Field boundary delineation with seasonal sentinel 2 imagery using Segment Anything Model (SAM)

Accurate field boundary delineation is critical for accurate modelling on crop yields and for precision agriculture (PA), enabling site-specific management to optimize resource use and crop productivity. Traditional boundary mapping methods, such as manual digitization and semi-automated extraction from farm machinery, are labor-intensive and challenging to apply at large scales. Advances in high-resolution land cover data and satellite imagery offer scalable solutions for automated field boundary extraction. In this study, we propose a fully automated workflow that integrates a pre-trained foundation model, the Segment Anything Model - SAM [1] with time-series Sentinel-2 imagery. Seasonal composites of Red, Green, and Blue bands were generated at different phenological stages to support segmentation. The method was applied across over 32 million hectares (79 million acres) of cultivated land in the Canadian Prairies, achieving an intersection-over-union (IoU) accuracy of 0.86 compared to manual segmentation. The workflow consists of four main steps: (1) setting the python working environment, (2) seasonal image acquisition and preprocessing using Google Earth Engine via Python API; (3) field boundary segmentation using SAM; and (4) post-processing and feature cleaning using ArcGIS Pro. This approach demonstrates a scalable, efficient solution for large-scale field boundary mapping to support PA applications.•Integrates a foundation segmentation model (SAM) with Sentinel-2 seasonal imagery•Demonstrates high-accuracy, large-scale automated field boundary delineation•Provides a reproducible workflow adaptable to other regions and datasets.