How far ahead can you trust a surf forecast? 344,616 forecasts checked
Every forecast run between 9 Sept 2026 and 17 Oct 2026 predicted the next nine days at 1,102 breaks. We compared each prediction with the score that day actually settled at: 344,616 pairs in all. This measures how much a forecast changes before the day arrives, not how well it matches what you see in the water.
The short version
84.9%
of Firing calls 7 days out settled Good or better (base rate 15%)
51.1%
of those Firing calls stayed Firing: a coin flip at a week
70.2%
of Firing calls 3 days out stayed Firing
4.3 → 10.1
average error in score points, 1 day vs 7 days ahead
A week out: worth planning around
A Firing call seven days ahead rarely turned into a bad day. It settled Good or better 84.9% of the time, against 15% for a random spot-day.
Three days out: worth booking on
By three days out, 70.2% of Firing calls stayed Firing and 97% settled Good or better.
No hype at long range
The average signed error stayed between -1.5 and -0.2 points at every lead: long-range forecasts were not systematically more optimistic than the day itself.
When the forecast says Firing
Of every forecast that called a day Firing (80 or above), how many settled Firing, and how many settled at least Good (65 or above)? Read across to see how much a Firing call is worth at each lead.
79.5% Firing · 99.4% Good or better · 1,759 calls
75.4% Firing · 98.4% Good or better · 1,673 calls
70.2% Firing · 97% Good or better · 1,676 calls
65.4% Firing · 95.2% Good or better · 1,682 calls
62.8% Firing · 93.9% Good or better · 1,599 calls
57.3% Firing · 89.5% Good or better · 1,644 calls
51.1% Firing · 84.9% Good or better · 1,664 calls
46.9% Firing · 83.7% Good or better · 1,690 calls
47.1% Firing · 81.7% Good or better · 1,538 calls
How much warning do you get?
Turn it around: of the days that ended up Firing, how many were already showing up in the forecast? Seven days ahead, 53.8% were forecast Firing and 87.1% were forecast at least Good. So most big days are visible a week out, but often as “good” rather than “firing”. They sharpen as the day gets closer.
| Days ahead | Firing days forecast Firing | Forecast at least Good | Average error |
|---|---|---|---|
| 1 | 75.6% | 98.4% | 4.3 pts |
| 2 | 69.8% | 97.1% | 5.3 pts |
| 3 | 66.9% | 95.7% | 6.3 pts |
| 4 | 64.5% | 93.9% | 7.2 pts |
| 5 | 60.2% | 90.8% | 8.1 pts |
| 6 | 58% | 88.8% | 9.2 pts |
| 7 | 53.8% | 87.1% | 10.1 pts |
| 8 | 51.5% | 83.4% | 11.1 pts |
| 9 | 48.3% | 80.5% | 12.4 pts |
Better than assuming nothing changes
The simplest possible forecast is “tomorrow will be like today.” That guess was off by 11.5 points a day ahead and 15.6 a week ahead. The forecast beat it at every lead, including nine days out (12.4 vs 16.5).
1 day ahead
4.3
forecast error (pts)
11.5
“same as today” error
3 days ahead
6.3
forecast error (pts)
15.2
“same as today” error
7 days ahead
10.1
forecast error (pts)
15.6
“same as today” error
9 days ahead
12.4
forecast error (pts)
16.5
“same as today” error
Where a week-out Firing call held best
Groups with at least 100 Firing calls at seven days. Read the base rate too: where good days are common anyway (Southeast Asia in this window), a Firing call adds less information than where they are rare.
| Group | Breaks | Firing calls | Settled Good+ | Stayed Firing | Base rate Good+ |
|---|---|---|---|---|---|
| Southeast AsiaRegion | 109 | 656 | 91.5% | 57.2% | 40.8% |
| Central AmericaRegion | 115 | 316 | 91.1% | 62.3% | 31.8% |
| ReefWave type | 417 | 786 | 86.8% | 52.4% | 15.8% |
| SouthernHemisphere | 488 | 916 | 85.8% | 49.9% | 16% |
| PointWave type | 208 | 241 | 84.6% | 51.5% | 13.9% |
| South AmericaRegion | 217 | 148 | 84.5% | 38.5% | 10.2% |
| EuropeRegion | 115 | 154 | 84.4% | 50.6% | 15.4% |
| NorthernHemisphere | 614 | 748 | 83.8% | 52.5% | 14.2% |
| BeachWave type | 427 | 634 | 82.8% | 49.4% | 16.4% |
Limits you should carry with every citation
- Checked against the day-of reading, not the ocean. Each forecast is compared with the score the same pipeline gave that day. That shows how much forecasts move before the day; error against what surfers actually see will be larger.
- One window, one season. Forecast runs 9 Sept 2026 to 17 Oct 2026: northern autumn and southern spring. Other seasons and swell regimes may behave differently.
- One scoring engine. The window starts the day after an 8 September 2026 scoring change, so every pair comes from the same engine. Earlier archive rows are excluded on purpose.
- Not independent experiments. Nearby breaks share swell and weather systems, so 344,616 pairs carry far less independent information than the count suggests.
- Uneven coverage. The archive skips a spot on a day when its forecast is stale or failed to load, so some spots contribute fewer pairs.
- The window ends early. Archiving paused after 18 Oct 2026 while the forecast job was timing out (fixed 27 October 2026). An update will extend the window.
Method and reproducibility
- 1
Forecast archive
Twice a day the forecast job scores every break for today and the next nine days. The last successful run of each date is archived: the day-0 score plus the nine predicted scores.
- 2
Pairing
A pair is the score predicted for date + k (k = 1 to 9) and the day-0 score later archived for that same date and break.
- 3
Thresholds, not just averages
Results use the Strike Score bands (Good 65+, Firing 80+) alongside mean absolute error, because "will it be good?" is the question that decides a trip.
- 4
Persistence baseline
The run date’s own day-0 score, used as the forecast for every later date, is the no-skill comparison.
- 5
Group comparison
Regions, hemispheres and wave types appear only with at least 100 Firing calls at seven days.
- 6
Public aggregates
The CSVs hold every lead time, every group and every break, with file checksums in the methodology manifest.
Download and cite the data
The aggregate files are free to inspect. When republishing results, link to this report, name Strike Mission, keep the date window, and say what the forecasts were checked against. Use is governed by the API terms.
Results by lead time
9 rows · CSV
Region, hemisphere and wave type
Every group × lead · CSV
Spot-level results
1,102 breaks · CSV
Methodology + checksums
Versioned manifest · JSON
Suggested citation: Cacciotti, Joel. “How Far Ahead Can You Trust a Surf Forecast?” Strike Mission, 27 Oct 2026, dataset v1.0.
Questions, corrections, or press use?
See the source notes or contact the data desk.