Methodology & Assumptions

Fog Atlas trades per-runway precision for honest, worldwide comparability. Every approximation is listed here. If you need certified minima analysis for a specific operation, use your AFM, OpSpecs, and Jeppesen — not this site.

Observation basis

• Source: routine hourly METARs (IEM ASOS archive; report_type=3), 2016–2025. Special reports (SPECI) are deliberately excluded so that frequencies are an unbiased sample of hours; including SPECIs would overweight deteriorating weather.
• Hourly dedup: the last routine report in each UTC hour (US stations file at ~:53–:56; many international stations file at :00/:30).
• Statistic: "% of hours" = hours in band ÷ hours with a valid visibility report. Station outages reduce the denominator, not the frequency. Coverage is reported per airport.

Bands (visibility OR ceiling)

A CAT I approach needs both visibility above minima and ceiling at or above the ~200 ft decision height, so both terms enter the classification (ceiling = lowest BKN/OVC/VV layer from the sky-condition fields, joined per hour):

Empirical note: ceiling-only hours (vis fine, ceiling < 200 ft) are rare everywhere — single digits to a few dozen hours per year even at stratus-prone airports — because in fog, visibility and ceiling collapse together and the visibility term already catches the hour. The famous counterexample cuts the other way: San Francisco's marine stratus sits at 500–1,500 ft, well above the decision height, so SFO shows almost no sub-CAT-I hours even with ceilings included. Its fog problem is arrival capacity (no paired visual approaches under the deck), not approach minima — a distinction this map now makes verifiably rather than by omission.

RVR cross-check (SFO). Because the visibility-as-RVR-proxy question matters most at the famous airports, we verified it against runway-measured ground truth at KSFO: parsing RVR groups from ten years of raw routine METARs gives 8.9 hrs/yr below the 1,800 ft CAT I RVR minimum (10.8 below a conservative 2,400 ft), against 9.8 hrs/yr from our prevailing-visibility bands — agreement within ~1 hr/yr. SFO's CAT III infrastructure is justified by consequence (a fortress hub losing even one morning a year is expensive) and by history (coastal California fog has declined measurably since the system was installed), not by frequency.

Known approximations:

1. Prevailing visibility ≠ RVR. METAR visibility is a human/sensor prevailing value for the aerodrome; RVR is runway-specific and often better than prevailing visibility in fog (high-intensity lights). Our bands therefore understate what's flyable on a lit CAT I runway and the split should be read as a climatological index, not an ops decision.
2. Thresholds are global constants. Real minima vary per runway, per approach, per operator. ½ SM / 300 m are defensible central values, not authoritative ones.
3. Reporting granularity. US ASOS reports fractions (¼ SM = ~400 m falls in the EFVS band; ⅛ SM = ~200 m falls below). International METARs report meters with their own steps. Band edges sit between common reporting steps where possible.

Persistence statistics (phase 2)

An event is a maximal run of consecutive hourly observations below CAT I (a missing hour breaks the run, so durations are conservative). Per airport we report event-duration quartiles and the survival curve P(still below CAT I k hours after onset), overall and conditioned on season × time-of-day of onset where ≥ 20 events support the cell; airports with < 25 events in ten years get no persistence claims. The live "right now" line classifies the latest METAR (via NOAA AWC) with the same band rules; when an airport is currently below CAT I, the quoted lift odds use the matching season/time-of-day survival curve. Caveat: survival is measured from event onset — a continuing event that began hours ago has different remaining-duration odds than a fresh one, and a single METAR can't tell us the elapsed time.

Nowcast model (phase 3)

A logistic model predicts P(sub-CAT-I at t+2h) from the current METAR (visibility, ceiling, temp-dewpoint spread, wind, visibility trend, prior-hour state, diurnal/seasonal harmonics) plus the airport's climatological rate as an input feature — so the model directly competes with its own baseline. Trained on 2016–2023, evaluated on 14.2M unsampled 2024–2025 hours (strict temporal split):

The logistic model cuts Brier 39% vs climatology and 14% vs persistence; gradient boosting adds another ~9% but doesn't ship (the browser runs the 15-coefficient logistic in one line). The live "Model:" line in each deep-dive is this model scored client-side against the latest METARs. It is experimental and not for operational use — it knows nothing about synoptic weather, fronts, or forecasts; it is a statistical nowcast from station observations alone.

Cause attribution

Present-weather codes, first match in priority order: FG (incl. FZFG) → SN → HZ/FU → BR → other/none. An observation with multiple phenomena is attributed to the highest-priority one.

The app's cause chart folds BR (mist) into the fog family: by definition BR is reported when visibility is ≥ 800 m, so a BR code attached to a sub-CAT-I observation is fog that the observer/algorithm coded conservatively. The pipeline output keeps the raw distinction.

Minima-aware EFVS opportunity, segmented by operator equipage

"EFVS-recoverable" against a global CAT I threshold is optimistic for some operators and pessimistic for others, because the floor an operator can actually achieve is whichever binds: flight deck or ground infrastructure. A CAT III-equipped Part 121 crew at a CAT III hub already lands at RVR 600; a CAT I-equipped Part 135/125 operator is held to CAT I minima even at that same hub — which is precisely the population EFVS retrofits serve. The EFVS opportunity metric is therefore computed per equipage profile:

International CAT I floors are eAIP-researched per runway for the top-ranked airports (agent-researched, spot-audited; pipeline/data/intl_floors.csv) and class-approximated (800 m) elsewhere.

