Open-Meteo
The principal marine and weather forecast history.
Data foundation live / Analysis & calibration in development
Illustrative image · Not an observed event01 / THE FOUNDATION
PORTUGAL’S SILVER COASTAn ocean of data.
A record that stays.
Adamastor is an empirical ocean-intelligence platform built above existing forecasts and observations. The aim is to understand how the Atlantic changes — and how the coast responds.
22Oeste Surf spotsOne shared historical pipeline
Every forecast has a history.
Successive forecast runs are archived rather than overwritten. Issue, generation, valid and ingestion times remain distinct, with historical and analytical storage in ClickHouse.
Issued→Generated→Valid→Ingested
A separate offshore estimate for Praia do Norte.
Ingested observations, progressively integrated into evaluation.
Independent perspectives remain independent. Models and observations are not merged into an artificial “better number”.
02 / THE ANALYSIS
MVP IN DEVELOPMENTDetection.
Not divination.
A sudden forecast revision. Models drifting apart. A pattern that persists across runs. Adamastor is being built to investigate these signals in their historical context.
Forecast AForecast BObservation
Agreement is a clue.
Not a guarantee.
A disagreement prompts checks on timing, location and source quality. Confidence must follow the evidence, not the appearance of a convincing curve.
The same event, seen days apart.
Compare successive forecasts from four days ahead to the day itself. Study the direction of revisions, persistence across runs and, where matched observations exist, error by forecast horizon.
03 / THE COAST
NAZARÉ → THE SILVER COASTOne Atlantic.
Different responses.
Nazaré is the principal laboratory.
The canyon, bathymetry, swell direction, period and wind make the relationship between offshore swell and Praia do Norte complex. A fixed multiplier cannot honestly describe it.
MODEL / FORECAST
Open-Meteo · Copernicus
Independent offshore estimates
OBSERVATION
MONICAN
Ingested observational dataset
INGESTION & ARCHIVE · LIVE
Historical storage
Forecast runs and observations retained with their timestamps
ANALYSIS · MVP
Local response
Correlation, bathymetry, exposure, wind and tide
Swell height, breaking-wave face height and event potential are distinct quantities. Observations inform evaluation; they are not generated by the model.
The foundation for Coast Radar.
The same method can extend across the 22 spots: learning how each location responds to a given ocean configuration. This empirical local-response layer is a development objective, not an operational prediction claim.
How calibration will be evaluated +
Evaluation must examine true positives, false positives, false negatives, signal stability, timing error and accuracy by forecast horizon. Seasonal baselines and historical analogues provide context; plausible-looking thresholds are not evidence of validation.
04 / BUILT ON EVIDENCE
PROJECT STATUSCollection & archive
Marine forecasts, successive runs and ingested observations. A shared historical foundation.
Analysis & calibration
Correlation, bias, forecast evolution, historical context and confidence synthesis.
Interpretation & prediction
Structured-signal interpretation and future local-response models. No trained proprietary forecast model is claimed.
Analysis first.
Language second.
The engine must produce the structured evidence and deterministic confidence before an LLM explains it. The LLM cannot alter a measurement, choose a status or turn offshore swell into an observed breaking wave.