Methodology

Principles

StormWorn is a record, not a model. Every figure on a place page is counted or measured from a public federal dataset; nothing is estimated, projected or scored by a proprietary method. Where we have to make a choice, such as how near a hurricane must pass to count, the rule is stated here and on the page that uses it.

Data from different agencies uses different geographies, eras and conventions. Our work is to place each record in the counties and cities it touched, keep the original values, and say plainly where a dataset is incomplete.

Places and boundaries

County boundaries are the U.S. Census Bureau’s 2024 cartographic boundary files (1:500,000), covering the 50 states and the District of Columbia: 3,143 counties and county equivalents. Connecticut replaced its eight counties with nine planning regions in 2022, but every historical dataset we use is keyed to the eight counties, so Connecticut pages use the 2020 county boundaries.

Cities are incorporated places with a 2025 population estimate of 10,000 or more, from the Census Bureau’s Gazetteer and Population Estimates. A city is represented by the Census interior point; its records use radius rules described below. Each county has a population centre (Census 2020 mean centre of population), which we use to choose its weather station.

Tornadoes

Source: the Storm Prediction Center’s severe weather database of tornadoes, 1950 through 2025. The database stores a multi-state tornado as one whole-track record plus one record per state segment, and lists up to four counties per segment, with overflow records for tornadoes that crossed more counties. We take the tornado’s attributes (date, rating, path length and width, injuries and deaths) from the whole-track record and assign it to every county listed in any of its segments. Since 2007 the record number is the start date and minute with a suffix; two different tornadoes that began in the same minute share the number and differ only in the suffix, so we keep them apart and join each state segment to the one multi-state record it belongs to.

• Counties: a tornado counts once for each county the Storm Prediction Center lists for it. Retired county codes are moved to the county that holds them now, such as Dade to Miami-Dade. Where the list cannot place a tornado, or never names a county at all, as with some independent cities in Virginia, the recorded path’s start and end points decide.
• Cities: a tornado counts when its straight-line path from start to end point passes within 25 miles of the city’s interior point, sampled at nine points along the path.
• Casualties: injuries and deaths in the tornado tables are for the whole path, which can extend beyond the county. Deaths recorded within the county come from the Storm Events Database.
• Ratings: the Fujita scale before February 2007 and the Enhanced Fujita scale after. Tornadoes without a rating are shown as F-U or EF-U.

As a check, the county tornado count from the Storm Prediction Center is compared with the count of tornado records in the Storm Events Database for our reference counties; the two agree within 15 percent.

Tropical cyclones

Source: the National Hurricane Center’s HURDAT2 best-track data for the Atlantic (1851–2025) and the Northeast and Central Pacific (1949–2025). HURDAT2 gives a storm’s position, maximum sustained wind and status every six hours. We interpolate each track linearly to hourly positions and find each storm’s closest approach to every place.

• Counties: distance is measured to the county boundary, so a storm that crosses the county is at zero miles. Measuring to a centre point would understate storms that strike a coastal county with an inland centre.
• Cities: distance is measured to the city’s interior point.
• Direct and near: a storm is counted as passing within 50 miles (direct) or within 100 miles (near).
• Strength: the category is assigned from the interpolated wind at the closest point, using the Saffir-Simpson scale. A storm is counted only if it was a tropical or subtropical storm or hurricane (34 knots or more) at that point. Post-tropical remnants, waves and lows with gale-force winds are listed but not counted.

Early records miss storms that never reached land or a shipping lane, so counts before about 1900 are low.

Wildfire

Source: Monitoring Trends in Burn Severity, a joint program of the U.S. Geological Survey and the U.S. Forest Service, which maps the perimeter of every large fire from satellite imagery since 1984: 1,000 acres or more in the western United States and 500 acres or more in the east. We include wildfires, wildland fire use and fires of unrecorded type, and exclude prescribed burns.

For counties, we intersect each perimeter with the county boundary and report the acres inside the county as a geodesic area, so the figure is correct anywhere in the country. For cities, a fire counts when its perimeter comes within 25 miles of the city’s interior point.

Storm events

Source: NOAA’s Storm Events Database, 1950 to the latest monthly release. Before 1996 the database records only tornadoes, thunderstorm wind and hail; from 1996 it covers 48 event types, which we group into severe weather, flood, winter, heat, cold, tropical, drought and other. Types the groups would hide are also listed on their own: rip currents, high surf and sneaker waves, storm surge and tide, lightning, dust storms and dust devils, dense fog and smoke, debris flows, avalanches, seiches and tsunamis.

