CROPSIGHT-US: Crop Type Ground Truth from Street View and Satellite Imagery, 2013-2023

A dataset from 2013 to 2023 provides object-based ground truth for 17 major crop types across the contiguous United States, featuring 124,419 cropland fields. It was created by Zhijie Zhou using Google Street View and Sentinel-2 satellite imagery, with each entry including a polygon boundary, predicted label, and uncertainty metrics. The dataset is hosted on figshare and was last updated in March 2026.

Use Cases

• Train and validate crop type classification models based on polygon boundaries and satellite imagery.
• Analyze prediction uncertainty in agricultural AI models based on the provided entropy and variance metrics.
• Study spatial and temporal patterns of crop cultivation across 294 Agricultural Statistics Districts.
• Develop methods for fusing street-level and satellite imagery for land use analysis.

Strengths

• Covers 124,419 cropland fields, providing a substantial sample size for analysis.
• Includes 17 major crop types and spans 294 Agricultural Statistics Districts across the contiguous United States.
• Provides explicit uncertainty metrics (entropy, variance, confidence) for each prediction.
• Offers a temporal range from 2013 to 2023, allowing for multi-year studies.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic and temporal bias inherent to the street view and satellite imagery sources.

Provenance

Source

figshare, with associated code and interactive application on GitHub and Google Earth Engine.

Collection Method

Created using Google Street View imagery and Sentinel-2 satellite data via a method described in Liu et al. (2024).

Time Range

2013 to 2023

Freshness

Last updated 2026-03-22 06:55:16; freshness should be verified.

Geography

Contiguous United States (CONUS)