Sailing Forecast

data captured
—

Before you go: — these decide whether you depart; the AI summary below is interpretation only.

Sources: ECMWF EC46 sub-seasonal ensemble mean, medium-range consensus of , sea state from Open-Meteo Marine, and synoptic charts from charts.ecmwf.int — all via aegean_forecast.py. Each place uses its nearest sea grid cell, not land.

How the medium-range wind numbers are calculated: each of the configured models () is sampled over a 3×3 grid of points around every place — roughly 30×25 km, not a single GPS pin — because a boat sailing that spot experiences the whole patch of sea around it, not one exact point. Wind speed, direction, temperature, and rain are the mean across those 9 points per model, which smooths out single-cell grid noise. Gust and wind max are taken as the max across those 9 points — both are worst-case figures where averaging would hide a real local peak. The models are then combined the same way on top: mean for wind/temp/rain/direction, max for gust — so the gust shown is the strongest plausible gust anywhere in that patch, across all models, a safety margin by design. Wind (max) is the daily maximum mean wind (the highest hourly average in the day), useful as a comparator for gusts: a gust/wind-max ratio of 1.3–1.5 is normal for open sea. A ratio well above that, or a gust much larger than wind-max, is worth noting.

Model disagreement: a place/day is flagged (⚠) when the models' wind speeds spread by more than , or their directions by more than while wind is at least . On a flagged wind cell, the cell shows the actual min–max span (e.g. 10–17 kt) instead of one blended number. On a flagged direction cell, it shows the per-model bearings (e.g. NE / SW) instead of the circular mean; if the model directions cancel completely, the cell shows VAR. Clicking any ⚠ cell opens a popup with the full per-model table, the numeric spread, and an AI note explaining what the disagreement means for that day. The per-model breakdown also goes into runs.csv and the AI sailing summary.

Sea state (wave height, period, direction) uses the same 3×3 kernel as wind, since a boat crossing a patch of sea can meet its roughest cell, not just the exact point: wave height is the kernel max (same worst-case reasoning as gust), period and direction are the kernel mean. Sea state is a single model (Open-Meteo Marine), so there's no further multi-model combine on top like wind has — the kernel is the only combining step here. Wave height is significant wave height (Hs), the average of the highest third of waves; the occasional largest wave is roughly 1.5–2× that. Cells highlighted in amber are steep seas (wave steepness Hs / 1.56T² > 0.04 — roughly 1.0 m at 4 s, 1.6 m at 5 s, 2.3 m at 6 s), which are uncomfortable and potentially dangerous regardless of height alone.

Ensemble/consensus wind smooths out individual fronts — read it as the week's regime, not a day-by-day forecast. EC46 (the coarsest, farthest-out tier: a single 51-member ensemble mean at ∼36 km resolution) is the one tier still read at a single point, with no kernel or per-model combine — it's already someone else's finished ensemble mean. EC46 also fills the back half of the trip when the medium-range horizon hasn't reached those dates yet (the source column in the raw data distinguishes which rows are which). A low EC46 mean late in the week signals uncertainty and ensemble smoothing, not a guaranteed calm. Rows show the region's actual spread (sheltered vs. exposed spots), never a single averaged number, since this route crosses both. This page regenerates on every run.

Poseidon regional forecast (independent, additive): a second, unofficial opinion from HCMR's (Hellenic Centre for Marine Research) own short-range Greek-seas model, shown alongside the ECMWF chart above — not a replacement for the tiers above it, and not folded into their kernel/consensus logic. Own model, own single point per place (no 3×3 kernel). Wind max is the day's highest 3-hourly sample, not a true gust — HCMR's model doesn't provide a separate gust figure, and its wave model gives no period either. This is not part of any documented, versioned public API (found by reading the public Poseidon map's own client code, the same request a browser makes) — it can change or vanish without notice, unlike every other source described above, and some coastal spots simply have no wave data (a masked/land grid cell in HCMR's model, shown as “–” rather than a real reading).