Descartes Labs: Predicting Farmer’s Fortunes from Space

A Los Alamos-based startup is disrupting the market for the oldest industry on earth: agriculture.

There’s arguably no more important industry to humanity than agriculture: it feeds us, employs one-third of the global workforce [1], and contributes 3 trillion dollars to global GDP [2]. Given the prominence, significant effort and money are invested in accurately predicting seasonal crop output. In the United States, traders, insurers, lenders, and many others utilize crop forecasts to support the agricultural financial ecosystem. Until recently, these players relied almost entirely on monthly production reports issued by the US Department of Agriculture (USDA) to glean information about future supply and demand. The USDA compiled this data by surveying samples of farmers around the country and using statistical techniques to extrapolate domestic-level estimates [3]. While these estimates were decent (usually less than 5% error) [4], unexpected weather events or other anomalies between monthly reports reduce accuracy and provide inaccurate signals to the market.

Enter Descartes Labs

Descartes Labs processes satellite imagery of US farmland (3.1 million square miles) [5] using neural networks and cloud computing horsepower to provide free crop yield forecasts on a weekly basis. This methodology has proven to be more accurate than the USDA survey and provides real-time price signals that improve overall market efficiency. In fact, the Descartes Lab forecast is so accurate (1% error relative to actual production) it is effectively measuring the weight of every corn kernel in America to an error of plus or minus 3 milligrams [4]. Descartes’ first harvest was in 2015 when it offered free state-by-state and country-wide corn yield forecasts on a weekly basis. In 2016, it expanded the forecast to include soy production and offered a further county-by-county breakdown of yield [5].

Disruptive Business Model

While Descartes can promise customers accuracy, its business model seems unlikely to be profitable at first glance; releasing a free weekly report has little top-line value. However, this free morsel is only to entice customers to pay for additional premium services. Currently, paying subscribers can have access to crop forecasts updated every two days enabling speculators to take positions based on data not yet publicly available. More notably, Descartes offers customizable forecast solutions that allow paying customers to apply proprietary machine-learning algorithms and staff expertise to their own big data questions (see below for detail) [6].


Operational Enablers

The operational backbone that supports Descartes’ customer promise relies on the combination of three emerging digital technologies:

  1. Artificial neural networks [7]
  2. Cheap cloud-based processing horsepower [8]
  3. Proliferation of small, cheap, earth-observation satellites [9]

Descartes utilizes these technologies in concert to produce superior agricultural forecasts. They have also developed a standardized forecast development process (shown below) with the potential to be applied broadly to other commercial applications.

Descartes Labs Forecast Methodology [6]


If this technology can be used to monitor changes in crop yields, what is preventing it from being retooled to assess construction activity in China or human migration patterns out of Syria? The white space for earth-observation AI applications is massive and as costs for satellite images and cloud computing space fall, this operational model is extremely scalable given the low marginal cost of new forecasting applications.

The Future of Earth Observation

Despite the seemingly boundless potential for this technology, Descartes will face several key challenges in the future. First, the frequently refreshed satellite imagery that their models rely on are controlled by only a few companies like DigitalGlobe and Planet [10]. What is stopping them from using their monopoly power to squeeze Descartes’ margins? Furthermore, what is preventing these companies or another capable start-up from simply replicating what Descartes has done? To get ahead of these challenges, I recommend that Descartes consider the following recommendations to expand capabilities and protect market position:

  1. Go Global – today’s markets are global and what happens to the wheat harvest in Russia can drastically impact US markets (see Great Grain Robbery [11]). Furthermore, while less accurate the USDA forecast provides adequate transparency into US markets – Descartes can make a bigger impact by expanding to more opaque regions.
  2. Go Upstream – farming inputs like water can have a major impact on crop yields. Descartes should factor in mountain snow pack and reservoir levels into its yield forecast.
  3. Democratize – Descartes should open their API to a larger audience free of charge. Gaining users and crowdsourcing their capabilities will generate amazing applications for this technology beyond the reach of a resource-limited Descartes team that can still be monetized with creative revenue generation mechanisms.

The Greatest Descartes?

Just as Rene Descartes redefined the way we look at and think about the world, Descartes Labs is using digital technology to revolutionize the global marketplace for goods and ideas. In a world with more transparency and efficient markets, we’re all better off.

784 words



1)      “Labour.”

2)      “Agriculture, Value Added (% of GDP).” Agriculture, Value Added (% of GDP) | Data,

3)      “Understanding USDA Crop Forecasts.” doi:10.15417/1881.

