Severe thunderstorm(+) Climatology and Trends

Loading…

Climatology

drag · scroll to zoom · click a box to select it

Trend

Trends are inherently far noisier: focus on coherent spatial patterns, not the value in any single gridbox.

both maps share one view

Methods

Source. NOAA/NCEI Storm Events Database — every hail, tornado and thunderstorm-wind report. Only county-type records (CZ_TYPE = C) are used. Puerto Rico lies outside this grid.
Position each report. Its own BEGIN_LAT/BEGIN_LON where present; otherwise the centre of the county that reported it (the average of its simplified boundary points, which can differ slightly from the true centroid for oddly shaped counties). The fallback matters almost entirely in 1993–1995, when hail and wind reports carry no coordinates at all.
Bin into a box. A 2° × 2° grid over the lower 48 (24–50°N, 126–66°W; 390 boxes, about 260 of which ever see a hail, tornado or wind report).
Apply the threshold. Each report is scored against the selected size or intensity cut. Thresholds are cumulative — a 2″ hailstone also counts as a 1″ one.
Collapse to hazard days. Count distinct calendar dates on which a box saw at least one qualifying report. One storm producing forty reports in a box on one afternoon is one hazard day, which is what removes most duplicate-report inflation.
Store those counts per box × year × month × threshold. Everything below is computed live in the browser from that table.
Climatology map. Mean annual hazard days over the selected period. A year with no days is a real zero and is counted in the denominator — every box exists every year.
Trend map. Ordinary least squares on the annual counts, reported per decade. A box is left grey unless it clears three coverage gates: enough years carrying a day (at least 8, or 30% of the period), those years spanning at least 70% of the period, and at least 2 of them in each half.
Annual and monthly panels. The series for the single selected box, fitted with a 95% confidence band — and subject to the same three gates, so a box the map declines to fit does not get a slope here either.
Smoothing. The two maps are drawn through a light Gaussian blur, sigma 2°, on by default and switchable to Raw. On this 2° grid that is exactly one box, and the kernel is cut off at two boxes (a 5 × 5 window). The county page uses the same 2° sigma, so the two are smoothed at a matched physical scale; it bites much harder there, because counties are roughly 0.3° while this field is already 2°-averaged by construction. A single 2° box holds few enough reports that its year-to-year value is noisy, and the blur pulls a stray box towards its neighbours without moving a coherent pattern. Three things it deliberately does not do: it does not fill gaps — only boxes that have a value contribute and the weights are renormalised by what was found, so edge boxes are not dragged towards zero by the empty ocean; it does not change which boxes are drawn, since a box the trend gates declined to fit stays grey either way; and it never touches the annual or monthly panels, which always show the unsmoothed series. Because it is a neighbourhood average, the drawn value is not the box's own — hover reports both when they differ. Totals are not conserved. It also interacts with the white “exactly 0” shading: with smoothing on, a box is white only when its whole neighbourhood is zero, so the zero boundary is drawn more conservatively than it really is. Switch smoothing off to see the true edge.
Colour scales. Set from the data on screen: the 98th percentile for climatology, the 95th of absolute values for trend, symmetric about zero.