Events are filed either by county or by National Weather Service forecast zone. County events map directly. Zone events are mapped to counties through the Weather Service’s zone–county correlation file when the zone’s name still matches; failing that, by a current zone of the same name, by a county of the same name, or by a zone name that is a county name with a direction such as “Coastal Walton”. An event in a zone that spans several counties counts in each of them. About 9 percent of zone-based events belong to zones that have since been renumbered or renamed and cannot be placed with confidence; they are left out rather than guessed.

Damage figures are as reported at the time and are not adjusted for inflation. The narratives quoted on pages are written by the National Weather Service.

Flood insurance claims

Source: FEMA’s National Flood Insurance Program redacted claims (version 3), more than 2.7 million claims since 1978. We add building, contents and increased-cost-of-compliance payments for each claim and group claims by county, five-digit ZIP code and year of loss.

FEMA’s terms forbid re-identifying policyholders. We never show claims below the ZIP code level, and we do not show any ZIP code or year with fewer than five claims separately. These figures count insured losses only: many flooded properties carry no federal flood insurance.

This product uses the Federal Emergency Management Agency’s OpenFEMA API, but is not endorsed by FEMA. The Federal Government or FEMA cannot vouch for the data or analyses derived from these data after the data have been retrieved from the Agency’s website(s).

River crests

Source: the National Weather Service’s National Water Prediction Service, which publishes for each river gauge its flood stages (action, minor, moderate and major) and two lists kept by the local forecast office: the highest crests on record and recent crests. We merge the two lists for each gauge. They often carry the same flood a day apart, one in local time and one in UTC, so crests at one gauge within three days of a higher crest are treated as the same flood.

• Counties: a gauge belongs to the county that contains it, or to the nearest county within two miles when its point falls just outside a coastline-clipped boundary. Only gauges with a flood stage are shown; reservoir pools and gauges without one cannot be read as floods.
• Cities: the gauges of the county that contains the city.
• Categories: each crest is placed against the gauge’s flood stages as they stand today. Levees, dams and channel changes since a crest may mean the same stage would not flood the same ground now. Crests flagged as measured on an older datum are listed but left out of rankings.

The lists are the crests each office chose to record, not every flood at the gauge.

High tide flooding

Source: NOAA’s Center for Operational Oceanographic Products and Services, which counts for each long-term tide gauge the days on which water reached its minor, moderate and major high tide flooding thresholds, by meteorological year (May to April). These are the counts in NOAA’s annual high tide flooding reports. The thresholds are NOAA’s own, set from each gauge’s tide range so that they mean the same thing on every coast; they are not National Weather Service flood stages.

A gauge belongs to the county that contains it, or, because gauges stand on piers and jetties, to the nearest county within three miles. Where a county has several gauges, the chart shows the one with the longest complete record. Years with less than 80 percent of data are left out. Great Lakes gauges have no tides and no counts. City pages use the county that contains the city.

Drought

Source: NOAA’s nClimDiv county Palmer Drought Severity Index, computed monthly from 1895 for every county in the lower 48 states from county-average temperature and precipitation. The index is relative to each county’s own climate: zero is normal, −3 or lower is severe drought and −4 or lower extreme drought. We count the months at or below each threshold, find the longest unbroken run at −3 or lower, and report the lowest month. Alaska and Hawaii are not covered. City pages use the county that contains the city.

The index measures the long-term balance of moisture. It does not measure water supply, reservoir levels or crop losses.

The national figures on the drought page are the share of lower-48 counties at −3 or lower each month, averaged over the twelve months for a year. Every county counts once, whatever its size.

Wildfire smoke

Source: NOAA’s Hazard Mapping System, whose analysts trace the smoke they see in GOES and polar-orbiting satellite imagery each day as polygons classed light, medium or heavy, from 2005. A smoke day for a county is a date on which a polygon covers the county’s population centre; for a city, its interior point.

We count medium and heavy smoke only, from 2011, the first year every polygon carries a density class. Since 2018 analysts have often drawn light smoke as one polygon over most of the continent (833 polygons wider than 40 degrees of longitude in 2023, against 40 in 2008), so counts of light smoke reflect drawing practice as much as smoke and are not compared across years. The polygons show smoke anywhere in the air above a place, not air quality at the ground, and smoke under cloud is missed.

