Method · report dated 2026-09-27
Nothing reached SOLID
29 features of offshore forcing and coarse geometry were tested on 31 breaks. None passes Holm correction and none survives every robustness check. Six size features and three wind features clear BH-FDR q < 0.06 (not < 0.05) with bootstrap CIs that exclude zero. The directional items are hypotheses to test on a larger, better-matched set, not findings.
Groups
Longboard (L, n = 16): waves widely named in mainstream English-language surf media as premier longboard or log waves. Ambiguous (A, n = 6): Rincon, Snapper/Rainbow, Raglan, Arugam Bay Main Point, Chicama and Pavones, ridden often on longboards but equally known as performance waves. Control (C, n = 9): Pipeline, Teahupo'o, Puerto Escondido, Jeffreys Bay, Cloud 9, Mundaka, Supertubos, Kirra and Cabo Blanco, chosen for fame as heavy waves. The main test uses L+A (22) against C (9). Selection is subjective and biased toward English-language media; the controls are not matched shortboard waves in moderate climates, and that matters for the size result. Ericeira/Ribamar, Doheny, Tea Tree and Huanchaco were left out as not clearly longboard waves or as grid-cell duplicates.
Take-off coordinates
Each candidate point was checked against OpenStreetMap natural=coastline ways and placed 100–600 m seaward of the coastline, by break type. Shore orientation is the normal of roughly 1 km of coastline down the line of peel (points) or centered on the peak (reefs and beaches). Twelve of 31 first-guess coordinates were on land before this step. As a check, the offshore mean wave direction combined with the shore normal predicts the correct peel direction at all 22 single-direction breaks.
Data (free sources only)
- Waves
- Open-Meteo marine archive,
era5_ocean: hourly wave height, mean period (Tm, not Tp) and direction, 2015-01-01 to 2024-01-31, ending before the 2024-02-01 ERA5-Wave shift (Hs about 9% low, Tm about 0.5 s low after it). A full-period rerun to 2025-12-31 changed no δ by more than 0.14. Cells are 3–53 km from the take-off (median about 26 km); 31 breaks map to 28 cells (Snapper and Kirra, Sayulita and El Anclote, San Onofre and Cardiff share cells). - Swell partitions
- Meteo-France MFWAM (
meteofrance_wave), 2022–2025, about 8 km grid, 1–17 km from the take-off. - Wind
- Open-Meteo archive, ERA5 0.25°, 10 m wind at the take-off, 2015–2025, local time.
- Tide
- Open-Meteo
sea_level_height_msl(Meteo-France SMOC with FES tides), 2024 hourly. Mean daily range agrees with NOAA great diurnal range within 0.04 m at Santa Monica (1.66 vs 1.65), La Jolla/Cardiff (1.64 vs 1.62), Santa Barbara/Rincon (1.68 vs 1.64) and Honolulu/Waikiki (0.62 vs 0.58). - Bathymetry
- GEBCO 2020 via OpenTopoData on seven transects (shore normal ±60° in 20° steps, 200 m spacing, to 19 km); slope between the 20 m and 50 m crossings, median over transects that do not cross land. GEBCO type ID from the GEBCO WMS at 1–8 km. ETOPO1 as a cross-check.
- Bottom type and wave direction
- Surf-guide consensus recalled from memory; web search was unavailable in the session, so these are unverified. Direction is supported by the geometry check above.
Statistics and grades
Each feature gets Cliff's δ (L+A vs C) with a 4,000-draw bootstrap CI and a Mann-Whitney p. BH-FDR and Holm corrections cover all 29 features. Robustness checks: leave-one-break-out, strict (L only vs C), ambiguous moved to control, and one break per ERA5 cell.
- SOLID q < 0.05, CI excludes 0, and the same sign with |δ| ≥ 0.33 under every check.
- DIRECTIONAL CI excludes 0, and either q < 0.25 or robust.
- KILLED otherwise.
