Mapping County-Level Rice Planting Areas by Joint Use of High-Resolution Optical and Time Series SAR Imagery

Timely and accurate mapping of rice spatial distribution is needed for ensuring food security, managing water usage, and optimizing agricultural production. Frequent cloudy and rainy weather during the rice growing season presents challenges in constructing comprehensive time-series features from optical images. In addition, the fragmentation and sparsity of farmland parcels within the county lead to low extraction accuracy. To address the above challenges, this study proposed an automated rice mapping framework for county-level rice mapping in cloudy and rainy regions by integrating the strengths of high-resolution optical and time-series Synthetic Aperture Radar (SAR) imagery. First, the HRTSNet model was developed to extract farmland parcels from GF-6 high resolution imagery. Subsequently, the long short-term memory (LSTM)-based temporal classification model was utilized to acquire rice cultivation information at parcel scale using time-series Sentinel-1 SAR data. The proposed method was validated at two counties in China. The results showed that the HRTSNet model achieved the highest mIoU and delineated the closest boundary maps to ground truth in extracting farmland parcels from high-resolution optical imagery. And the proposed method effectively integrated limited GF-6 imagery with time-series Sentinel-1 data and performed better than traditional machine learning algorithms like Random Forest and pixel-based methods, achieving an overall accuracy of over 88% and a Kappa coefficient of over 86% for the Dangtu and Rudong counties. In addition, the classification accuracy was effectively improved by incorporating the DPSVIm index. The results provide a potential solution for mapping county-level rice planting areas with limited optical imagery.