At TellusLabs, we use satellite imagery and other earth data to deliver actionable insights to customers in agriculture, water, forestry, energy, and many other sectors. While some of this imagery has been collected since the 1970s (theLandsat program), only in the last decade has it beenopened for free use in public and commercial domains. This democratization of satellite dataempowers industry, academia, and government sectors, worldwide, with access to earth observation research and insights.
This spaceborne perspective is also beautiful — in this post, we combine some amazing scenes of agriculturally active areas (like this scene below from Storm Lake, Utah) with a little more detail on how this whole process works.
A clear summer day in Jefferson County, Oregon (pop ~21,720). Mount Jefferson (left) is a snow covered stratovolcano in the Cascade Range. The city of Madras (right) features prominent agricultural features despite its rain shadow induced semi-arid climate.
As compelling as these images are, there is a wide gap between pixel values and a useful signal. We bridge that gap by (1) getting satellite and other earth data ‘analytics-ready’ via decades of domain expertise, and (2) connecting those signals to world-class algorithms via a scalable cloud-based computing infrastructure. Our first product,Kernel BETA, is delivering daily agriculture metrics to hundreds of customers from around the world and, as of this posting, is theonly forecast of its kind to cover both US corn and US soybean forecasts on a daily cadence at national, state, and county scales.
Glaciers and meltwater wind and flow through the eastern limits of Lake Clark National Park (est. 1980), just west of Anchorage, Alaska. Infrared sensors allow the clear delineation of vegetation (red) from the surrounding snow and ice (white and blue).
For those unfamiliar with satellite data, one might wonder how Kernel’s precise agricultural forecasts are gleaned from satellite imagery. How can an image, from space, hold so much information pertaining to the intricacies of life on earth? The answer lies, in part, with themultispectral capabilities of satellite sensors when detecting light reflected from the earth’s surface. While the human eye is only able to detect the blue, green, and red wavelengths of theelectromagnetic spectrum, many satellite sensors are sensitive to infrared light and other wavelengths like thermal and microwave (radar). See ourTellusCam post for a sense of what this vision would look like here on earth.
These super-human visual abilities allow satellite signals to track vegetation health, moisture, crop maturity and a host of other vital signs — in many cases on a daily basis. Each wavelength of light that a spaceborne sensor can detect is a separate data attribute representing a reflectance signature unique to the interaction between that type of light and a given land surface. It’s a lot of data to wade through — many terabytes for even coarse resolution sensors. We train machine learning algorithms to translate all of this satellite, weather, and other earth data intoprecise forecasts and metrics that matter for our customers.
Agriculture, cooled lava flows, and rainforest in Maui, Hawaii, USA – true color composite (above), 753 false color composite (above).
All of these handpickedLandsat 8 scenes demonstrate the power of multispectral imagery by revealing, or further defining, what may not be visible to the human eye. The true color variants at the top of each image-pair are created from red, green, and blue reflectances (human-visible light). The false color composite variants at the bottom of each image pair substitute data that is not visible to the human eye (that’s the ‘false’ in ‘false color’) in place of the red, green, or blue data. For today’s images, the team at TellusLabs substituted shortwave infrared, near infrared, and green reflectance signatures for RGB. This is commonly referred to as a ‘753 composite’ because the Landsat data schema refers to these as bands 7, 5, and 3.
NewOrleans, Louisiana, sits at the end of the Mississippi River, along the Gulf of Mexico. Agriculture can be seen along the banks of the Mississippi (in light pink and blue), sediment swirls in Lake Pontchartrain, and the loss of coastal land is apparent to the south and east of New Orleans.
In order to make our pioneering forecasts, we need to know where crops are, which is why we’re pioneering new techniques for crop mapping.