Using Machine Learning and Satellite Imagery to Estimate Corn Crop Production

Elizabeth Borneman


Descartes Lab, a start-up organization, has created crop production analysis that uses millions of satellite images and machine learning to produce accurate data about the production of agricultural crops. Descartes Lab created a tiled, spatio-temporal mosaic of the planet to make predictions about agricultural production. Despite the relative youth of the project, it has already made more accurate predictions than the US Department of Agriculture has.

The project takes new satellite images and updates them daily, giving the data set a high level of present day accuracy as well as a historical data set. The data has so far been focused on the United States’ extensive corn and soy crops, which is heavily subsidized by the government. Predictions about the success of the corn crop could show farmers who they should sell their corn to and where they should sell it, in addition to keeping an eye on greater economic and environmental trends.

Field of corn in Iowa.
Field of Corn in Iowa. Photo: Peter Van Metre, USGS. Public domain.

The Descartes Lab has utilized multiple satellite systems including Landsat and Sentinel, among others. Every day, nearly 5 terabytes in data is collected.

From Descartes Lab:

This data consists of satellite imagery from multiple NASA, ESA, and commercial constellations. We also collect the latest weather readings and additional agronomically significant signals in real time. To enable data collection and daily processing at this rate, we’ve developed our own hyper-scalable, machine-learning platform. Forecasting at this scale simply cannot be accomplished by sampling individual farms.

In the next year over 100 new Earth-watching satellites are estimated to take to the sky, providing a massive amount of imagery data every day. Descartes Lab isn’t the first to use this data to look at the growth and strengths of cash crops, but they are doing it in an incredibly accurate way that remains unprecedented.

The implications for this kind of data set stretches from the individual farmers growing corn to the US government, and the many points of contact this crop has between the two.

Tim Kelton, one of the co-founders of Descartes Labs, speaks about the technology in this Google Cloud Platform podcast:

More: 2016 Corn & Soy Forecast for United States – Descartes Labs


Photo of author
About the author
Elizabeth Borneman
My name is Elizabeth Borneman and I am a freelance writer, reader, and coffee drinker. I live on a small island in Alaska, which gives me plenty of time to fish, hike, kayak, and be inspired by nature. I enjoy writing about the natural world and find lots of ways to flex my creative muscles on the beach, in the forest, or down at the local coffee shop.

Free weekly newsletter

Fill out your e-mail address to receive our newsletter!