Earthquakes

Source: the U.S. Geological Survey’s Advanced National Seismic System Comprehensive Catalog (ComCat): every earthquake of magnitude 2.5 or more since 1900 in and around the 50 states. Quarry blasts, explosions and other non-earthquake events are excluded.

• Counties: an earthquake is “inside the county” when its epicentre falls within the county boundary, and “within 50 miles” when the epicentre is within 50 miles of the boundary, measured like tropical cyclone distances.
• Cities: within 50 miles of the city’s interior point.
• Counts: earthquakes of magnitude 2.5 and 3 or more are counted from 1973, when the modern U.S. networks began; earthquakes of magnitude 4 and 5 or more are counted from 1900.
• Dates: every date and clock time is local to the county nearest the epicentre, as the earthquake is known there; the record page also gives UTC. Loma Prieta struck on 17 October 1989 at 5:04 pm, already 18 October in UTC.
• Record pages: each earthquake of magnitude 5 or more within 50 miles of a county in the lower 48 or Hawaii, and of magnitude 6.5 or more in Alaska, has a page. Its home county is the county the epicentre is in, or the nearest within 50 miles; its rank is by magnitude among every earthquake in that state’s record since 1900.
• Aftershocks: earthquakes of magnitude 2.5 or more in the 30 days after, within the surface rupture length expected for the main shock’s magnitude (Wells and Coppersmith, 1994), at least 20 and at most 600 miles. Foreshocks use the same distance and the 7 days before.

Small earthquakes were under-recorded before the 1970s and are still missed in remote areas. Magnitudes are as the catalogue reports them and mix scales (moment, local and body-wave), so two earthquakes of the same listed magnitude may not have released the same energy.

Tsunamis and volcanoes

Tsunamis: NOAA’s Global Historical Tsunami Database (NCEI and the World Data Service) records the places where each tsunami’s water was observed or measured and the height it reached. We keep the records of tsunamis NOAA rates probable or definite and not flagged doubtful, and assign each record to the county that contains it or, since coastal points often fall just offshore of Census shapes, to the nearest county within five miles. Heights mix eyewitness accounts and tide gauge readings, and older tsunamis are less completely recorded. City pages use the county that contains the city.

Volcanoes: the U.S. Geological Survey’s list of 161 U.S. volcanoes, with the threat category from its 2018 National Volcanic Threat Assessment (very high, high, moderate, low, very low). The category ranks a volcano by how hazardous its past eruptions were and how many people and how much infrastructure lie near it; it is not a forecast. A place lists every volcano within 100 miles of the county boundary or the city centre.

Landslides

Source: the U.S. Geological Survey’s Landslide Inventories across the United States, version 3.0 (2025), which joins 56 inventories made by federal and state agencies into one set of points and mapped areas. Each landslide is placed in the county that contains its point (for a mapped area, its centre). We keep landslides the compilation rates probable or better and leave out those rated only possible.

The inventories were made for different purposes and cover different areas: a state survey may have mapped every slope in some counties and none in others. The count in a county therefore reflects how closely it has been surveyed as much as how often its slopes fail, so landslide counts are not ranked, and each page names the inventories behind its count. Deaths come from the inventories; a slide recorded by two inventories on the same day within five miles is counted once, with the higher figure.

Disaster declarations

Source: FEMA’s Disaster Declarations Summaries, every major disaster, emergency and fire management declaration since 1953. A declaration is listed for each county it designated. Statewide declarations are counted on state pages but not listed on county pages.

Weather records

Source: NOAA’s Global Historical Climatology Network Daily, which holds daily maximum and minimum temperature, precipitation and snowfall for thousands of U.S. stations. Each place gets one station, chosen by a published rule:

1. Consider stations within 40 miles of the county’s population centre (or the city’s interior point) that still report temperature and precipitation.
2. Estimate the place’s ground elevation from the seven nearest weather and rain-gauge stations, and keep only candidates within 200 metres of it when any exist, so a valley city is not measured on a nearby mountain.
3. Require 30 or more years of temperature records, at least 85 percent of days complete in the last 30 years, and data within the past 400 days.
4. Score the rest: years of record count up to 60; first-order stations, usually at airports, earn a bonus for their quality control; each mile from the population centre and each metre of elevation difference costs points.
5. Use the highest scorer, and name it, its distance and its completeness on the page.

