NASA Harvest Uses Satellites and AI to Map Every Farm Field on Earth

Knowing how much food the world will produce this year sounds like a simple question. It is not. Most farming statistics arrive months after harvest, compiled from paper surveys and guesswork. When war, drought, or flood disrupts a growing season, governments and food aid organizations are often flying blind. NASA Harvest, a consortium led by the University of Maryland with NASA and the USDA, exists to fix that. It applies machine learning to satellite imagery so that crop maps and yield forecasts arrive while there is still time to act, and it matters because early, accurate information determines whether food aid reaches the right people before hunger turns into famine.

The scale of what the team has built is quietly enormous. Since November 2017, NASA Harvest has initiated or contributed to roughly 30 projects around the world, from East Africa to Latin America to Eastern Europe. The common thread is turning pixels into policy grade agricultural intelligence.

Counting Wheat in a War Zone

The program's most famous moment came in 2022. When Russia invaded Ukraine, one of the world's largest wheat exporters, global grain markets panicked and forecasting collapsed. Nobody could verify how much wheat had actually been planted or harvested. Working with satellite operators including Planet, the NASA Harvest team analyzed imagery across Ukraine and calculated that farmers had harvested 26.6 million tons of wheat in 2022, several million tons more than leading forecasts expected, and found that a larger share than predicted had been gathered from fields inside the war zone.

That single number mattered far beyond commodity markets. It informed decisions on grain corridor negotiations, food prices, and aid planning across import dependent countries in Africa and the Middle East. Without satellite based machine learning, the world would have relied on rumor and stale projections for one of the most consequential harvests in decades.

Why Smallholder Fields Are the Hard Problem

Mapping industrial farms in Kansas or France is comparatively easy. The fields are huge, uniform, and well documented. The genuine technical challenge is the smallholder farm, often smaller than a football pitch, sometimes intercropped with two or three species in the same plot, and completely absent from any official record. This is how most of sub Saharan Africa farms, and it is precisely where food security data is weakest.

NASA Harvest's researchers developed transfer learning and weak supervision techniques that let models trained on a few well labeled regions generalize to vast unlabeled ones. The result is automated crop type mapping and field delineation at national scale, within the growing season rather than a year after it ends. Their published work on unlocking large scale crop field delineation in smallholder systems has become a foundational reference for the field.

From Maps to Food on Tables

The maps are not academic exercises. They feed directly into programs that move money and food. In Uganda, NASA Harvest customized its satellite based GLAM monitoring system for East Africa, enabling the World Bank's Disaster Risk Financing and Insurance Program. That program has supported more than 300,000 people in the Karamoja region, providing alternative livelihoods to smallholder farmers hit by drought. When the model signals crop failure, payouts and support can trigger before the situation becomes catastrophic.

In Togo, a NASA Harvest crop map built in 2019 became the foundation for the YOLIM program, which served more than 50,000 people with agricultural support. In each case the pattern is the same: satellite data plus machine learning produces a trustworthy, timely picture of harvest conditions, and that picture unlocks financing and aid that would otherwise arrive too late.

Building Capacity, Not Dependency

A distinctive part of the NASA Harvest model is that it does not simply hand over maps. The consortium works directly with national ministries and regional agencies, training local analysts to run the monitoring systems themselves. The stated goal under the Zero Hunger agenda is to help developing countries integrate Earth observation and machine learning into their existing agricultural monitoring frameworks. The tools transfer. The expertise stays.

The people behind the program have received real recognition. Harvest Africa director Catherine Nakalembe was named a 2020 Africa Food Prize laureate for her work using satellite data to protect millions of farmers across the continent. Former Harvest AI lead Hannah Kerner was named to the Forbes 30 Under 30 list in science, and chief scientist Christopher Justice received a NASA Distinguished Public Service Medal.

The Bigger Picture

Climate change is making harvests more volatile, and the countries least responsible for emissions often face the sharpest food shocks. In that world, the gap between what is actually growing and what governments think is growing can be measured in lives. NASA Harvest's contribution is narrow and profound: close that gap with physics grade observation and industrial scale machine learning, then hand the capability to the institutions that need it most.

A satellite passes over a patch of sorghum in Karamoja. A model trained on thousands of labeled fields decides what is growing there and how it is doing. A ministry in Kampala sees the forecast weeks early. A payout reaches a village before the empty season peaks. None of this requires a science fiction future. It is happening now, one orbit at a time.