Mapping of paddy field by Sentinel-1 SAR images and cadastre data in Yunlin, Taiwan

Paddy rice is a prominent agricultural commodity in Taiwan; however, conventional ground surveys and censuses proved to be both time and resource-intensive. The utilization of spaceborne Synthetic Aperture Radar (SAR) technologies for monitoring paddy phenology emerges as a transformative approach, offering advancements in crop management and fortification of food security. This study focuses on mapping paddy rice parcels by extracting paddy-specific phenology time series using ESA’s Sentinel-1 SAR data in Yunlin, Taiwan. The analysis incorporates 60 SAR images in 2019, employing a systematic three-step process: (1) Discerning specific paddy phenology curves in designated training sites through the temporal behaviour of SAR backscattering coefficients (BC) in VH and VV polarization and employing Wavelet Transform for signal decomposition and reconstruction across various time scales to eliminate noise; (2) Identifying the start of the season (SOS), end of tillering (EOT) and end of the season (EOS) in a paddy growing cycle based on backscattering coefficients time series; (3) Recognition of paddy fields by matching the specific paddy phenological pattern. Additionally, cadastral maps are integrated as geometric features for object-based classification with enhanced accuracy. The validation process reveals a kappa coefficient of 0.64 for Yunlin, with an overall accuracy exceeding 0.8 across all townships. Notably, Douliu City, Dounan and Dapi Township exhibit superior accuracy due to their high paddy field density. Conversely, Tuku and Yuanchang Township results are less accurate and are influenced by crop diversification.