For the Digital Public Good

Innovation Portfolio

Fields of
The World

Expanding the world's first open ecosystem for global agricultural field boundary detection from satellite imagery — combining a global benchmark dataset, baseline ML models, inference tools, and web applications.

Fields of The World is built on open collaboration. Connect with the team and join us in our mission to use AI and ML to identify important information in satellite imagery at global scale.

Explore FTW

Dig in to the full ecosystem behind Fields of The World

Find Us on GitHub

Contribute to the open source ecosystem on github.com/fieldsoftheworld

Community Discussions

Join the lively group of experts discussing FTW on Slack

Slack invite link coming soon

User Advisory Group

Contact us to join the experts providing input and guidance as work progresses

Technical Community Meetings

Join the Google Group to receive invitations to regular meetings

Access all of the FTW Data

Access all of the data through source.coop

Program Overview

Fields of The World is a groundbreaking program designed as an "innovation bridge" between academic research and industry. In its first phase, FTW released the largest benchmark dataset for training models to infer field boundaries from satellite imagery. In its second phase, FTW released an assessment of over 70 geospatial AI models and their comparative performance, in addition to releasing a high performing model architecture for inferring global field boundaries. By making this ecosystem openly available, we're empowering researchers, NGOs, and governments to better understand and manage agricultural systems worldwide.

Fields of The World Explorer showing global agricultural field boundary predictions on a world map

Global Field Boundaries Explorer web app showing field boundaries along with confidence in the model output

Global Coverage: Comprehensive field boundaries across the globe

Open Dataset: Freely accessible to researchers and innovators worldwide

AI-Powered: Created using advanced machine learning and satellite imagery

In-Browser Analysis: You can run AI models directly in your web browser. No coding or software setup is required to generate field boundaries.

Agricultural Tracking: Supports food security, crop types, and land cover / land use assessments.

Custom AI Development: Developers can download the benchmark dataset from Source Cooperative to train custom geospatial AI models.

Does your organization have a use case for field boundary data? We'd love to learn about your real-world application so it can guide our work.

Research Focus
Benchmark Datasets

AI-ready datasets for training and evaluating field boundary segmentation models, covering vastly different agricultural landscapes across the globe.

Model Training and Evaluation

Off-the-shelf field boundary segmentation models and code for custom training and evaluation. Evaluation metrics that reflect real-world deployment conditions.

Global Inference

Global, multi-year field boundary maps created using FTW-trained models and global satellite mosaics, providing field boundary polygons anywhere in the world.

Open Source & Cloud Native

All data, models, and code are published freely with permissible licenses on public platforms in cloud-native geospatial formats.

Research Partners
Kerner Lab, Arizona State UniversityMultimodal Vision Research Lab, WashUAi2Microsoft AI for Good Lab
Technical Fellows Development

Taylor Geospatial uses technical fellows to bridge the gap between academic research and impactful use. Fellows are elite geospatial experts who work closely with research teams.

Their role ensures that complex science translates into products and tools the industry can actually use. For example, on large projects like global field boundaries mapping, fellows work shoulder-to-shoulder with academic modelers to scale the data for planetary use.

We identify technical fellows right from the community contributing to our work. See the box above for ways to get involved.

User Advisory Group

The Fields of The World User Advisory Group will serve as a standing body of practitioners who ensure that the initiative is designed with users rather than for users.

The group will provide continuous feedback on data products, workflows, documentation, and research priorities while serving as early pilot partners for new datasets and tools. Modeled after the human-centered design principles, the Advisory Group will emphasize real-world application, iterative improvement, and diverse perspectives from organizations using geospatial data to make decisions.

Contact us to join the experts providing input and guidance as work progresses

Related Papers
Related News
Next →

Features of The World
Next →

Benchmarks of The World