How Accurate Are Surf Forecasts?
NOAA's WaveWatch III model is accurate within ±1-2 feet for wave height and ±2-3 seconds for period, 0-3 days out. Accuracy degrades beyond 5 days. Quiver's per-beach ML corrections — trained on 30,000+ buoy observations from CDIP and NDBC — achieve a 90.4% directional match rate and 59% improvement over raw NOAA output.
3 min read
Key Takeaways
- Surf forecasts are ±1-2 feet accurate for 0-3 days. Quiver's ML corrections improve raw NOAA output by 59% at specific beaches.
- Global models miss local effects. Quiver's ML learns each beach's systematic biases and corrects for them — see live metrics at /forecast-accuracy.
- Plan sessions from the 3-day window. Always cross-check the forecast against buoy data before paddling out.
The Short Answer
NOAA's WaveWatch III global wave model is accurate within ±1-2 feet for significant wave height and ±2-3 seconds for period, 0-3 days out. Beyond 5 days, accuracy drops to ±3+ feet. Quiver's per-beach XGBoost ML corrections — trained on 30,000+ buoy observations from CDIP and NDBC — achieve a 90.4% directional match rate and 59% improvement over raw NOAA baseline. No forecast is perfect; buoys are always the ground truth.
Where Forecasts Are Accurate — and Where They Fail
WaveWatch III runs at 0.5° resolution (~30 miles per grid cell). It captures open-ocean swells well but misses the "last mile" — local bathymetry (underwater canyons, reefs, sandbars), coastal wind patterns, and swell refraction around headlands. That's why buoy readings at La Jolla's Scripps Pier can differ from WaveWatch III by 1-3 feet: the Scripps Canyon focuses and amplifies certain swell angles in ways the global model can't resolve.
Quiver's ML layer fixes this. Per-beach XGBoost models learn the systematic bias at each break — when the model consistently over-predicts at your spot, ML corrects it. This is evaluated against holdout data (data the model never saw during training) to prevent overfitting. The public accuracy dashboard at quiversurf.app/forecast-accuracy shows real-time metrics by region and beach.
“Global models miss local effects. Quiver's ML learns each beach's systematic biases and corrects for them — see live metrics at /forecast-accuracy.”
What This Means for Your Session
Trust the 0-3 day forecast window for planning sessions. Use the 5-7 day view to track incoming swell systems, but don't commit to plans based on it — too much can change. Always cross-check the forecast against real-time buoy data 1-3 hours before your session. If the nearest NDBC buoy shows 3 feet but the forecast says 5 feet, the swell hasn't arrived yet or the model over-predicted. Build local knowledge: after 20+ sessions comparing Quiver's forecast to what you actually see, you'll develop an intuitive calibration for your break that no model can replace.
Frequently Asked Questions
Why are some forecasts more accurate than others?+
Accuracy depends on data density. Beaches near CDIP or NDBC buoys get more ML training data and better corrections. Remote breaks far from buoys rely more on the global model, which misses local effects. Quiver's accuracy dashboard shows per-beach metrics.
How does Quiver's ML correction work?+
Per-beach XGBoost models trained on 30,000+ buoy observations learn systematic forecast errors at each spot — when the model over-predicts height or misreads direction due to local bathymetry. The correction layer runs every 3 hours with fresh buoy data.
Is Quiver more accurate than Surfline?+
Quiver publishes its accuracy metrics publicly at /forecast-accuracy. Surfline does not publish comparable metrics. Quiver's ML achieves a 90.4% match rate vs. observed buoy readings and 59% improvement over raw NOAA — judge for yourself.
Should I trust a 10-day forecast?+
No. Use it to spot incoming swell systems and general trends. Weather chaos makes anything beyond 7 days unreliable. By 3 days out, the forecast is much more accurate — check again then before committing.
