Integrating sentinel-1 SAR and sentinel-2 optical data for crop mapping and flood inundation assessment in agricultural systems

Abstract

Precise and timely evaluation of the impacts of flooding on agricultural systems is essential for planning food security strategies and responding to disasters, especially in key crop-producing areas. In this study, an integrated remote sensing approach that uses Sentinel-1 and Sentinel-2 multispectral imagery is proposed to identify crop types and assess the extent of flooding and its impacts on agricultural systems in the prominent agricultural area of the Jianghan Plain, China, specifically in the Jingzhou study area, from 2018 to 2022. Flood inundation mapping during a significant flood event in July 2020 was conducted using dual-polarization (VV and VH) Sentinel-1 C-band SAR data. Flooded regions were extracted using a change-detection approach coupled with histogram-based thresholding (Otsu’s method) of the VV/VH ratio, enabling a clear distinction between floodwater and non-flooded regions, as well as between permanent water bodies and non-flooded regions. The random forest classifier for Sentinel-2 time-series imagery in 2020 achieved an overall accuracy of 92.7% and a Kappa of 0.892. It classified the total agricultural area as 11,442 km², of which rice, wheat, and cotton were grown on areas of 888 km², 384 km², and 43 km², respectively. The study demonstrates the potential of combining optical and SAR data for effective, crop-specific flood monitoring in near real time.