US Extreme Weather and Climate Change Dashboard

NCEI Storm Events annual tornado count, by F/EF-scale rating. Unit: tornadoes per year.

Hail, tracked on the same source page, has no reliable trend window at any point in the record and is deliberately excluded from trend assessment entirely — see /severe-storms. Each tornado's F/EF rating here is the peak intensity across the reconstructed track (Storm Events records one row per county a tornado crosses, not one row per tornado; county-segment rows are merged into a single physical-tornado track before counting — see the Methodology page).

Source: NCEI Storm Events Database (reconstructed tornado tracks)

Computing trend…

Annual count of distinct calendar days with at least one tornado -- less sensitive than the raw annual count to the NEXRAD-era weak-tornado detection artifact, since a busy outbreak day still counts as one tornado-day regardless of how many weak tornadoes within it modern radar newly detects. Unit: days per year.

Distinct calendar days (local time, as Storm Events records it) with at least one reconstructed tornado track -- a standard alternative to raw annual counts in the severe-weather literature, far less sensitive to the NEXRAD-era weak-tornado detection artifact than the raw total: a busy outbreak day still counts as one tornado-day no matter how many additional weak tornadoes modern radar/spotter coverage would newly detect within it. This series uses an earlier reliable-trend window (1954+) than the raw total's 1997+: tornado days settle into a sustained ~150-210/year band decades before the raw total's NEXRAD-era jump. See the Methodology page.

Source: NCEI Storm Events Database (reconstructed tornado tracks)

Computing trend…

Mean tornado count on 'outbreak days' (calendar days with 6+ tornadoes, per Galway 1977 / Tippett, Lepore & Cohen 2016) each year -- some literature finds outbreak clustering trending upward even where raw annual counts don't. Unit: tornadoes per outbreak day.

An "outbreak day" is a calendar day with 6 or more reconstructed tornado tracks. The threshold and day-based (not rolling-window) bucketing are disclosed judgment calls -- 6 is the low end of the published range of tornado-outbreak definitions and matches Tippett, Lepore & Cohen (2016, Science), though that paper uses a 6-hour rolling start-time-gap cluster restricted to (E)F1+ tornadoes rather than calendar-day, all-tracks bucketing, so these numbers don't claim to reproduce that paper's published counts. Unlike tornado days, outbreak size reuses the raw total's 1997+ reliable-trend window: it carries essentially the same NEXRAD-era weak-tornado artifact. See the Methodology page.

Source: NCEI Storm Events Database (reconstructed tornado tracks)

Computing trend…

CONUS convective available potential energy (CAPE) from ERA5 reanalysis at 21Z daily, independent of any human tornado/hail/wind report -- an environment-based instability proxy, not a report count. This is CAPE only: it deliberately omits wind shear (the ingredient that specifically differentiates tornadic from ordinary severe environments), so read it as 'was the atmosphere unstable enough for organized severe thunderstorms,' not a tornado-probability index. Unit: J/kg.

This is CAPE (instability) only -- it deliberately does not include wind shear, the ingredient that specifically differentiates tornadic storms from ordinary severe thunderstorms/hail in the literature (e.g. the Significant Tornado Parameter). Read this series as "was the atmosphere unstable enough for organized severe thunderstorms," not as a tornado-probability index: a real, major tornado outbreak on Dec 15, 2021 had unremarkable CAPE, because it was a high-shear/low-CAPE event. See the Methodology page.

Source: ARCO-ERA5 (Google Research)

Computing trend…

Mean daily share of the CONUS reanalysis grid at or above a moderate (1,000 J/kg) or extreme (2,500 J/kg) CAPE threshold -- a spatial-coverage view, since severe weather is regional, not continent-wide, and a plain national average dilutes any real signal. Unit: % of CONUS grid.

Unlike every other severe-weather series here (which counts human reports/observations of tornadoes or hail), this series is computed purely from reanalysis-assimilated atmospheric fields -- it doesn't depend on anyone observing or reporting anything, so it isn't subject to the population-density/spotter-network/radar-deployment reporting-practice issues that affect tornado and hail counts. It has different limitations instead (reanalysis quality by era, spatial/temporal sampling choices) -- see the Methodology page.

Source: ARCO-ERA5 (Google Research)

Computing trend…

Annual count of days meeting a disclosed instability threshold -- 'widespread' days (≥10% of the CONUS grid at moderate CAPE) and 'high-CAPE' days (any grid cell at extreme CAPE) -- structurally parallel to this site's tornado-outbreak-day metric, but built from reanalysis fields rather than human reports. Unit: days per year.

The reliable-trend window reflects ERA5's documented reanalysis-quality break, not a reporting-practice artifact like the tornado record's window: ECMWF's back-extension documentation states reanalysis skill "improves dramatically following the first assimilation of satellite data from the TOVS satellites at the end of 1978." This independently matches the severe-weather-environment reanalysis literature's practical start year (NARR, and the Tippett/Gensini-Brooks STP climatologies built on it, also begin in 1979). See the Methodology page.

Source: ARCO-ERA5 (Google Research)

Computing trend…