Early identification of rice growing areas based on Sentinel-1 SAR time series data

Rice is a crucial staple crop, and timely information about its cultivation areas is essential for effective resource allocation and food security. Traditional optical remote sensing faces challenges due to cloud cover and spectral similarities with other crops. In contrast, radar remote sensing provides all-weather capabilities, making it ideal for extracting rice planting areas over time. This study analyzed backscattering characteristics in three major rice-producing regions—Chongming Island, Liaohe Plain, and Poyang Lake—using machine learning techniques. The findings revealed that dual-polarization backscatter data (VV+VH) significantly improved identification accuracy. The optimal time window for achieving the highest accuracy (over 0.92) was six acquisitions, particularly during the rice transplanting stage. The developed model demonstrated strong generalization capabilities across varied phenological conditions in the three regions.