Skip to content
Coastal IntelligenceVerification

The beach decides: a nearshore wave forecast, locked before the test and graded by 5 buoys

A research verification: on 17 October 2026 we locked a method for estimating wave height close to Southern California's beaches, then let 15 days of buoy readings grade it. The main method missed the target we set in advance and stays off the map. Two backups passed: wave physics with one regional correction cut the typical nearshore error from 0.36 m to 0.28 m, and a correction learned from each beach's own buoy cut it to 0.21 m. Local knowledge, measured, beat the better model.

Case reconstructions

Research film: the nearshore wave study, from the locked method to the buoy score. Satellite imagery and real buoy results; synthetic narration.

AYON Coastal Intelligence · contains modified Copernicus Sentinel data 2026 · buoy data CDIP · 83s

What the map shows today

The wave numbers riders see for a beach come from NOAA's global wave model at its nearest grid point, 5 to 30 km offshore. Near the beach, islands, headlands and shallow water change the waves, and that point does not see them. This study asked how much of that gap a better estimate can close, and how much a beach's own measurements are worth.

Locked first, graded once

The method, its constants and every correction factor were fitted on a 30-day training window and committed on 17 October 2026, before the test window opened. The test ran from 18 October 00:00 UTC to 1 November 23:00 UTC, 360 hours, and was scored once, on 14 November, with the frozen code.

Ground truth is CDIP's directional wave buoys: 7 shallow buoys near the shore, 11 to 41 m deep, and 11 deeper ones. Two shallow buoys, Red Beach and Leucadia, sent no data during the test, so the nearshore score rests on 5 buoys and 1,705 hourly pairs. That is stated, not adjusted.

The scorecard

Every rule below was written before the test window opened. A method that loses is reported as losing.

  • Main method: sheltering plus one regional correctionMiss
    Called
    Error at least 20% below today's, and bias no worse at 5 of 7 shallow buoys
    Measured
    10.5% below (ratio 0.895, 95% interval 0.842–0.981); bias no worse at 4
  • Wave physics (refraction) plus one regional correction, for beaches without a buoyHit
    Called
    Error below a plain regional correction, with the whole 95% interval below 1.0
    Measured
    Ratio 0.881 (0.811–0.934): typical error 0.28 m against 0.31 m
  • A correction learned from each beach's own buoyHit
    Called
    Error below a plain regional correction, with the whole 95% interval below 1.0
    Measured
    Ratio 0.685 (0.619–0.734): typical error 0.21 m
  • Wave direction from wave physicsHit
    Called
    Mean direction error below today's
    Measured
    19.3° against 29.0°

What it means

Today's forecast reads low near the shore, by 0.21 m on average, and most of the gain comes from correcting that. Wave physics helps the direction and helps where a beach has no instrument. Once a beach has its own record, physics adds nothing measurable on top of it (ratio 1.051, interval 0.991–1.102).

By the rules fixed in advance, the map may now show the physics estimate at Southern California beaches without a buoy, labelled MODELED, with its measured error of ±0.28 m and the number of buoys behind it, and never as a surf or breaking-wave height. The main method stays off the map.

Limits

One autumn season, 15 days, 5 shallow buoys. Long Beach Channel kept the largest error under every method. Nothing here measures breaking waves at the sand; that needs field observations, scored the same way.

Evidence discipline

This is research. Estimates are models, not guarantees; a model is never an observation, and a machine estimate is never “confirmed” — only a human reviewing it in the field makes it verified. Nothing on this page is an operational-capability claim, a safety promise, or a prediction of a specific event at a specific place.

References & data

  1. 1.CDIP — Coastal Data Information Program, Scripps Institution of Oceanography — Directional wave buoys used as ground truth, read through the program's public THREDDS service.
  2. 2.NOAA/NCEP GFS-Wave, NOAA Open Data on AWS — The wcoast 0.16° model files for training and test windows, each kept with its sha256.
  3. 3.AYON Coastal Intelligence study record — Protocol registered 09b7c53, method and factors frozen 5770793, test score 1d64f6ca; results-test.json and report-test.md.