5 Q’s for Mark Johnson, Founder of Descartes Labs
The Center for Data Innovation spoke with Mark Johnson, founder of Descartes Labs, a crop forecasting startup in Los Alamos, New Mexico. Johnson discussed how his company beats official government sources at predicting crop yields and how his company was able to commercialize software developed with federal research and development funding.
Joshua New: Descartes Labs forecasts crop production around the world, even better than government agencies. Before we talk about how you do it better, can you explain why forecasting data is valuable, and why it is valuable to produce more timely and accurate data?
Mark Johnson: In the case of crop production, there are all sorts of reasons businesses, governments, and farmers want to forecast the future. Insurers of farms want to better understand the risk in their portfolio when a storm affects their farms. Companies who move around grain want to figure out how to optimize their logistics network. Farmers want to know how they’re doing with respect to their neighbors to maximize profit when they sell their grain. Traders want to know how much crop is being grown to predict price. And governments want to know when there might be a major food shortage to deploy humanitarian resources before a major famine.
New: So, how does Descartes Labs do this better than anyone else?
Johnson: We have a lot of respect for what our friends in the U.S. Department of Agriculture (USDA) are doing. The United States does better than any other country at forecasting crop production. However, the techniques they are using have inherent limitations. USDA builds a report from field visits and surveys to farmers. Not only are both very costly, but you can only mail out so many surveys and visit so many farms each month. That means you are limited in the amount of data you can collect.
Descartes Labs believes that software can make forecasts faster, more frequently, and with greater accuracy than traditional methods. We process a massive amount of satellite and weather data and use artificial intelligence to interpret that data and make forecasts. We get a daily picture of all 3 million square kilometers in the corn belt every day, which means we get a sample of every farm every day, instead of tens of thousands of farms. Our edge is being able to use that data effectively to get crop predictions with less than 1 percent error.
New: The software behind Descartes Labs is based on technology developed at the Los Alamos National Laboratory, which is run by the U.S. Department of Energy. How do you go about commercializing government research?
Johnson: National Labs actually have an excellent “tech transfer” program at each lab. Since our tax dollars go into this research, the labs are encouraged to license technology when it’s mature enough and not a threat to national interests.
In our case, we worked with Los Alamos National Laboratory’s tech transfer group, the Richard P. Feynman Center for Innovation. We were able to get a license to the technology that our founding scientists had been working on for seven years with $15 million of federal research and development. This technology was a series of patents around deep learning artificial intelligence, which seeks to model the brain in a computer. I think it’s really cool that our national labs still work on blue sky technology like this.
New: How does Descartes Labs get all of its satellite imagery? Is it from public or private sources, or both?
Johnson: Descartes Labs uses both public and private data sources. Our 2016 Corn Production Forecast uses primarily public data from NASA constellations, specifically MODIS and Landsat. We also pull in all of the European Space Agency’s brand-new Sentinel constellation. Through our partnership with satellite imaging company Planet Labs, we have access to their RapidEye constellation and are very excited about their cubesat constellation, scheduled to launch later this year.
New: Theoretically, if the entire agricultural sector were use your software to make much more informed decisions, what would the larger implications be? Lower food prices overall? Less hunger?
Johnson: Already, technology has greatly improved global food production. In the United States before World War II, corn was produced at around 35 bushels per acre; today we’re forecasting a corn crop yielding just over 170 bushels per acre. There are lots of companies focusing on continuing to increase yield. What we do is make that yield more efficient once the corn gets into the market. We hope that lowers the price of food, maximizes the profit of farmers, allows grain traders to operate more efficiently, and ultimately means we can deploy our precious food resources more efficiently to the people who need them.
No one has tried to build artificial intelligence at planetary scale—Descartes Labs believes we’ll be able to better understand our impact on Earth and subsequently make smarter decisions to save our planet.