Accuracy
Forecast accuracy, in the open.
Frothed publishes its forecast accuracy out loud. Every night a job pulls the previous day's model output for wave height and primary swell period at each region, joins it against NDBC and CDIP buoy observations at the same hour, and records mean absolute error at day one, day three, and day seven horizons.
The methodology is documented and reproducible. The scoring window is rolling thirty days. The baseline benchmark is the NOAA WaveWatch III global run with NWPS nearshore enhancement where coverage exists, blended against CDIP directional spectra in California. Bad weeks and good weeks both ship.
Baseline scoring begins at open beta. Daily updates start the day the public beta opens. Numbers below render from the nightly run once live.
Day-three wave-height error, by region
| Region | Day-3 MAE, wave height | Day-3 MAE, primary period |
|---|---|---|
| Southern California | pending | pending |
| Central California | pending | pending |
| Northern California | pending | pending |
| Hawaii | pending | pending |
| Outer Banks, NC | pending | pending |
| New York and New Jersey | pending | pending |
| Pacific Northwest | pending | pending |
Methodology
- Models: NOAA WaveWatch III global run, NOAA NWPS nearshore where available, HRRR for surface wind, CDIP for buoys with directional spectra.
- Truth: NDBC buoy hourly observations and CDIP directional spectra at the buoys closest to the spot polygon.
- Scoring window: rolling 30 days, recomputed nightly.
- Horizons: day-1, day-3, day-7 forecast lead time.
- Metrics: mean absolute error for wave height in feet and primary period in seconds.