It combines a physical world model with terrain-adjusted weather and a snow metamorphosis engine to predict for every point on the mountain the surface state of the snowpack, across time. Think of a simulated snow globe, where we've accounted for sun, wind, temperatures, precipitation, and more.
It's been fun working on non-LLM-"AI" (whatever that means nowadays) - just good ol' GIS, meteorology, and physics. The model is cool - while I have pretty good spatial reasoning abilities, I suppose it's not unsuspected that running this rigorous computation returns results that I would not have otherwise thought about.
Hope it's an interesting read, and would love any thoughts and feedback.