Frostwise — know your frost dates, plant with confidence

Description and use

Frostwise leverages TabPFN and ERA5 climate data to compute precise frost dates and generate planting calendars, enabling gardeners to plan confidently without risking crop loss.

Problem

Gardeners lack accurate, location‑specific frost dates, leading to crop loss.

Outcome

Frostwise delivers accurate, hyperlocal frost dates and planting plans without proprietary APIs, enabling gardeners to plant confidently.

Workflow

• 1. User enters city or selects sample location.
• 2. Backend fetches 2005–2024 daily minimum temperatures from Open‑Meteo ERA5 archive for that location.
• 3. TabPFN performs an in‑context prediction: a classifier decides if frost occurs, two regressors output last spring and first autumn frost day‑of‑year.
• 4. The app generates a planting calendar for 12 common crops, anchoring sow windows to the predicted frost dates and adding crop notes.
• 5. Result is displayed in a clean single‑page UI; static demo precomputes predictions for offline use.

Evidence

Frostwise — a hyperlocal garden planner that computes your frost dates from real climate history using TabPFN, the open‑source tabular foundation model — then turns them into a planting plan you can actually follow.

Source

At request time, TabPFN performs the whole prediction in‑context: a classifier for frost/no‑frost, two regressors for the dates — a single forward pass per model, no gradient training.

Source

The real challenge was making the AI genuine instead of decorative. A frost date isn't in any API — it has to be derived, so I built the entire label pipeline from raw reanalysis data...

Source