Wildfires are one of nature’s most destructive forces, exacerbated by the effects of climate change and land development. The recent fires in Sonoma County killed 22 people and destroyed thousands of homes, and the Ventura and Santa Barbara County wildfires burned for more than 281,800 acres before the flames were finally extinguished.

Early warnings are key to minimizing loss of life and property, which is why Descartes Labs, a geospatial analytics startup headquartered in Santa Fe, New Mexico, is today launching the Wildfire Watch system. It’s a Twitter bot (@WildfireSignal) that updates every six hours with two time-lapse videos — one in full color and one in infrared — that track the progression of wildfires across the continental U.S., highlighting their smoke plumes and heat signatures in bright red and white pixels.

Each tweet also contains the name of the fire, its latitude and longitude, and a unique hashtag.

“You can see the smoke obscure the landscape,” Caitlin Kontgis, lead applied scientist working on the Wildfire Watch system, told VentureBeat in a phone interview, “and watch the heat signature develop and evolve over time.”

Descartes Labs

Above: A graphic of the Box Car Fire generated by Wildfire Watch.

Image Credit: Descartes Labs

So how does it work? The system scrapes Inciweb, a government interagency incident management platform, for a list of active fires verified by the U.S. Forest Service. It then sources imagery from GOES-16 — a geostationary satellite that snaps pictures of the entire Western Hemisphere every five minutes — and from two other satellites, Landsat 8 and Sentinel 2. (The time-lapse frames come from GOES-16; Landsat 8 and Sentinel 2 are used to confirm the “burn scar,” or scorched earth, left in the fires’ wakes.)

The updates could help evacuees in the path of a wildfire avoid smokey areas, Kontgis said. And because it tweets constantly, even at night when it’s harder to see smoke plumes with the naked eye, the system could provide more warning time than other systems for those affected.

Alongside the launch of the Twitter bot, Descartes is working to get a version of the system in the hands of first responders. The company has built a dedicated feed for the Santa Fe National Forest and is developing a computer vision platform that can detect fires in satellite imagery before they’re flagged by the Forest Service. This is in the final development stages, with a tentative fall 2018 launch window.

“We’d like it to remain a public service,” Kontgis said.

The Wildfire Watch system is part of Descartes Labs’ ongoing outreach effort. In addition to datasets that it makes freely available to the public, the company has published tools like GeoVisual Search, which uses a 50-layer neural network and satellite imagery from NAIP Arial Imagery, Planetscope, and Landsat 8 to highlight wind turbines, solar farms, football fields, and other structures on a country-wide scale.

On the commercial side of the house, Descartes Labs’ bread and butter is data analysis. The company’s in-house Python-based platform ingests and processes about four petabytes of imagery from NASA, commercial vendors, and public sources like OpenStreetMap. (It once spun up 30,000 CPU cores in Google Cloud to process a petabyte of satellite imagery in 16 hours.)

The startup leverages that data to help analytics-driven clients make decisions and generate predictive models. One such customer is the U.S.’s Defense Advanced Research Projects Agency (DARPA), which is investigating food security in the Middle East and North Africa.

Descartes is also in talks with insurance providers, though it won’t name names.

The company, which spun out of Los Alamos National Laboratory, got its start in agriculture. In 2015, using a model trained on satellite data and spectral information like chlorophyll levels, it began releasing state- and country-wide U.S. corn yield estimates. Descartes founder Mark Johnson claims that the daily reports regularly beat the U.S. Department of Agriculture’s own by a few percentage points.

Descartes Labs raised $30 million in August 2017 in a funding round led by March Capital. Additional investors include Crosslink Capital and Cultivian Sandbox.