Billions of dollars have been invested in helping American farmers adopt conservation practices that protect soil, improve water quality, and strengthen agricultural resilience. But how much are these investments changing what happens in fields across the country?

A team of researchers at Virginia Tech, in collaboration with the University of Maryland, is testing whether a new generation of artificial intelligence models called Geospatial Foundation Models (GFMs) can make it easier to identify where conservation practices are being adopted and understand their impacts over time.

The work is being advanced through Elinor Benami’s selection as a 2026–2027 Virginia Tech Institute for Society, Culture, and Environment Scholar, which includes seed funding to support the research and development of the project.

“Our work connects advances in geospatial technology with the economic questions that matter for conservation decisions,” said Benami, an assistant professor College of Agriculture and Life Sciences’ Department of Agricultural and Applied Economics, who is leading the project. “By applying these tools to economic and policy-relevant questions, we aim to develop more useful information for farmers, policymakers, and others making decisions about conservation investments.”

The researchers will focus on two major conservation practices: cover crops and tillage.

Currently, much of the nation's information about agricultural conservation comes from Agricultural Census statistics released every five years at the county level. These data rely heavily on surveys with declining response rates and generally do not show where practices are implemented within a county, making it difficult to track adoption at the individual field level or understand changes from year to year.

Using AI to fill the information gap

To address this gap, the team will test OlmoEarth, a GFM developed by the Allen Institute for Artificial Intelligence with input from domain experts, including scientists from NASA Harvest and the International Food Policy Research Institute.

OlmoEarth can learn from changes over time and combine multiple sources of satellite information, including optical imagery and radar. This allows it to capture seasonal patterns even when cloud cover limits visibility. In addition, its fully open design and lightweight options for local analysis make it easier to evaluate, adapt, and deploy for public-interest applications, including those with data confidentiality concerns.

The researchers will initially examine field-level tillage and cover crop records from the Chesapeake Bay region, including Virginia and Maryland. They will evaluate how accurately the models identify conservation practices across different landscapes and farming systems, how well the approach transfers to other places and years, and whether it can reduce the need for costly additional field data while still maintaining performance. Observations from other states may also help assess its broader applicability.

Reliable maps of conservation practices could help researchers better understand how adoption affects farm productivity, values, and resilience to issues like floods and droughts. The information could also support more timely monitoring, incentive targeting, and evaluation of major conservation investments.

“Farmers, researchers, and policymakers all want to better understand which conservation practices work, where they work, and what benefits they deliver,” said Mark Reiter, director of the Eastern Shore Agricultural Research and Extension Center. “Finding ways to more quickly and cost-effectively identify where those practices are being adopted, with or without subsidy programs, is an important step toward more robust impact assessment — and ultimately toward directing resources where they can generate the greatest returns for producers and the public.”

The research team also includes Michael Cecil, a NASA Harvest postdoctoral associate at the University of Maryland, and Inbal Becker-Reshef, a research professor at the University of Maryland, the founding and current director of NASA Harvest, and the managing director of Microsoft AI for Good. Reiter is also a professor at the School of Plant and Environmental Sciences and research director for the Alliance to Advance Climate-Smart Agriculture.

The pilot project is slated for completion in mid-2027.

Share this story