fog/time

✕

how fog/time works

THE NOW LINE

The big colored readout is the field's latest METAR, decoded — lowest ceiling group and visibility — colored by flight category (green VFR · blue MVFR · red IFR), plus current wind (direction° speed, gusts), with the exact observation time and its age. Tap it to see the raw METAR. Observations refresh every 5 minutes and whenever you return to the tab.

THE TODAY TIMELINE

One bar, 05:00–22:30 Pacific (airport-local) time — all times on this page are. Grey = when a deck (ceiling < 3,000 ft) has actually been observed over the field so far today. Lighter grey = hours where the model expects a deck from now on (HRRR low-cloud ≥ 50%) — same substance as observed, fainter certainty. Prediction windows: green box = clearing (good), amber box = fog returning, each a best-time marker inside an empirical ± band. Red is reserved for fog itself. Dashed verticals are sunrise and sunset; the white line is now, and it moves.

THE FOUR ESTIMATORS

CLARO — our fitted model (Ceiling Lift And Return Outlook): ridge regressions for burn-off and return time plus a logistic model for hourly P(ceiling < 1,000 ft), trained in your browser on three warm seasons (Apr–Oct) of archived METARs, 05:00 SFO→SAC / SFO→RNO pressure gradients, HRRR archive fields, and T₈₅₀−T_2m as an inversion-strength proxy — the marine layer is capped near 850 hPa, and inversion structure governs stratus (Iacobellis & Cayan 2013). Architecture follows the KSFO stratus systems (Hilliker & Fritsch 1999; MIT-LL Marine Stratus Forecast System). HRRR — NOAA's 3-km hourly-refresh model, read directly: first hour under 40% low cloud = clears; first over 60% = returns. Recent days — medians of what actually happened, last 7 days. Gradient — live SFO→SAC onshore push: stronger push, deeper deck, later burn-off, earlier evening return.

THE PROBABILITY STRIP

Eighteen cells, 05–22h: CLARO's P(ceiling < 1,000 ft) for each hour today — anchored to today's dawn ceiling (the feature it was trained on), so the strip is stable through the day and identical for every viewer, refreshing with observations. Green low, red high. Tap a cell for the number.

THE 14-DAY DRAWER

Verification view: each row a day, time across. White = observed events (dashed/hollow burn-off, solid/filled return); green/amber = the model's burn-off/return for that same day; gutter ticks: white = an observed event that day, red = model kept the deck all day. This is where you audit the model against reality.

VERIFICATION & HONESTY

CLARO is scored walk-forward — refit on strictly earlier days, predict blind — against HRRR, persistence, and climatology; the full table and coefficients live in the CLARO Lab at the bottom of the page. Card ± bands are the 80th percentile of CLARO's own hindcast errors. When a deck is present, a break-today confidence also appears: a logistic model over every deck-day in the archive (did it clear before late afternoon?), verified the same walk-forward way and shown only where it beats climatology (positive Brier skill) — otherwise the field's own 3-season base rate is shown, labeled as such. A coastal-fog tool. Experimental — not for flight decisions.

CLARO — VERIFICATION & COEFFICIENTS

Walk-forward scores (last 45 event-days, refit on strictly earlier days, predicted blind). MAE in minutes; baselines: raw HRRR, persistence, 7-day median. BSS = Brier skill vs climatology for P(<1,000 ft). Coefficients are standardized β (minutes per SD; positive = later clearing). Card ± bands are the 80th percentile of CLARO's blind errors.

DATA & CADENCE

Observations: freshest of NWS, AviationWeather, and the AWOS/ASOS live feed (via IEM), polled every 2 min · archive: Iowa Environmental Mesonet · model: Open-Meteo (HRRR + historical runs) · pressures: SFO/SMF/RNO. Observations every 5 min · now-line every minute · model rebuild hourly · CLARO refit daily.