For quant researchers & trading teams
Versioned, point-in-time weather forecasts and observations
With forecast revisions, signed errors, calibrated uncertainty, and complete provenance — packaged for quantitative research. Not just another weather API.
Point-in-time correct
Every forecast row records exactly when it was knowable, so your backtests can avoid look-ahead bias.
Full provenance
Model run timestamps, retrieval timestamps, and source URLs ship with every forecast row, not just the headline prediction.
Calibrated uncertainty
Probability curves computed from strictly-prior-date training errors only — a genuine out-of-sample calibration, documented with its own methodology/version tags.
Rights-aware by design
Every field in the commercial export is audited for redistribution rights before it ships — see the data dictionary for each field's license and status.
What's in the box (v1.0)
- 5 cities: New York, Chicago, Miami, Los Angeles, Austin — one primary NOAA GHCND station each.
- Daily observations back to 1869, through the latest published date.
- Forecasts, errors, revisions, and probability curves from 2021 onward, each tagged with exactly when it was knowable.
- 5 tables:
weather_observations,weather_forecasts,forecast_errors,forecast_revisions,probability_curves— see the live manifest and schema for exact row counts.
No account needed to browse the schema, data dictionary, or a live row preview. Want to try it in your own notebook first? Every table has a free 500-row sample Parquet download, no signup required — see the explorer.