A Multi-Temporal Rice Mapping Method Combining SAR Interferometric Coherence and Backscattering Coefficient Features

Accurate rice mapping using synthetic aperture radar (SAR) in cloudy and rainy areas is of great importance to achieve the United Nations Sustainable Development Goal 2 for 2030. Considering the influence of water and shores on rice mapping results in current multi-temporal SAR rice mapping methods, a deep learning multi-temporal SAR rice mapping method combining temporal SAR interferometric coherence features and backscatter coefficient features is proposed in this paper. By analyzing the temporal backscatter coefficients of rice VH polarization and the temporal coherence of VV polarization, three temporal features of temporal backscatter coefficient variance, range and coherence time-varying features are constructed for deep learning rice mapping. Taking Kampng Chhang and Kampng Chham provinces in Cambodia as the study area, the experimental results show that the proposed three features well suppress the influence of water and shores on rice mapping and achieve an overall accuracy of 91.96% on the test sample. Thus, the proposed method can extract rice distribution information efficiently and accurately in cloudy and rainy areas.