Opportunity = hours/yr below the achievable floor yet within EFVS range (≥ 300 m / ~RVR 1000).

• US floors: per-runway published minima from the FAA ILS Master report (CAT I visibility, SA CAT I / CAT II / SA CAT II / CAT III RVR; lowest across runways) — e.g. SFO: CAT I floor RVR 1800, CAT III floor RVR 600. LPV from FAA CIFP SBAS path points; no-ILS fields get 800 m with LPV, 1600 m without (LNAV-class).
• International: approximated from capability class — CAT III ~175 m, CAT II ~350 m, CAT I 800 m, no-ILS 1600 m — refined by (a) eAIP-researched per-runway CAT I minima for the top-ranked airports, and (b) EGNOS LPV procedures from ESSP's official per-runway dataset (485 airports with operational LPV; LPV200 → 800 m floor at no-ILS fields, plain LPV → ~1100 m).

Counting is conservative: visibility bins count only when entirely below the floor; ceiling-limited hours count only where a ~200 ft DH binds (floor ≥ 450 m). Rankings rank by the selected equipage profile.

Fog chase board (CHASE)

A launch-decision board for EFVS flight testing: which airports within still-air range of a base are below CAT I right now, which the nowcast expects to drop within two hours, and which merely qualify.

• Infrastructure filters are per runway end from FAA NASR (approach light system, runway length, RVR sensor locations) joined with NASR ILS categories and CIFP LPV lines of minima. US airports only; a curated Canadian tier is planned.
• Go-around height filters on the lowest published minima height of the runway end, as a tier proxy: ILS CAT I/II/III → 200/100/50 ft, LPV → ~250 ft, anything else ≥350 ft (LNAV/circling MDAs). These are not chart DAs — terrain-driven exceptions exist. The operational point: an approach whose minima sit at 400+ ft AGL goes missed before an EFVS sensor can acquire the lights in a real fog layer.
• Distance/ETE are great-circle still-air at the user's cruise speed — no winds, no climb/descent allowance.
• IN FOG NOW joins the same ~3-minute NOAA AWC cache as NOW mode, keeping raw METAR text: reported RVR groups (lowest across runways), ceiling, weather string, observation age. Observations older than 75 minutes grey out and never alert.
• LIKELY SOON scores the phase-3 nowcast per candidate from the same cache observation, with trend features (previous-hour visibility / sub-CAT-I state) accumulated client-side from successive cache snapshots plus one bounded multi-station history fetch per session. Top five shown regardless of probability — an honest "<1%" beats a hidden list.
• Alerts (notification + chime on entry into IN FOG NOW) work only while the tab is open. A static site cannot wake a phone; server-side push would be a different architecture, deliberately not built yet.

Reporting reliability

"% of hours" assumes observations sample hours impartially. Two failure modes are detected per station and flagged:

1. Low coverage (< 40% of possible hours in the archive): too thin to support frequency claims.
2. Suspect reporting — the signature of an encoding artifact rather than weather: a majority of sub-CAT-I observations at literal-zero visibility combined with no diurnal structure (real fog is strongly morning-skewed; haze and marine advection fog are flatter, so the threshold is deliberately loose).

Flagged stations are excluded from the fog field and rankings, demoted on the map, and carry a warning banner in their deep-dive. Their raw numbers remain visible for completeness.

CAT II/III capability

• US: derived from the FAA NASR ILS database (ILS_BASE.csv CATEGORY field, best category per airport) — 97 airports with CAT II/III, 586 confirmed CAT I, and the remainder flagged "no ILS on record." At RNAV-only fields, real minima are typically higher than CAT I (LPV ~200–250 ft at best, LNAV/VNAV often 350–400+ ft), so the sub-CAT-I bands understate blocked hours there — and the EFVS case is correspondingly stronger. The map's "ILS only" toggle hides known no-ILS fields; international airports with unknown ILS status remain visible.
• International: the FAA's published OpSpec C060 list of foreign facilities approved for CAT II/III operations — 158 airports. Everything else is assumed CAT I and labeled accordingly; the C060 list reflects FAA approval for US carriers, so a foreign airport with CAT II/III capability not used by US carriers may be missing.
• The EFVS-value framing: at a CAT III airport, suitably equipped airlines already land in fog — EFVS value concentrates where low visibility is frequent and CAT II/III is absent.

Cancellation validation (US only)

US DOT/BTS on-time data (2023–2024, 13.9M scheduled departures) joined to our hourly bands at each flight's origin airport:

Two honest readings of the structure. First, the below-300m multiplier being lower than the EFVS band's looks backwards until you notice where that exposure lives: mostly at CAT III-equipped hubs (ORD, DTW, SEA) where autoland keeps the operation running — which is the CAT II/III-absence thesis showing up in cancellation data. Second, BTS "weather" cancellations are generic: Denver's high multiplier is blizzard-driven, not fog-driven; treat the aggregate as validation that the bands mark operationally hostile hours, not as a fog-specific cost model. JFK's below-baseline rate during sub-CAT-I hours (0.05%) likely reflects proactive schedule thinning being coded as carrier/NAS rather than weather.

Sources

• Iowa Environmental Mesonet ASOS/METAR archive (primary)
• NOAA ISD (international backfill, planned)
• OurAirports (airport metadata)
• FAA CIFP (US approach capability)
• US DOT BTS (US cancellations, planned)

Live world fog map · all airports · by city · forecast verification · methodology · about · llms.txt. Not for operational use.