Data Sources & Methodology

This site exists to settle a specific kind of argument with sources instead of recall. That only works if the method is inspectable, so here is the whole thing: what counts, where each field comes from, how the datasets are joined, and what was left out on purpose.

What Counts as a Florida Impact

A storm is in this dataset if, and only if, NOAA's Continental United States Hurricane Impacts/Landfalls table records hurricane-force sustained winds in Florida for it. That table is built by NOAA's Hurricane Research Division from estimated wind fields, which matters: a hurricane's wind field extends far beyond its center, so a storm can batter Florida while its center never crosses the state (1852's "Great Mobile" made landfall in Alabama yet produced hurricane-force wind in two Florida regions), and a storm whose center crosses Florida at tropical-storm strength does not count. Center tracks alone cannot make that distinction; NOAA's wind-field analysis can, so its table is the sole authority here for the yes/no question.

Florida regions follow NOAA's own dividing lines: north/south from Cape Canaveral (28.45N) to Tarpon Springs (28.17N); west/east from 82.69W at the Georgia border to Lake Okeechobee, then south along 80.85W. Codes like SW4 mean Southwest Florida, Category 4, per the table's state-by-state Saffir-Simpson estimates. An I- prefix marks an inland impact, per the same table.

The Two-Dataset Join

Track, intensity, and lifecycle facts come from HURDAT2, NOAA's Atlantic best-track database (1851 to present). Each Florida row in the impacts table is matched to its HURDAT2 storm as follows.

The join key is the season year plus the month the storm affected Florida, never its formation month; a storm that forms in late August and reaches Florida in September joins under September, exactly as the table lists it. A storm lingering near Florida across a month boundary is eligible under every month it spent there. Named storms (1950 onward) match by name. Earlier, unnamed storms match by nearest central pressure between HURDAT2's minimum and the table's listed landfall pressure, with same-month candidates assigned one-to-one so two rows can never silently claim one storm.

When the join cannot separate candidates with full confidence, the record is disclosed as such rather than guessed silently, marked with a * wherever it appears on this site. This is a permanent limit of 19th- and early-20th-century source data, not an open task -- there is no further investigation that would resolve most of these. 32 of the current 128records carry this disclosure, all of them pre-1950, split three ways:

• 12 are genuine near-ties: two or more real historical storms have real, differing pressure readings close enough that either could be the match. One guaranteed example: June 1886 produced two different Florida hurricanes with identical 970mb listed pressures; no algorithm should pretend to tell those apart.
• 12 have more than one possible historical storm, but no pressure reading survives for any of them to disambiguate at all -- unresolvable from this source data, regardless of methodology.
• 8 had only one possible historical storm by year and month, so there was never a real choice to make -- the match is correct by elimination, just not independently confirmable by pressure.

A fourth possible flag, pressure_mismatch_large, checks whether a storm's pressure at its actual Florida-relevant point (not its lifetime minimum, which can come from anywhere in its life) is still far from the table's figure before raising it -- a storm that legitimately weakened before reaching Florida is not a wrong match. No current record trips this flag.

ENSO Classification

Each season's phase comes from NOAA CPC's Oceanic Nino Index: the 3-month running mean of ERSST.v5 sea-surface-temperature anomalies in the Nino 3.4 region. A hurricane season takes the phase of its ASO (August-September-October) value under CPC's official episode rule: the +/-0.5°C threshold must hold for at least five consecutive overlapping 3-month seasons to count as an El Nino or La Nina episode. A brief spike that touches the threshold outside such a run stays Neutral.

That rule produces answers that sometimes contradict casual memory, which is rather the point. The 2017 season (Irma) is often remembered as La Nina, but ASO 2017 sat at -0.34, before that winter's episode began: officially Neutral. ONI coverage begins in 1950; the 84 storms before that are shown without a phase and are excluded from phase-based counts rather than backfilled from proxy reconstructions.

Derived Fields

Approach side (Gulf vs Atlantic) is derived from the HURDAT2 track by walking backward from the Florida-relevant point to the first clearly Gulf-side or Atlantic-side position. It is a documented heuristic, useful for aggregate comparison, and it is labeled as derived rather than NOAA-published. Peak category is computed from HURDAT2 maximum sustained winds on the Saffir-Simpson scale, using the current HURDAT2 reanalysis values even where they differ from older popular figures.