What held (directional)
- Glassy mornings. Share of 06:00–10:00 hours with wind under 2 m/s: 0.39 vs 0.15, δ +0.64 (CI +0.26 to +0.94), p 0.006, q 0.054, Holm 0.17. Leave-one-out δ +0.60 to +0.79; strict +0.71; ambiguous-as-control +0.58; tropics +0.56, outside +0.73. Not explained by latitude (ρ 0.01) or distance to the wind cell (ρ 0.07); keeps δ +0.39 after ranking out the wave ceiling. Caveat: ERA5 wind at 25 km cannot see sheltering by headlands or bays. Strongest result.
- Longboardable daylight hours (Hs 0.6–1.5 m and glass or light offshore): 0.31 vs 0.15, δ +0.66, q 0.054; weakens to +0.35 with ambiguous-as-control. Combines items 1 and 3.
- Low wave ceiling and frequent hits of the size window. Hs p90 1.61 vs 2.66 m (δ −0.58); hours above 2.5 m 0.4% vs 12%; hours at 0.6–1.5 m 81% vs 29% (δ +0.58); days with 3+ window hours 88% vs 39%; months with half their hours in the window 12 vs 3. One factor. The same signal appears in MFWAM (p90 δ −0.53, window δ +0.45). Collapses to |δ| 0.15–0.31 with ambiguous-as-control: it likely reflects the choice of controls.
- Year-to-year reliability of the window share. CV 0.07 vs 0.18, δ −0.70 (CI −0.94 to −0.38), the best raw p (0.003). Mostly "a high share is also a stable share"; shrinks to about −0.40 when the window level is ranked out.
- Clean mornings (glass or light offshore): 0.69 vs 0.49, δ +0.50, q 0.10; weakens to +0.29 with ambiguous-as-control.
- Size after a crude refraction factor: δ −0.48, q 0.11, but ambiguous-as-control removes it (+0.06). Treat as killed.
What did not hold (kills list)
- An obliquity "sweet spot": longboard obliquities run from 1° (Cardiff) to 174° (Imsouane) with no band; δ(|obliquity|) +0.14; hours at 20–60° δ −0.20; regional obliquity +0.03.
- Consistency mattering more than size: hourly Hs CV δ −0.16.
- A slope threshold: GEBCO 20–50 m slope δ −0.22 (CI −0.70 to +0.29). J-Bay 0.003, Supertubos 0.004 and Kirra 0.004 sit on gentle offshore slopes.
- An Iribarren (breaker-type) separation: longboard 0.08, control 0.14, δ −0.15; at the offshore slope almost every break reads as "spilling".
- Longer period: Tm δ −0.07; swell Tm +0.15; hours at 12 s or more +0.23.
- A narrower directional window: circular SD δ +0.03.
- Tide range and relative tide range: δ +0.16 and +0.33, CIs through zero.
- Swell share of energy: δ +0.32, CI through zero; little spread.
- Seasonality and afternoon onshore wind: δ −0.40 and −0.33, CIs through zero.
- Coast turn at the take-off: δ +0.27; wrap-around geometry is common among controls too.
- Geometry-filtered size: removed by ambiguous-as-control.
Full test table (29 features, sorted by p)
Group medians are longboard (L+A, n = 22) and control (n = 9). δ is Cliff's delta: +1 means every longboard break is higher than every control. Copied from the report of 2026-09-27.
