Skip to content

Agriculture · 2026 · Self-built prototype

Frostwatch: micro-climate frost risk, block by block

A prototype that scores frost risk at the orchard-block level from terrain, cold-air drainage, and weather, as the template for any climate advisory that has to be right about a specific patch of ground rather than a region.

Sector
Agriculture
Year
2026
Engagement
Self-built prototype
Team
Solo build, with Claude Code

A regional frost forecast is nearly useless to a grower, because frost pools. Cold air drains downhill and settles, and the block at the bottom of the slope loses fruit while the block up the hill is fine. Frostwatch is a proof that you can score that difference before the night it matters.

What we built

A PostGIS model that scores frost risk per block from terrain, cold-air drainage paths, and weather, rather than from a single regional number. The output is a map a grower can actually act on: which blocks to protect first when a cold night is coming.

What we found

The spatial reasoning is the whole product. Getting the micro-climate right, where the cold settles and how fast, is what separates a useful advisory from a weather app a grower already ignores.

What we learned

This one is honest about its own calendar. Frost risk concentrates into a few months; the rest of the year the same system is a pleasant vegetation dashboard nobody urgently needs. We built it to prove the micro-climate method, and we would only stand up a production version for a grower who has the frost problem badly enough to pay for it in season. Until a paying pilot says otherwise, it stays a proof, not a product.

Outcomes

Block-level
risk resolved below the regional forecast
Terrain-aware
cold-air drainage and slope, not just temperature
1 template
the method generalizes to other ag-climate advisories