Storm Genesis Type & the Current Events Base Rate

Every storm is also classified by where its very first HURDAT2 track point falls, into one of four basin types: Straits (the Florida Straits / Cuba shelf corridor -- a distinct, well-documented environment where a narrow, shallow shelf disrupts storms differently than open water), Caribbean, Gulf ("regional/homegrown" in the conversational sense -- a storm with no long ocean-crossing track to speak of), and Atlantic (everything else, including the classic Cape Verde corridor). This exists because a live storm on the Current Events page and a historical "analog" should only be compared on track shape when they're genuinely the same kind of storm to begin with -- a storm born already in the Gulf has no long approach to diverge from a typical recurve; a storm born off Africa does. Boxes are checked in a fixed priority order (Straits, then Caribbean, then Gulf, then Atlantic as the catch-all), so a point in more than one box's range resolves consistently; see etl/genesis_type.php in the codebase for the exact boundaries.

This four-way split answers a narrow question -- what kind of storm is this, for the purpose of a fair live comparison -- and it is a genuinely different system from Zone Lab's ten named zones (GEZ, Highrider, Lowrider, Carib East, Carib West/Yucatan, three lower-confidence drag zones, Florida_Land, and Atlantic_Zone). Genesis type is a single fixed label assigned once, from a storm's very first tracked point. Zone Lab's zones are a sequence: they describe every zone a storm's inbound track actually passed through on its way to Florida, in order, which is a different and more detailed question -- a storm born in the eastern Caribbean can transit Carib East, then Carib West/Yucatan, then the Yucatan Channel Drag Pocket, then GEZ, then Florida_Land itself at landfall, a five-zone sequence that "Caribbean genesis" alone says nothing about. The two systems are deliberately not collapsed into one classification; they answer different questions at different resolutions, from separate code paths, and a storm's label in one carries no implication for its zone sequence in the other.

Detection-Era Reliability

Genesis-point reliability is not constant across the historical record, and this site does not pretend otherwise. Per NHC's own HURDAT2 format documentation: a storm's formation (tropical depression) stage was not consistently recorded in best track before 1882; aircraft reconnaissance began in 1944, improving detection near reconnaissance range but not basin-wide; and "routine satellite imagery for the Atlantic basin" -- NHC's own words -- began in 1968. Before that year, a storm's first recorded position is disproportionately likely to be wherever a ship or coastal observer first happened to notice it, near land or a shipping lane, rather than where it actually formed. This is not a small effect: of the storms in this dataset whose genesis point falls in the Straits corridor specifically -- a heavily-trafficked shipping lane -- roughly seven in ten predate 1968.

Zone Lab's heat overlay tool needs a finer version of this same reliability ladder than the single 1968 cutoff used here, since it interpolates across the entire 175-year record rather than gating one yes/no eligibility check. It splits detection era into four regimes instead of two: Pre-1911 (ship reports and land stations only, reconstructed via the Fernandez-Partagas & Diaz method), 1911-1943 (station data plus the COADS ship-observation network, still decades before aircraft or satellite), 1944-1967 (aircraft reconnaissance, pre-radar and pre-satellite), and 1968-present (the same modern era this page's binary threshold marks). The two systems agree where they overlap -- both treat 1968 as the start of reliable, satellite-era detection -- they just differ in how finely they subdivide everything before it, because they're built for different jobs: this page is gating a single comparison; Zone Lab is interpolating a heat map across the whole record and needs to know exactly how much to trust each era's points.

Because of this, genesis type is shown for every storm regardless of era -- nothing is hidden -- but it is only used for live-storm comparisons on the Current Events page when the historical candidate's own genesis falls in 1968 or later. A live storm's own genesis point is always current, high-confidence NHC data; comparing it against a pre-1968 storm's detection artifact would not be a fair fight. Storm-to-storm comparisons in the Analog Finder's Compare mode are not restricted this way, since both sides of that comparison are already historical.

The Island Drag Area Signal (1911-1943)

Zone Lab's Island Drag Area zone is negative -- drag-consistent, storms weakening as expected -- in every detection era except one: 1911-1943, where it flips positive (+2.86kt/step, n=63 steps, 9 storms). This has been independently confirmed by two different computation methods and is not a data-quality artifact.

It was investigated further. The signal is not unique to this zone -- every low-traffic zone shows the same era dominated by just 2-3 storms, since 1911-1943 has far fewer storms than surrounding eras in every zone except Florida_Land, which sees every landfalling storm and averages out over a much larger sample. Splitting the window at 1926 -- the documented start of a well-known active Atlantic hurricane era -- does not reveal a regime break either: the zone reads positive on both sides of that line, so the flip isn't explained by that seam. Looking at the zone's full 175-year history in 5-year windows, 1915-1943 is the longest unbroken same-direction stretch on record for this zone -- but a permutation test shows a run that length has roughly an 11% chance of occurring by chance alone, well short of a defensible significance threshold.

