Why a scoreboard
Forecast centres verify their own models carefully, but the results mostly live in technical reports, image charts and login-gated archives. Anyone asking the simple questions ("they said 90°, it hit 86°: was that a bad forecast?", "which forecast should I trust for my town?") has nowhere to look. AI weather models have made the question urgent: new systems arrive every few months, each with a benchmark chosen by its builder.
Forecast Scores grades every system the same way, on the same cases, against the same truth, every day, and publishes the result as standings. Sports have box scores, records and ratings because they make performance legible. Forecasts deserve the same.
What's on the board
Station forecasts vs reality. The National Blend, the NWS, our HRRR-MOS and RRFS-MOS, classic MOS and a dozen raw and AI models, checked every day against the official high, low and rain at about 1,200 U.S. weather stations, from the same morning to a week ahead. Global models. Global models from ECMWF, NOAA, DWD, Environment and Climate Change Canada, the Met Office and Google, physics-based and machine-learned, scored on the large-scale pattern out to ten days. Humans vs Machines vs Money. NWS forecasters, the National Blend, raw models and the prediction market on daily high temperatures at U.S. stations. The Hurricane League. NHC's official forecasts against every hurricane model, on track and intensity.
What's next
Hourly temperatures and the nowcast. Week 3–4, monthly and seasonal outlooks (CPC and the seasonal models) scored against the same station record. Super MOS, our consensus of every station forecast. Seasonal hurricane outlooks graded when the season ends. Forecast Scores is free; the same verification feeds Weather Trader's products.
Independence
Forecast Scores is independent and is not affiliated with or endorsed by any forecast centre, model developer or market. We publish the method in full, flag known advantages (such as a model graded against its own centre's analysis), and we'd rather show an awkward result than a flattering one.
Scores are computed by the Weather Trader verification pipeline and refreshed hourly. See the methodology for exact definitions.