Skip to main content
Back to timeline
Ai2 BlogSource publication:

Ai2 and Taylor Geospatial open up geospatial AI: models, benchmarks and field-boundary data free to use

Synopsis

Ai2 and Taylor Geospatial announced a partnership that releases geospatial AI models, benchmarks and datasets under open licenses, so conservation, agriculture and humanitarian practitioners do not have to start from scratch. The partnership combines Benchmarks of The World for comparing models, OlmoEarth, an open family of Earth-observation foundation models with a no-code fine-tuning platform, and Fields of The World, an open ecosystem producing a global farm field-boundary dataset from satellite imagery. That dataset is being expanded for higher boundary precision and to include crop type and yield, and all outputs are free on Source Cooperative and GitHub.

AI-generated editorial illustration: Ai2 and Taylor Geospatial partner to make geospatial AI easier to compare, run, and build on

Interpretation

Open geospatial AI resources are being organized into a shared public foundation that conservation, agriculture and humanitarian practitioners and policymakers can compare, fine-tune and reuse. Such work has often lagged behind model advances because few teams have the resources to pretrain, evaluate and maintain open Earth-observation models. The partnership releases model comparison, model operation and dataset production under one set of open licenses, covering the path from evaluation to application.

The partnership consists of three separately usable components that address choosing a model, running a model and producing data. Benchmarks of The World compares models, OlmoEarth is an open Earth-observation foundation-model family with a no-code platform for fine-tuning and running models globally, and Fields of The World is an open ecosystem producing a global farm field-boundary dataset from satellite imagery. Fields of The World already has a user advisory group of 13 experts from government agencies, UN organizations, nonprofits, universities and geospatial companies, who review its data and models and test them in their own work.

Perspective

The resources target people who build, evaluate or depend on geospatial foundation models, and outputs are released under open licenses on Source Cooperative and GitHub; the partners plan to bring similar practitioner input to the shared benchmarks and are asking for input on which benchmarks belong in a shared core.

The Fields of The World expansion for higher boundary precision and crop type and yield is still in progress; how practitioner input will shape the shared benchmarks, and which benchmarks they will finally contain, is still open to input.