Sample-free automated mapping of double-season rice in China using Sentinel-1 SAR imagery

Abstract

Introduction: Timely and accurately mapping the spatial distribution of rice is of great significance for estimating crop yield, ensuring food security and freshwater resources, and studying climate change. Double-season rice is a dominant rice planting system in China, but it is challenging to map it from remote sensing data due to its complex temporal profiles that requires high-frequency observations. Methods: We used an automated rice mapping method based on the Synthetic Aperture Radar (SAR)-based Rice Mapping Index (SPRI), that requires no samples to identify double-season rice. We used the Sentinel-1 SAR time series data to capture the growth of rice from transplanting to maturity in 2018, and calculated the SPRI of each pixel by adaptive parameters using cloud-free Sentinel-2 imagery. We extensively evaluated the methods performance at pixel and regional scales. Results and discussion: The results showed that even without any training samples, SPRI was able to provide satisfactory classification results, with the average overall accuracy of early and late rice in the main producing provinces of 84.38% and 84.43%, respectively. The estimated area of double-season rice showed a good agreement with county-level agricultural census data. Our results showed that the SPRI method can be used to automatically map the distribution of rice with high accuracy at large scales.