Values flagged by NOAA’s quality checks are excluded. Annual figures use only years with at least 329 days of data, and snowfall seasons, which run July to June, need 90 percent of winter days observed. A station whose snowiest day exceeds its snowiest complete season shows no season figures, because the snowy seasons were the ones left out. Station records can differ from a city’s official threaded record, which joins several stations over time. Extremes are cross-checked against the Applied Climate Information System for the reference places.

Ranks and comparisons

Ranks compare each county with the other counties in its state and with all 3,143 counties, highest first; ties share a rank. “The typical county” means the median county in the state. State totals count unique events, so a tornado that crossed three counties counts once for the state.

The hazard fingerprint

The eight-spoke figure on each page shows where a county stands nationally on eight hazards. Each spoke is the county’s percentile among U.S. counties where that hazard has occurred at all, so a county with a handful of storm remnants does not read as mid-scale just because most counties have none. A county where the hazard has never been recorded sits at the centre.

Spoke	Measure
Tornado	Tornadoes since 1950 per 1,000 square miles of land
Tropical	Tropical storms and hurricanes within 50 miles since 1900
Wildfire	Share of land burned by large fires since 1984
Flood	Flood events since 1996 per 1,000 square miles, plus flood insurance claims per 1,000 residents
Hail and wind	Severe weather events since 1996 per 1,000 square miles
Heat, cold, winter	Heat, cold and winter weather events in the Storm Events Database since 1996

City pages show the fingerprint of the county that contains the city.

Record pages and rankings

Besides places, the site has a page for every tropical cyclone that passed within 100 miles of a U.S. county, every tornado rated F4/EF4 or stronger or that killed five or more people, every tornado outbreak day (40 or more tornadoes on one Storm Prediction Center date, or 6 or more rated F3/EF3 or stronger), every severe wind day (300 or more damaging wind reports on one date, or 15 or more of 75 knots or more), every National Weather Service river or coastal gauge that lists a crest above its flood stage, every wildfire of 25,000 acres or more, every year since 1851 and every day of the calendar. Tornado pages quote the National Weather Service’s own account where its Storm Events record can be matched to the tornado by date, county and rating. A tornado path follows the county-by-county segments of the Storm Events Database where they join up cleanly, and is otherwise drawn as a straight line between the recorded start and end points.

Rankings list the top 100 counties or cities on one measure, taken from the place pages and sharing their rules. Density rankings exclude counties under 100 square miles, per-resident rankings exclude counties under 5,000 people, and city weather rankings include only cities of 50,000 or more.

Event replays

Each tornado, tropical cyclone and wildfire page opens with a replay of the event, as does each earthquake page with ten or more aftershocks, the drought page and the page of every year in which a quarter or more of counties were in severe drought in its worst month. Replays are historical reconstructions from the same records as the rest of the page; they are not forecasts and add nothing the records do not contain.

Tropical cyclones move between the National Hurricane Center’s six-hourly HURDAT2 positions, interpolated in a straight line. Wind speed, pressure, category and the extent of 34-, 50- and 64-knot winds (recorded from 2004) come from the same fixes. A county lights at the hour the storm passed closest to its boundary. Times are UTC. The relief beneath the map is Natural Earth’s 1:10 million shaded relief, which is public domain.

Tornadoes follow the recorded path at the pace of the times the National Weather Service logged for each county segment, which it has done for nearly every tornado since 1996. Earlier tornadoes move at an even pace because only the touchdown time survives. The Storm Prediction Center records every tornado in Central Standard Time; replays show the county’s own time zone. Towns are today’s Census places with 2025 population estimates, so a town shown may not have existed at the time.

Wildfires are shown with the burn-severity classes mapped by Monitoring Trends in Burn Severity from Landsat images taken before and after the fire. The order they burned in comes from NASA FIRMS active-fire detections (MODIS from November 2000, VIIRS from 2012) that fall inside the final perimeter: each 30 to 120 metre pixel takes the time of its nearest detection, so the sequence is observed by satellite but interpolated between passes, which come roughly every twelve hours; playback runs in proportion to real time until 98% of the area has burned. Fires before November 2000, and the few later fires with no detections inside the perimeter, have no recorded order: their severity map is revealed outward from the reported origin when it lies within three miles of the burned area, otherwise from the centre, and the page says so. Comparison areas use Census land area.