4)      “Descartes Labs Raises 5M to Make Agricultural Predictions with Deep Learning.” Venture Beat,

5)      “Seeing Corn with Satellites.” The Atlantic, Atlantic Media Company,

6)      “Descartes Labs: Solutions.” Descartes Labs,

7)      “How Imaging Technologies Are Changing the World.” Forbes, Forbes Magazine,

8)      Lynley, Matthew. “Deep-Learning, Image-Analysis Startup Descartes Labs Raises $3.3M After Spinning Out Of Los Alamos National Lab.” TechCrunch, 1 May 2015,

9)      Wieczner, Jen. “Traders’ New Edge: Satellite Data.” Fortune, 22 Dec. 2015,

10)  “A Sudden Light.” The Economist, The Economist Newspaper, 10 Aug. 2016,

11)  “The Great Grain Robbery: Lessons Learned from Earth Imaging’s Early History.” Earth Imaging Journal Remote Sensing Satellite Images Satellite Imagery,


Omada Health: A Leader in the Digital Health Revolution


root9B’s Specialized Approach to Cybersecurity

6 thoughts on “Descartes Labs: Predicting Farmer’s Fortunes from Space

  1. Very interesting company. Building on the final recommendation you give Descartes to strengthen its competitive position, I think the democratization of the company’s intelligence would help drug enforcing agencies throughout the world, especially those operating in areas where terrestrial patrolling is complicated. As pointed out by the World Drug Report, there are some agencies already using satellites to look for drug plantations; nevertheless, the current model requires those agencies to purchase a library of satellite images of areas typically associated with growing drugs, and then they proceed to examine them. (1) The new technology could enable these agencies to take a more divergent approach, taking extensive segments of land looking for areas that resemble drug plantations, instead of focusing in the same areas over and over. By casting their net wide, they could significantly increase the number of plantations detected and, more importantly, they could signal drug organizations that their vigilance is no longer constrained to “usual suspect” areas.

    (1) United Nations Office on Drugs and Crime, “World Drug Report 2010,”, accessed November 18, 2016

  2. Great post! It is fascinating to see the high level of accuracy of the analysis of these detailed satellite images. In the climate change discussion earlier this month, we saw many posts of how changes in temperature and precipitation patterns are leading to significant agricultural challenges (e.g., cacao trees) and even altering entire habitats. I wonder if this technology could also be used for the detailed tracking and prediction of changes in plant productivity in unique, at-risk ecosystems like the Amazon rainforest.

  3. Thanks for the interesting article! I agree that to most benefit the world, Descartes should continue the democratization of its insights as it has done so to date. It is interesting to wonder, though, what Descartes should do if its goal is to maximize profits. For the past few decades, USDA reports have moved billions of dollars of trades on the Chicago Mercantile Exchange due to the price sensitivity this information has on agricultural commodities (for a funny picture of what this looks like in orange juice, see this scene from Eddie Murphy’s 1983 film Trading Places – ). If Descartes has more accurate information than the USDA, there would be many commodity-focused hedge funds that would buy Descartes, make its information proprietary, then trade on these insights (both on physical and financial markets). Through leverage and use of derivatives products, there are billions of dollars of profits to be made (in fact, I’m certain there is a hedge-fund-meets-Rocket-Internet play here to copy Descartes technology for other markets outside of corn, like dairy or soy).

    The two paths of action aren’t mutually exclusive, of course. Descartes could make its information freely available, only after taking positions in commodity markets, then use its profits to reinvest in growth in future frontiers in earth observation.

  4. Great post! As you mention in your article, why is this technology not being applied to other industries and regions of the world? Besides all of the benefits you mention, the use of earth-observation artificial intelligence could be extremely useful to both governments and companies in emerging economies because large portions of the economy are informal and don’t get reported, which makes the available official information even less reliable. Having precise forecasts of agricultural yields, construction rates or traffic between specific cities would allow governments to better manage their investments in infrastructure and entire industries to better manage their operations.

  5. Really enjoyed this article. I know the assignment’s purpose wasn’t to write about the technology, but I would love to see a plot of “data acquisition cost” vs time. From this perspective, and to some of the previous comments, my guess is that the price is still too high for other uses. Although they use CubeSats instead of more traditional satellites, the costs are still quite high to launch things into space (~ $1.7M for 50kg payload [1]). This price, however, will continue to decrease.

    I liked your recommendations for how Descartes can improve. One additional thought would be to put satellites into higher orbits like middle-earth-orbit or Geo-stationary. Gaining valuable intelligence on items like Syrian Refugee crisis or drug movements may need a “persistent stare” versus a satellite that passes infrequently for a short period (as what low-earth-orbit satellites provide). Revisiting each location every few hours may not be enough to actually track the shipment of goods or people. But, this is a rapidly growing field with many exciting opportunities ahead.


  6. Interesting post about a company I had not previously heard of, Rocky. Given SpaceX’s and Blue Origin’s recent successes in landing reusable rockets, I think it is likely that the cost per payload for Descartes cubesats will decrease significantly over the next several years. This will lower their variable cost per payload, and if they are willing and able to invest the capital it seems likely that they could expand their agricultural intelligence offerings substantially (why not have a goal of monitoring every possible commodity?). I also agree with Carl that it would be nice to put Descartes satellites into geostationary orbits; however, the cost of doing this is significantly higher than putting a cubesat into low-earth orbit (LEO) and would require Descartes to launch thousands of satellites. And so, constant, real-time monitoring of specific fields may not be cost-effective. Instead, launching dozens of cubesats into sun-synchronous LEOs, which are “often used for remote-sensing
    missions because they pass over nearly every point on Earth’s surface” [1], seems to be a more cost efficient, and even long term solution that could propel Descartes into market domination for real-time commodity industry monitoring.


Leave a comment