| Feature | Longboard median | Control median | Cliff δ | 95% CI | p (MWU) | q (BH) | p (Holm) | LOO δ range | Strict δ | Ambiguous→control δ | One-per-cell δ | Grade |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
window_frac_interannual_cvyear-to-year variation of the 0.6–1.5 m window share | 0.070 | 0.175 | -0.70 | -0.94 to -0.38 | 0.003 | 0.054 | 0.08 | -0.80 to -0.66 | -0.67 | -0.33 | -0.76 | DIRECTIONAL |
frac_daylight_window_and_cleandaylight hours with Hs 0.6–1.5 m and glass or light offshore | 0.312 | 0.152 | +0.66 | +0.29 to +0.95 | 0.005 | 0.054 | 0.14 | +0.62 to +0.76 | +0.65 | +0.35 | DIRECTIONAL | |
morning_glass_fracglassy mornings (06–10 h, wind under 2 m/s) | 0.386 | 0.152 | +0.64 | +0.26 to +0.94 | 0.006 | 0.054 | 0.16 | +0.60 to +0.79 | +0.71 | +0.58 | DIRECTIONAL | |
frac_days_3h_in_windowdays with at least 3 window hours | 0.884 | 0.390 | +0.60 | +0.23 to +0.90 | 0.010 | 0.054 | 0.27 | +0.55 to +0.70 | +0.53 | +0.21 | +0.66 | DIRECTIONAL |
hs_p9090th-percentile Hs (wave ceiling) | 1.610 | 2.660 | -0.58 | -0.88 to -0.21 | 0.013 | 0.054 | 0.33 | -0.68 to -0.53 | -0.56 | -0.31 | -0.63 | DIRECTIONAL |
frac_hs_gt_2.5hours above 2.5 m | 0.004 | 0.119 | -0.58 | -0.88 to -0.21 | 0.013 | 0.054 | 0.33 | -0.67 to -0.53 | -0.53 | -0.17 | -0.62 | DIRECTIONAL |
frac_hs_0.6_1.5hours at 0.6–1.5 m | 0.811 | 0.292 | +0.58 | +0.19 to +0.88 | 0.013 | 0.054 | 0.33 | +0.53 to +0.68 | +0.50 | +0.17 | +0.63 | DIRECTIONAL |
months_window_ge_50pctmonths with half their hours in the window | 12.000 | 3.000 | +0.54 | +0.14 to +0.85 | 0.016 | 0.056 | 0.36 | +0.49 to +0.66 | +0.48 | +0.15 | +0.61 | DIRECTIONAL |
hs_medianmedian Hs | 1.170 | 1.680 | -0.56 | -0.88 to -0.16 | 0.018 | 0.056 | 0.37 | -0.66 to -0.50 | -0.52 | -0.23 | -0.61 | DIRECTIONAL |
morning_clean_fracclean mornings (glass or light offshore) | 0.689 | 0.491 | +0.49 | +0.02 to +0.87 | 0.035 | 0.101 | 0.70 | +0.43 to +0.68 | +0.50 | +0.29 | DIRECTIONAL | |
hs_geomfiltered_medianHs after a crude refraction factor | 0.512 | 0.911 | -0.48 | -0.82 to -0.08 | 0.041 | 0.108 | 0.78 | -0.57 to -0.43 | -0.33 | +0.06 | DIRECTIONAL | |
season_ratio_hsseasonality of Hs | 1.396 | 1.600 | -0.40 | -0.79 to +0.04 | 0.090 | 0.216 | 1.00 | -0.51 to -0.34 | -0.47 | -0.46 | -0.32 | KILLED |
afternoon_onshore_gt5_fracafternoon onshore wind over 5 m/s | 0.023 | 0.044 | -0.33 | -0.70 to +0.11 | 0.157 | 0.339 | 1.00 | -0.41 to -0.25 | -0.36 | -0.19 | KILLED | |
relative_tide_rangetide range relative to wave height | 1.768 | 1.210 | +0.33 | -0.13 to +0.75 | 0.164 | 0.339 | 1.00 | +0.24 to +0.47 | +0.33 | +0.18 | KILLED | |
swell_energy_frac_medianswell share of wave energy | 0.998 | 0.979 | +0.32 | -0.13 to +0.71 | 0.177 | 0.343 | 1.00 | +0.23 to +0.44 | +0.40 | +0.33 | +0.24 | KILLED |
frac_geomfiltered_0.6_1.5window share after refraction factor | 0.322 | 0.473 | -0.29 | -0.67 to +0.11 | 0.215 | 0.389 | 1.00 | -0.35 to -0.24 | -0.08 | +0.24 | KILLED | |
coast_turn_degcoast turn at the take-off | 17.000 | 14.000 | +0.27 | -0.16 to +0.67 | 0.248 | 0.423 | 1.00 | +0.18 to +0.35 | +0.22 | +0.03 | KILLED | |
frac_tswell_ge_12hours with swell period 12 s or more | 0.099 | 0.053 | +0.23 | -0.23 to +0.67 | 0.327 | 0.527 | 1.00 | +0.15 to +0.38 | +0.28 | +0.29 | +0.16 | KILLED |
