How Accurate Is Our Surf Forecast? The September 2026 Numbers

We publish surf forecasts every day. Twice a week we check them against what the ocean actually did, and this is that check for the week ending September 14, 2026. The numbers below are not flattering, and we are publishing them anyway — an accuracy figure you only see when it looks good is not an accuracy figure, it is marketing.

What we measure, exactly

Our verification compares forecast open-ocean swell height against observed open-ocean swell height. That is a specific claim and it is narrower than it sounds, so it is worth being precise about what it is not:

• It is not a measure of breaking wave height at the sand. Beach-level face height depends on bathymetry, tide, local wind and swell direction, and it is a harder problem than the open-ocean number.
• It is not a timing or direction score. It is height only.
• It is a rolling 7-day window, recomputed continuously, so it moves.

We hold ourselves to the open-ocean number because it is the part of the chain we can verify cleanly against buoy and model observation. The step from there to the sand is where our remaining error lives, and we say so rather than quietly folding it into a friendlier headline.

The current reading

As of September 14, 2026, over 205 verified forecasts in the previous seven days:

• Average miss: 1.67 ft
• 67% of height calls landed within 2 ft

Which means roughly one call in three missed by more than 2 ft in the last week. That is the honest version of the sentence.

The direction of travel, which matters more than the number

A single accuracy figure with no trend attached is technically true and materially misleading, so here is the series. Through late August our weekly reads sat in the 1.0–1.3 ft range. Then:

• Sept 7: 1.55 ft average miss, 75% within 2 ft
• Sept 10: 1.97 ft average miss, 62% within 2 ft
• Sept 14: 1.67 ft average miss, 67% within 2 ft

Five consecutive reads got worse. This one got better. We are not calling that a recovery. One read down after five up is a single data point, and the same discipline that made us publish the slide stops us from declaring it over. We will say the run has turned when two consecutive reads say so, and not before.

The sample size has stayed steady at roughly 200 verifications per read, and the metric definition has not changed. So this is movement in the forecasts themselves, not an artifact of counting different things week to week.

What we think happened

Our leading hypothesis — and we want to be clear that it is a hypothesis, not a finding — is that the verified window has been unusually tropical-heavy. Late-season eastern Pacific systems were feeding much of the swell in the set being scored, and open-ocean height from tropical systems is where wave models miss most: the storm is compact, it moves, and small differences in intensity and track produce large differences in the swell that radiates out.

We have not confirmed that the composition of the verified set explains the slide, so we are not claiming it does. We are checking. If it turns out the composition is ordinary and the model still drifted, that is a problem in our physics and we will publish that too.

What this changes in what you read from us

While the share of calls landing within 2 ft is below 70%, we have a standing internal rule: no bare point size figure. Every height we publish carries a range or an explicit caveat in the same sentence. A naked number implies a precision the current data does not support, and we would rather look less confident than be less honest.

Two other things that do not change, ever:

• We publish the misses. When we name a day and the day does not deliver, the follow-up goes out within 48 hours saying so. Busted calls do not quietly expire.
• We state confidence explicitly. A swell already in the water and a storm that has not formed yet are not the same forecast, and we will tell you which one you are reading.

Check us yourself

The most useful thing you can do with this article is stop taking our word for it. Pick a spot, read the forecast a few days out, then go look at what showed up. These are the pages our readers check most:

• The Wedge, Orange County — a serious hazard break; treat the size numbers there with real respect
• Shaw's Cove, Orange County
• Trestles
• Manasquan, New Jersey
• Ocean City, New Jersey
• Robert Moses State Park, New York

Set an alert on the break you actually surf, watch our call against the day, and hold us to it. That is the only accuracy test that matters to you, and it is the one we are trying to earn.

Figures in this article are a point-in-time snapshot taken September 14, 2026 (n=205, rolling 7 days, open-ocean swell height). They will change. We will publish the new ones whichever way they move.