An automated rice mapping method based on flooding signals in synthetic aperture radar time series

Abstract

Paddy rice is one of the most important staple foods in the world, feeding over 50% of the global population. A quick and accurate mapping of the extent of paddy rice is of critical importance for ensuring food security, studying climate change and monitoring water resources. Based on Sentinel-1A data and rice flooding features, we proposed a rice mapping method called the Automated Rice Mapping using Synthetic Aperture Radar Flooding Signals (ARM-SARFS), in which the key “V” shaped feature in the Sentinel-1A VH backscatter time series rising from the flooding before and after rice transplanting was used for rice mapping. The ARM-SARFS was validated at three study sites in Hubei, Liaoning and Guangdong provinces in China under different rice cropping systems and different geographical and climate conditions. The results showed that even without any training samples, the ARM-SARFS was able to provide a satisfying classification result with an overall accuracy of over 86% and an F1 score of over 0.85 at all three study sites. With the aid of training samples, the classification performance increased further. When compared with the previously proposed Sentinel-1-based rice mapping methods, the ARM-SARFS improved the overall accuracy by 13.3–37.2%, and the most significant improvement was in the producer's accuracy. The sensitivity test showed that the ARM-SARFS is not sensitive to thresholding, and a high classification accuracy can be achieved at thresholds ranging from −0.025 to 0. These results demonstrated the robustness of ARM-SARFS for automated rice mapping with a high accuracy at large scales.