AI for Earth Observation & Global Field Boundaries, Phase 2

TGE: AI for Earth Observation Data

TGE’s mandate is to speed up the development and commercialization of geospatial innovation emerging from academic research. We identified a broad technology gap in geospatial science that has potential to significantly expand the market—applying AI and Machine Learning to detect distinct features in satellite imagery. We are singularly focused on opening this bottleneck to lead to even more groundbreaking innovations in using AI to analyze geospatial data.

TGE connects academia and industry to turn research-stage concepts into technology that is quickly usable by many organizations, in order to expand market opportunities. We do this in four distinct ways: gathering stakeholders to understand market needs; ensuring the technology is commonly usable; publishing to the cloud for accessibility; and expanding broad awareness. In our first year, we’ve proven our approach is effective. We are excited to kick off our second initiative.

Phase 1: Building Fields of The World (FTW)

In 2024, TGE supported the development of the largest benchmark dataset for extracting field boundaries from satellite imagery, called Fields of The World (FTW). Fields of The World was published as an academic paper that has been accepted for publication and is recognized for both its contribution to the advancement of AI and also for its social impact. TGE ensured that the effort was aligned to market needs, commonly usable, and accessible on the cloud: users can view the benchmark data, examine the fiboa schema repository, explore tutorials for running the model, get instructions to further experiment, and access the data itself.  To expand awareness, we published more than a dozen blog posts describing the work as it progressed.

Phase 2: Expanding FTW

Building on Phase 1, TGE will continue to bridge academic research teams and industry, adding a particular focus in Phase 2 on producing field boundaries data in collaboration with users in sustainable agriculture, supply chain, and food security in order to get feedback on usability.

We are looking to add two roles to our initiative:

• Sponsor: We want our sponsors to shape our priorities and give us feedback on early progress by engaging weekly with our community. In return, these organizations have access to early data and results. A sponsor is interested in influencing the direction of FTW because they can see a future benefit from tools to create global field boundaries from satellite imagery.
  

• Stakeholder: Stakeholders are organizations interested in staying aware of our progress and evaluating results in the middle and end of our effort. For this group, we’ll host progress sessions every other month with a forum to incorporate general feedback.

Work threads will include data publishing, continued academic research and usability/scaling.

1. Data publishing teams will publish FTW field boundary outputs and ship datasets in key/high-performing ROIs in collaboration with end-users and domain specialists
2. Academic research teams will
   1. improve the research methods for field delineation in low-performance areas
   2. Increase global coverage through more data and imagery sources to FTW 1.0
   3. Improve scalability of the boundary delineation machine learning process
3. Usability/Scaling teams will increase the usability of FTW in a number of ways
   1. Scale the FTW model input to be applied to broader regions
   2. Create a QGIS plugin for running the models and viewing output
   3. Expand a web app to upload an area, run the model on the cloud, and get results.
   4. Expand an existing GPT/LLM demo to make it more robust