Earthquakes show every event the USGS catalogue holds within the sequence radius, from seven days before the main shock to 30 days after, at its logged time and position, sized by magnitude. The clock is not even: the week before and the month after each run on a logarithmic scale, so the first minutes after the main shock take as long as the last weeks, which is how an aftershock sequence (whose rate falls roughly as one over time) can be seen at all. Between them, the main shock’s two shaking fronts are drawn spreading at 6.0 and 3.5 kilometres per second, typical speeds for the faster and slower seismic waves in the crust; they show how quickly shaking reached each place, not a measurement, and counties within 50 miles and towns light as the slower front reaches them. Times are local to the epicentre.

Tornado outbreaks replay every tornado of the day from its logged touchdown time. The Storm Prediction Center records when a tornado touched down but not when it lifted, so a path is drawn at a typical forward speed of 35 mph unless the tornado has its own record page with Storm Events segment times, in which case those times set the pace. Counties light as a path reaches them. A day is the Storm Prediction Center’s calendar date in Central Standard Time, shown in Central Daylight Time when daylight time applied (from 1967; earlier days are shown in standard time). Playback slows between tornadoes so an overnight lull does not crowd out the afternoon; the timeline marks the real hours.

Severe wind days replay every damaging thunderstorm wind report of the day at its logged time, sized by speed, filled where the speed was measured and drawn as a ring where it was estimated from damage, with the day’s hail reports and tornadoes. A report is one damage or measurement report, so one gust front can produce hundreds; reporting has grown with population and spotter networks, so counts from different decades are not comparable. The Storm Prediction Center has flagged measured against estimated speeds since 2006. A hail size of 9.99 inches in a few early rows is a fill value and is treated as unknown. So is any size above the certified national record on its date (5.7 inches before June 22, 2003, 7.0 before July 23, 2010, and 8.0 since): the report still counts, without its size.

Flood gauge pages list every crest the local National Weather Service office keeps for a gauge (the highest on record and the recent crests, merged, with crests within three days of a higher one counted once), against the gauge’s flood stages as set today. Gauges sharing a river name are treated as one river while each lies within 80 miles of another; for a gauge’s worst crests, the page lists the other gauges on that river that crested above flood stage within seven days. Crests measured on an earlier gauge datum are listed apart and never ranked. A gauge page is indexed when it lists five or more crests and at least one at major flood stage.

Drought replays colour every lower-48 county by its monthly Palmer index class: extreme (−4 or lower), severe (−3 to −4), moderate (−2 to −3), near normal, and very wet (+3 or higher). Each frame is one month’s published reading; counties ease between classes for legibility only. The drought page replays the ten calendar years with the highest average share of counties in severe drought; year pages replay July of the year before through December.

Satellite images and burn-severity maps. Storm pages since 2000 show the storm in true colour on the last daytime satellite pass before its strongest U.S. landfall (or, without a landfall, before its closest approach), from NASA’s Global Imagery Browse Services: VIIRS from late 2015, MODIS before that, with gaps between orbits filled from a second satellite and the leading sensor named in the caption. Wildfire pages show Monitoring Trends in Burn Severity classes inside the final perimeter, from the change in a Landsat burn index between images before and after the fire; acres per class are each class’s share of the mapped area. Both are U.S. government works.

Which pages exist

Every county and every city of 10,000 or more has a page. A page is offered to search engines only when it has something to say: at least ten recorded tornadoes, a tropical cyclone within 100 miles, a large wildfire or 50 storm events, and a weather station that meets the rules above. Pages that fall short remain available to anyone who follows a link to them.

Known limits

• Weak tornadoes were under-recorded before Doppler radar coverage became national in the early 1990s.
• Tropical cyclone records before about 1900 are incomplete at sea.
• Satellite fire mapping begins in 1984 and omits fires below the size thresholds.
• Storm Events coverage before 1996 is limited to three event types, and some zone-based events cannot yet be placed.
• Flood insurance claims reflect insured properties only.
• A single station’s record is not the same as a region’s climate.
• River crest lists are the crests each Weather Service office recorded, not every flood, and are rated against today’s flood stages.
• Smoke counts cover medium and heavy smoke seen from space since 2011, not ground-level air quality.
• Small earthquakes before the 1970s and in remote areas are missing from the catalogue.
• Landslide counts depend on where inventories have been made; an unmapped county can show none.

Found something wrong? Tell us through the corrections page.