A Living Map of the World's Food Supply

    December 13, 2018

    By Geoffrey von Maltzahn

    Today, we’re excited to announce Indigo’s acquisition of TellusLabs, a satellite imaging and artificial intelligence company based out of Somerville, Massachusetts. This joining of their mission with ours has been over a year-and-a-half in the making, and is a product of our Indigo Research Partners’ engagement with innovative startups around the world. Here’s the story of how this came about, what we’ve been working on together, and how our work will offer powerful tools to Indigo growers in the years to come.

    Like many exciting efforts at Indigo, this one started with a wild idea. In the summer of 2017, our Innovation Team asked: What if we could learn as much about the health and potential yield of agricultural fields with the help of satellites as we can by walking through them by row?

    Our point of inspiration came from a completely different field, specifically Anne Carpenter’s work at the Broad Institute of MIT and Harvard. She and her team have been using machine learning to generate remarkably deep insights from a seemingly simple source: videos of microscopic cells in culture. For over a century, scientists have used their own eyes and memory to analyze what they see through a microscope to discover the causes and cures for disease. Anne’s work takes a fundamentally different approach; rather than rely on the skills of a particular scientist to understand the view from under a microscope, her team applies cutting-edge machine learning tools that can recognize patterns across thousands of videos at a scale far surpassing capabilities of the human brain.

    Our Innovation Team asked if we might apply this lesson in an agricultural context. Specifically, we wondered whether spaceborne cameras capturing videos of the world’s agricultural fields while orbiting at 17,000 mph could be combined with modern machine learning tools to provide understandings just as useful and profound. Many necessary prerequisites for this hypothesis were already well established. For one, the instruments aboard satellites orbiting our planet are capable of viewing the millions of farms constituting the global food system. Images taken with this state-of-the-art equipment detect wavelengths of light both within and outside of the visible spectrum, allowing these cameras to “see” into crop canopies. Many of these images, thanks to the resources of NASA and the European Space Union, are made available in massive public repositories.

    Last year we launched Indigo Research Partners (IRP), an agricultural “lab” designed specifically to test ambitious hypotheses like this. In this program we work with nearly 100 growers around the country to test agricultural technologies — Indigo’s and others — on tens of thousands of acres. Over the past year in IRP, we have evaluated hundreds of companies and tested dozens in areas as diverse as grain quality and storage sensors, soil moisture probes and weather stations, drones and autonomous vehicles. To test this particular hypothesis, we set out to find the world’s experts and leaders in applying machine learning approaches to satellite imaging. After compiling a holistic assessment of the all of the companies in the field, TellusLabs stood out as the clear leader.   

    Their team had already completed ambitious work, including fielding a global agricultural intelligence platform, Kernel, that provides daily monitoring of nearly half of the human calories. Kernel is also a forecasting system; for the past three seasons, Tellus forecasted US corn and soy yields throughout the growing season. The implications for a model that could do this with accuracy, speed, and breadth would be profound. The United States Department of Agriculture (USDA) shocked the industry in October of 2017 when it made a record-setting upward revision to the harvest very late in the season; TellusLabs had been correctly forecasting the corn yield since August. This success highlighted TellusLabs’ belief and trust in the platform, not to mention the prospect that their technology could help determine county- and country- level estimates for a variety of crops.

    Collaborating closely with TellusLab, we kicked off a series of projects to push the technology’s boundaries, moving from country- and county-level estimates to those for individual agricultural fields. By combining TellusLabs’ satellite imagery and machine learning systems and Indigo’s ground-truth data — collected through Indigo Research Partners’ and commercial growers’ acres — we began training algorithms for predicting field boundaries, crop types, and field-level yield estimates.

    Over the past year, the progress we’ve made has surprised all of us. This work paved the way for recent conclusions around a 12.7% median yield increase for fields growing Indigo Wheat™ compared with neighboring fields that were not. Since then, we’ve expanded such analysis to more than half-a-million acres to compare the performance of other Indigo microbial seeds growers with neighboring fields.

    The promise of this work includes the treatment of all fields, across all geographies, with the delivery and selection of agricultural inputs and practices, from seed variety to fertilizer application, tillage to irrigation, that best fits those areas. Growers face the complex challenge of managing their fields on an acre by acre basis. With the tools we are building, we can predict critical determinants of farm profitability across all acres of farmland simultaneously. These insights – delivered in a streamlined form from Indigo – will allow growers to strategize for their most lucrative growing seasons.

    At Indigo, our mission — harnessing nature to help farmers sustainably feed the planet — is what drives our best work. With TellusLabs, we saw a similar devotion; their mission is building a living map of the world’s food supply. More and more, we found opportunities to advance both of our central objectives by combining our talent and technology. As a result, we decided to join forces to work on future projects, ones with the potential to inform decisions on farms around the world.

    Our completed work together only scratches the surface of what this new, innovative technology can mean for all participants in the agricultural industry. With a vision for a food system that is transparent, efficient, and better for farmers, consumers and the environment, we will continue solving the meaningful problems affecting our world today, for the betterment of present and future generations.