Building a Hyperlocal Rain Nowcaster on a $0 Stack: Benchmarking Open-Meteo Against Airport Radar

Описание и применение

Создано приложение для прогнозирования дождя в Мумбаи, использующее локальные данные METAR и Open‑Meteo, обучающее логистическую регрессию и выдающее прогнозы в браузере.

Задача

Providing a next‑two‑hour rain forecast for Mumbai streets requires more local evidence than a global model alone can provide.

Результат

The system achieves a Brier skill above the raw rain call and climatology, demonstrating that a lightweight logistic correction trained on local METAR labels can improve short‑term rain forecasts for Mumbai.

Рабочий процесс

• Collect hourly forecasts and METAR reports via GitHub Actions scheduled at minute 5
• Store forecasts in data/log.csv and label hours with rain/drizzle from METAR
• Filter forward forecasts and train logistic regression daily at 01:30 UTC
• Evaluate candidate model with Brier score and walk‑forward folds
• Promote model if it meets holdout and skill criteria, export model.json to CDN
• Serve model in browser to compute hourly rain calls and next‑two‑hour verdict

Доказательства

I collect forecasts, compare them with observations at Mumbai airport, and train a small correction that visitors run in their browsers.

Источник

I do not ingest airport radar. I use METAR surface-weather reports from station VABB, Chhatrapati Shivaji Maharaj International Airport.

Источник

I fit six features in this order: best_match_mm, ecmwf_mm, relative_humidity, sin(2π hour / 24), cos(2π hour / 24), recent_rain_mm.

Источник