An automatic rice mapping method based on constrained feature matching exploiting Sentinel-1 data for arbitrary length time series

Abstract

Rice is one of the staple food crops worldwide, and timely and accurate paddy rice mapping (PRM) is essential to coordinate agricultural production and assure grain security. In cloudy and foggy regions, there are low exploitation rates of optical images, and accurate PRM is a commonly occurring difficulty. To achieve a precise and timely PRM in these regions, an automatic PRM method based on constrained feature matching (Auto-CFM) for arbitrary length time series was proposed using Sentinel-1 Synthetic Aperture Radar (SAR) data, which takes into account the local shape differences of the time-series σ0 VH curves of different land cover types and the overall deviation of the curves due to the discrepancy in rice planting time. Moreover, it solved the problem of high precision extraction of rice when the Sentinel-1 data might suffer from partial missing images. In this study, the PRM was conducted in Hunan Province and validated in Hubei, Guangdong, and Heilongjiang Provinces, which featured different planting times, climates, and topographies. The results showed that the Auto-CFM improved the accuracy by 3%–9% compared to current competing methods and the PRM accuracy exceeded 92% in different validation areas, proving the effectiveness and robustness of the method. To obtain cultivation areas as early as possible, the PRM was performed by reducing the number of images one by one to eventually acquired the rice maps in mid to late August with an overall accuracy of no less than 90%, achieving the goal of access to the spatial distribution of rice with high accuracy before harvest.

Keywords

• Paddy rice mapping
• Sentinel-1
• Constrained feature matching
• Time-series σ0 VH
• Earliest identifiable time