Automated Paddy Rice Extent Extraction with Time Stacks of Sentinel Data: A Case Study in Jianghan Plain, Hubei, China

Paddy rice serves as one of the most important crop food globally. The spatial distribution of paddy rice fields plays a fundamental role in describing the rural landscapes, and the precise location and extent of paddy rice fields are of key importance to analyze the subsequent resource allocation, rice yield prediction, and food security. Paddy rice has a distinct character relative to other crops in that the paddy fields need flooding during the initial period of rice seeds preparing and rice transplanting. In order to map paddy rice, remote sensing techniques have long been used to extract and monitor rice crops. Throughout many approaches, phenology-based paddy rice mapping algorithms have been introduced and tested in coarser remote sensing images such as MODIS (Moderate Resolution Imaging Spectroradiorneter), AVHRR (Advanced Very High Resolution Radiometer) and Landsat images. However, the average size of paddy rice fields is commonly smaller than 0.09 ha (e.g., the area of a Landsat pixel). Therefore, a large number of mixed pixels exist, which leads to misclassification. Meanwhile, the phenological indicators used in the previous studies such as LSWI (Land Surface Water Index), and MNDWI (Modified Normalized Difference Water Index) are not feasible to detect the land surface water content at the initial stage of paddy rice seeds preparation and transplanting. Therefore, this study first proposes a new index PMI (Perpendicular Moisture Index) to identify the irrigation in paddy rice fields, and then combines other Vegetation Indices to map paddy rice in Jianghan Plain using time stacks of Sentinel-2 imagery. The results indicates that the proposed PMI can be effectively used as phenological indicator for paddy rice mapping and the Sentinel-2 images provide more detailed spatial distributions. The accuracy assessment suggests that our method has a high accuracy with Overall Accuracy (>97%) and Kappa (>93%). This study suggests that the Sentinel-based automated paddy rice mapping algorithm could potentially and effectively be applied at large spatial scales to monitor paddy rice agriculture.