Key Takeaways

  • Wildfire smoke causes roughly 300,000 excess deaths annually worldwide, according to World Health Organization estimates.
  • Fires are easy to suppress when they are room-sized (5x5 meters), but containment costs explode once flames spread across an acre.
  • A constellation of 50 to 80 low-Earth-orbit satellites equipped with mid-wave infrared sensors can scan the entire planet every 15 to 20 minutes.
  • Sensor hardware provides native 50x50 meter resolution, but machine learning super-resolution sharpens the signal 10x to isolate 5x5 meter ignitions.

The 20-Minute Containment Window

Wildfires kill people long before flames reach city limits. Wildfire smoke triggers about 300,000 excess deaths every year worldwide, according to World Health Organization data cited by Google Fellow John Platt.

The core bottleneck in firefighting is not water volume or crew size. It is detection latency.

“For wildfires, a lot of these wildfires you could... if you only caught them early enough, it's very easy to put out a wildfire the size of this room,” Platt explains. “But even if it's like an acre, it gets much, much harder.”

Once a fire covers an acre, radiant heat dries surrounding brush, local wind patterns warp, and suppression shifts from a single truck to multi-agency air drops. To catch fires at the room-scale stage, detection cannot rely on human 911 calls or sporadic lookout towers. It requires orbital surveillance with refresh cycles measured in minutes.

Solving the Physics Problem with Mid-Wave Infrared

Standard visual cameras struggle with fires because smoke plumes obscure the ground and sunlight creates blinding surface glare. Platt points to orbital physics as the clean solution: mid-wave infrared sensors.

“If you had a global constellation of low Earth orbit satellites that could detect in the midwave IR, that goes back to the black body essentially that's the temperature of fire,” Platt notes. “The fires stand out in the midwave IR.”

Hot combustion generates distinct black-body radiation peaks in the mid-wave spectrum. In that band, a small flare glows brightly against the cool Earth background.

Building a global safety net requires scale. Platt calculates that deploying “roughly 50 to 80 of them” in low Earth orbit allows constant coverage. With that orbital geometry, the constellation scans any coordinates on the globe “within like 15 to 20 minutes.”

Replacing Heavy Glass with Machine Learning

The hard engineering tradeoff in satellite design is weight versus resolution. Flying optical lenses sharp enough to capture a 5-meter square from 500 kilometers up requires massive, expensive satellites. A 50-satellite constellation built with traditional space optics would cost billions.

Platt and the Earth Fire Alliance bypassed this hardware limitation by shifting the burden to software.

The physical sensors produce raw data at 50x50 meter resolution. That is far too coarse to spot an early ignition. Google applies super-resolution models on top of the raw multispectral telemetry to reconstruct the scene down to 5x5 meters.

“The resolution is about 50 by 50 meters, but you can use super resolution because it's essentially multispectral and you sort of know where fires are,” Platt says. “There's a fair sprinkling of AI on top of them to reach that 5x5 meter.”

Instead of launching heavier glass into space, the system uses algorithmic post-processing on orbit and ground stations. Machine learning extracts a 10x resolution gain directly from the physics of multispectral thermal emission.

What to Do With This

Audit your core product stack for expensive hardware or heavy compute workarounds. Identify one sensor, server, or data pipeline where you can swap expensive hardware capture for algorithmic super-resolution or model priors this quarter.