Field-Scale Detection of Crop Seeding Date Using Sentinel-1 Coherence Time Series

Seeding date plays a crucial role in crop monitoring and yield estimation. Traditional methods, which estimate seeding dates by observing phenological stages such as the start of the season and back-calculating, are often affected by environmental and management variations. Detecting seeding dates directly from field changes before and after seeding offers a simpler and more robust alternative. However, existing approaches face challenges in accurately and efficiently determining seeding dates at the field scale. This study proposes a novel method for estimating seeding dates using Sentinel-1 constellation interferometric synthetic aperture radar (InSAR) coherence time series. The method identifies seeding activities by detecting changes in InSAR coherence caused by soil disturbances during seeding and applies thresholding and decision rules to analyze the time series. It also addresses practical challenges, such as successive seeding operations and reseeding due to adverse weather. Additionally, the method leverages the partial overlapping orbital footprints of Sentinel-1 to integrate data from multiple orbits, improving the reliability and accuracy of seeding date estimation. Applied to 89 and 79 experimental corn fields in 2019 and 2020 near London, ON, Canada, the method achieved root mean square errors (RMSEs) ranging from 5.99 to 9.89 days for individual orbits, outperforming the benchmark method. Integrating data from two orbits further reduced the RMSE to 6.74 days in 2019 and 5.59 days in 2020. When tested on over 1000 soybean fields and 2000 corn fields in 2019 and 2020, the method demonstrated strong agreement with local agricultural progress reports.