Recognition of multi-season rice in a complex tropical agronomy zone using time-series SAR data: a case study of Hainan, China

Abstract

ABSTRACT Rice plays a significant role in regional food supply and international food trading, especially for Asian countries. However, accurate mapping of tropical rice cultivationis a challenging task due to the flexible planting calendar and the complicated topography. With the all-day and all-weather imaging ability, Synthetic Aperture Radar (SAR) provides an encouraging resolution for this task. In this research, we presented a case study in Hainan, China that identify the cultivation patterns of rice. As the only tropical province in China, Hainan has abundant crop resources and diverse rice cultivation practices. First, our previously proposedrice mapping method based on time-series Sentinel-1 data and U-Net modelwas applied to Hainan to generate the candidate rice fields in 2019. Then, a start of season (SOS) detection strategy was proposed to discriminate rice cultivation patterns. The accuracy of the annual rice map and the multi-season discrimination results were validated in pixel level and parcel level, respectively. The annual rice map achieved an overall accuracy of 95.03%. Besides, for 313 out of 360 rice parcels, the cultivation patterns were correctly identified. These results proved the effectiveness of phenology information of time-series SAR, and the proposed scheme will be easily extended to complex tropical agronomy zones.