New AI Study: Severe Wildfires Are Changing California's Snowpack
The state's biggest water source may react to fires in surprising ways.
California's Sierra Nevada snowpack is the state's biggest natural reservoir — when it melts, it supplies roughly a third of the water used by farms, cities, and homes. In 2020, the Creek Fire tore through the southern Sierra, and researchers wanted to know something practical: did the burn change how long snow sticks around each winter? To find out, they tracked snow on the ground for eleven winters, from 2016 through 2026, using satellite images that spot white snow from space, combined with weather records and an AI model.
The clever part was the comparison. The team paired 3,778 burned spots with unburned spots that were otherwise nearly identical — same elevation, same slope, same pre-fire snow habits — like matching twins for a fair test. Across the whole burned area, the average change was essentially zero. But in the most severely burned patches, snow lasted about 2.6 percentage points more often than in the matched unburned spots. One likely reason: when trees burn away, less snow gets caught in branches and shaded, so more of it lands on the ground where satellites can see it. And that's the honest catch — the cameras only count snow under clear skies, so some of the 'extra' snow may have been there all along, just hidden under a canopy.
The study's bigger message is about AI itself. Their model (called XGBoost) predicted raw snow levels very well — roughly 81% accurate — but it was close to useless at predicting how much the fire itself changed things. Elevation and temperature alone explained about two-thirds of its decisions. In plain terms: a tool that draws an accurate map isn't the same as a tool that proves cause and effect. That distinction matters everywhere AI is used to make decisions.
The researchers shared their code and data openly, so water managers, fire planners, and ski towns can reuse the method. Expect more studies like this as the West grapples with hotter fires, thinner snowpacks, and a water supply that depends on both.
- In the most severely burned parts of the 2020 Creek Fire, snow lingered about 2.6 percentage points more often than in similar unburned areas.
- Across the entire burned landscape, the average effect was basically zero — the dramatic change showed up only in the worst-burned patches.
- The AI predicted snow levels well (about 81% accurate) but was nearly useless at predicting the fire's effect, showing accurate maps aren't proof of cause.
Why It Matters
Sierra snowmelt is a third of California's water — how fires change it affects farms, cities, and drought planning.