping of ratoon rice areas using time-series normalized difference vegetation index imagery

Over recent decades, the global demand for food has continued to grow, owing to population growth and the loss of arable land. Rice ratooning offers new opportunities for increasing rice production and has received renewed interest because of the minimal additional labor input required for its adoption. Regular, regional-scale monitoring of the spatial patterns of both traditional and ratoon rice cropping systems provides essential information for agricultural resource management and food security studies. However, the similar phenological characteristics of traditional double rice and ratoon rice cropping systems make it challenging to accurately classify these cropping practices based on satellite observations alone. In this study, we first proposed an improved phenology-based rice cropping area detection algorithm using moderate resolution imaging spectroradiometer (MODIS) normalized difference vegetation index (NDVI) imagery. A new index, ratoon rice index, was then developed to automatically delineate ratoon rice cropping areas with the aid of a base map of rice in Hubei Province, China. The accuracy assessment using ground truth data showed that our approach could map both traditional and ratoon rice cropping areas with high user accuracy (91.25% and 91.43%, respectively). The MODIS-retrieved rice cropping areas were validated using annual agricultural census data, and coefficient of determination (R²) values of 0.60 and 0.41 were recorded for traditional and ratoon rice cropping systems, respectively. The total area of ratoon rice was estimated to be 1 283.6 km², 5.0% of the total rice cropping area, in Hubei Province in 2016. These demonstrated the feasibility of extracting the spatial patterns of both traditional and ratoon rice cropping systems solely from time-series NDVI and field survey data and strides made in facilitating the timely and routine monitoring of traditional and ratoon rice distribution at subnational level. Given sufficient historical satellite and phenology records, the proposed algorithm had the potential to enhance rice cropping area mapping efforts across a broad temporal scale (e.g., from the 1980s to the present).