gebco_slope_20_50seabed slope, 20–50 m | 0.007 | 0.019 | -0.22 | -0.70 to +0.29 | 0.361 | 0.550 | 1.00 | -0.34 to -0.12 | -0.16 | +0.04 | KILLED | |
frac_hours_obl_20_60hours at 20–60° obliquity | 0.131 | 0.209 | -0.20 | -0.64 to +0.25 | 0.396 | 0.574 | 1.00 | -0.34 to -0.11 | -0.03 | +0.25 | KILLED | |
hs_cvhour-to-hour variability of Hs | 0.298 | 0.320 | -0.16 | -0.60 to +0.31 | 0.514 | 0.646 | 1.00 | -0.29 to -0.07 | -0.12 | -0.05 | -0.12 | KILLED |
tide_mean_daily_range_mmean daily tide range | 1.507 | 1.361 | +0.16 | -0.32 to +0.60 | 0.514 | 0.646 | 1.00 | +0.06 to +0.29 | +0.17 | +0.07 | KILLED | |
iribarren_tmIribarren (surf-similarity) number | 0.075 | 0.136 | -0.15 | -0.63 to +0.38 | 0.528 | 0.646 | 1.00 | -0.27 to -0.06 | -0.07 | +0.09 | KILLED | |
tswell_medianswell mean period | 9.325 | 8.350 | +0.15 | -0.32 to +0.61 | 0.542 | 0.646 | 1.00 | +0.07 to +0.28 | +0.17 | +0.13 | +0.06 | KILLED |
obliquity_absobliquity of mean swell to the shore | 106.316 | 65.142 | +0.14 | -0.32 to +0.59 | 0.557 | 0.646 | 1.00 | +0.05 to +0.25 | +0.03 | -0.17 | KILLED | |
tm_medianmean wave period | 8.950 | 8.950 | -0.07 | -0.51 to +0.38 | 0.794 | 0.885 | 1.00 | -0.18 to +0.02 | -0.03 | +0.04 | -0.16 | KILLED |
obliquity_regional_absobliquity to the regional coast | 71.179 | 70.142 | +0.03 | -0.42 to +0.48 | 0.913 | 0.931 | 1.00 | -0.09 to +0.11 | -0.07 | -0.17 | KILLED | |
obliquity_hourly_iqrspread of hourly obliquity | 30.500 | 32.000 | -0.03 | -0.44 to +0.39 | 0.931 | 0.931 | 1.00 | -0.10 to +0.09 | +0.01 | +0.08 | KILLED | |
swell_dir_circstdswell direction spread | 24.448 | 21.232 | +0.03 | -0.40 to +0.45 | 0.931 | 0.931 | 1.00 | -0.07 to +0.15 | +0.21 | +0.35 | -0.04 | KILLED |
Method checks and limits
- Bimodal direction. At Waikiki and Pipeline the energy-weighted mean direction mixes trade-wind sea with groundswell. Dropping Waikiki moves no δ by more than 0.06; dropping Pipeline no more than 0.13.
- Grid-cell distance. Restricting to breaks within 35 km of their ERA5 cell (n = 20) leaves glass (+0.75), window (+0.53) and ceiling (−0.59) essentially unchanged.
- Multiple comparisons. No Holm p below 0.05; best BH q 0.054.
- Selection bias. Longboard breaks chosen by reputation, controls by fame as heavy waves. A fairer next test would use moderate-climate performance waves as controls (Trestles, Burleigh, Bells, Superbank and similar). The ambiguous-as-control run is a first look: it removes the size signal and keeps the wind signal.
- Geography. 4 of 22 longboard breaks are in California and 4 on the Australian east coast. Cell-level duplicates were handled; coast-level clustering was not.
- A natural contrast. Snapper/Rainbow and Kirra share the same ERA5 cell, tide and nearly the same wind. They differ only in the orientation of the peel stretch (342° vs 20°), obliquity (119° vs 81°) and coast turn (46° vs 0°). Whatever separates their character sits in local geometry and sand.
- Scope. Offshore forcing, coastline orientation and 450 m bathymetry only. Nothing here measures peel angle, peel speed, section structure or wave shape; claims about wave shape wait for a fixed camera and a nearshore survey.
Data: CSV · JSON · table, map and charts · per-break pages.