A field boundary dataset for the Canadian Prairies derived from Sentinel-2 imagery using the Segment Anything Model

Content:
This dataset contains delineated agricultural field boundary polygons for the Canadian Prairies covering Alberta, Saskatchewan, and Manitoba. The field boundaries were derived from automated image segmentation applied to satellite imagery and represent individual crop field extents across agricultural regions. Each polygon includes geometric properties and identifiers that link to spatial regions (RMs). The dataset comprises 457 shapefiles encompassing 656,082 field polygons across the three provinces, providing a comprehensive geospatial representation of agricultural land parcels suitable for agricultural monitoring, land management, and regional analysis applications.

Location:
The data were collected across the Canadian Prairies, including Alberta, Saskatchewan, and Manitoba, covering major agricultural regions between approximately 49°–55° N latitude and 96°–114° W longitude. The processed dataset is stored and maintained at the authors’ institutional affiliation.

Structure :
The dataset is organized as a structured repository containing vector data representing agricultural field boundaries across the Canadian Prairies. The repository consists of the following hierarchy:
/Field Boundaries/
├── /Alberta_Shapefiles/ (zipped field boundary shapefiles for Alberta)
├── /Saskatchewan_Shapefiles/ (zipped field boundary shapefiles for Saskatchewan)
├── /Manitoba_Shapefiles/ (zipped field boundary shapefiles for Manitoba)
├── /geoparquet_AB/ (Geoparquet maps for Alberta)
├── /geoparquet_SK/ (Geoparquet maps for Saskatchewan)
└── /geoparquet_MB/ (Geoparquet maps for Manitoba)
/Metadata/ (metadata files describing the dataset)

Each field boundary shapefile is packaged as a zip file containing polygon geometries in ESRI Shapefile format with standard geometry files (.shp, .shx, .dbf, .prj) and is projected in a consistent coordinate reference system suitable for regional-scale analysis. Additionally, the dataset includes Geoparquet format files for each province, providing a modern, efficient alternative format for the same field boundaries. The boundary data represent delineated field boundaries derived from automated image segmentation.

Format:
Field boundary polygons are provided in two formats:
1. ESRI Shapefile format (as zip archives) with associated geometry files and projection information
2. Geoparquet format for efficient storage and processing

How To Access:
All scripts used for image preprocessing, segmentation, post-processing, dataset assembly, and GEE App development are publicly available in the associated GitHub repository: https://github.com/thuanhavan/CSA_Field_Boundary_Segmentation