The storms driving it are real, not data errors -- among them the Great Miami Hurricane (1926) and the Lake Okeechobee Hurricane (1928). The honest conclusion: this is a genuine, correctly-computed signal whose cause remains unresolved, most plausibly a small number of intense storms clustering in a low-traffic corridor during a data-sparse era, but not statistically distinguishable from coincidence. It is kept on the record rather than smoothed over, and flagged here as an open question rather than a finding.

The "Went On To Become" Percentage

The Current Events page shows a live tropical storm or hurricane's historical base rate: of all Atlantic storms that formed the same way (same genesis type) in the same calendar month since 1968, what share went on to produce hurricane-force wind in Florida. This is computed from the full, unfiltered HURDAT2 basin -- 2006 Atlantic storms total, 978 of them in the modern (1968+) era used for this calculation -- not just the storms that hit Florida. It is bucketed by genesis type and month only, deliberately not ENSO phase: adding ENSO as a third bucket dimension was tested and found to sharply thin real sample sizes without meaningfully changing the numbers. When fewer than eight historical storms share a live storm's exact genesis type and month, no percentage is shown at all -- an honest "not enough data" rather than a precise-looking number built on two or three storms. That minimum-sample threshold, like the pressure-gap threshold above, is a considered starting point rather than a formal statistical derivation, and is open to revision.

Storm Surge

Surge data is landfall-zone scoped: one curated peak figure for the zone where the storm came ashore, not a full-coastline analysis. Values come from NHC Tropical Cyclone Reports (or the earlier Preliminary Reports) and are recorded with the datum the source used; datums vary by era, from storm tide above sea level in older reports to inundation above ground level in modern ones, so cross-era comparisons carry that caveat. Coverage is deliberately null-tolerant: only storms whose figures have been verified against an NHC source carry a value, and everything else, including all pre-TCR-era storms, shows no curated record. Reliable surge observation drops off sharply before the modern gauge and survey network existed, which skews surge comparisons toward modern storms. That skew is disclosed here rather than papered over: it is a property of the historical record, not a flaw this tool can fix.

Deliberately Excluded

Rainfall is excluded outright, not deferred. Hurricane rainfall varies too sharply over short distances (15 inches to 2 inches within a mile is documented) to honestly represent as one number per storm; a single figure would misrepresent the storm rather than describe it.

Casualty figures are excluded outright. Direct-versus- indirect death attribution is genuinely contested even within official records, and this site does not present contested numbers as settled fact.

Storm Tags

Each storm carries machine-derived tags computed from its HURDAT2 track. Rapid Intensification: maximum sustained wind rose at least 30 knots within any 24-hour span of the storm's life, the standard NHC threshold, scoped to the whole lifetime (a Gulf intensification counts even if the storm weakened before landfall). Cross-Peninsula Track: the storm center traversed the Florida peninsula, judged by an interior track point plus presence on both the west and east flanks; track segments are interpolated between six-hourly fixes so a coast exit between recorded positions still registers, while Keys-only, Panhandle-only, and offshore-looping tracks correctly do not qualify. Major Hurricane: lifetime Category 3 or higher; note a storm can be a lifetime major yet strike Florida below Cat 3, and the per-region impact categories remain the Florida truth. Tags are stamped with the enrichment version that produced them and are re-derived, never hand-edited, whenever that logic changes.

Sources

HURDAT2 (Atlantic best track)

NOAA National Hurricane Center. Tracks, winds, pressures, landfall records, 1851 to present.

Continental US Hurricane Impacts/Landfalls

NOAA AOML Hurricane Research Division. The authoritative per-state, per-region hurricane-force impact record, 1851 to present.

Oceanic Nino Index

NOAA Climate Prediction Center. Season phase classification, 1950 to present.

Current Dataset Vintage

generated 2026-07-27T16:09:15Z
enrichment logic v2.6.0
HURDAT2: 1851-2025 complete (2025 post-season data, via GitHub mirror of NOAA HURDAT2; production re-pulls from NOAA)
impacts table: 1851-2025, March 2026 revision (fetched from aoml.noaa.gov 2026-07-20)
ONI: DJF 1950 - AMJ 2026 (fetched from cpc.ncep.noaa.gov 2026-07-20)
zone geometry (Zone Lab): 2026-07-25T07:50:26Z
provisional current-season storms: 2026-07-24T20:36:19Z

Raw source files are archived exactly as fetched, in dated snapshots that are never overwritten, so every dataset build can be reproduced and audited against the originals. Zone geometry runs on its own, separately hand-maintained track from the annual storms.json rebuild above, and provisional current-season storms update independently of either -- each line above reflects that source's own last update, not the storms.json rebuild date.

Download the full dataset as CSV and check the numbers yourself; the export includes join methods, confidence levels, and review flags for every storm.