If Hail Falls in the US, How Big Is It?

418,487 measured hail reports, 1955–2025: two-thirds of all stones are quarter-size or smaller, 99.3% of the record lands on thirteen named sizes, a tenth of counties catch 42% of the big hail — and one South Dakota county has taken golf-ball hail every single year since 1996, the only county in America with a perfect 30-year streak.

Abstract

We count, over the complete NCEI Storm Events hail record — 418,487 county-level reports carrying a measured stone size, 1955–2025 — where hail falls in the United States, in what months, at what sizes, and how often it returns to the same county. Two-thirds of all reports are one inch (quarter size) or smaller, and 99.3% of the record lands on just thirteen named sizes, from pea to grapefruit: the spotter vocabulary of coins and balls is the measurement instrument, and sizes between the named values are almost never reported. Big hail is geographically concentrated far beyond hail generally: in the coordinate-complete era 1996–2025, the top tenth of counties collect 42% of the 20,642 reports of two-inch hail or larger, and among states with at least 1,000 severe reports, the share that reach two inches runs from 15.2% in Texas to 4.2% in Pennsylvania — the Plains do not merely get more hail, they get disproportionately bigger hail. May and June carry 45% of all reports, the peak month arriving in May in Texas and Oklahoma but June in Nebraska, Colorado, and Minnesota, and hail one inch or larger fell somewhere in the country on 220 days per year on average (192–250, 2010–2025). Repeat-strike geography is the record's sharpest structure: of 2,933 counties with any golf-ball (1.75 in) report since 1996, the median county has one in 8 of 30 years, 312 counties have one in at least 20 — and exactly one, Meade County, South Dakota, has one in all 30. Every figure is a count over measured reports; nothing is fitted, forecast, or interpolated, and the reporting-practice discontinuities of the record — including the January 2010 change in the severe-hail criterion, plainly visible in the data — are shown rather than smoothed.

Headline numbers

The size ladder is the instrument. 99.3% of 418,487 reports land on thirteen named sizes; only 191 distinct diameter values appear in 71 years. Quarter (123,481 reports) and penny (105,842) dominate; golf ball is third (69,472); baseball outnumbers the smaller tennis ball (8,067 vs 5,428) because spotters reach for the more familiar object.
Two-thirds of everything is an inch or less. Median 1.00 in; p90 golf ball; p99 teacup (3 in); 41 reports of six inches or more in 71 years. The verified ceiling is the Vivian, SD stone (Lyman County, 2010-07-23): 8.00 in, the US record. One unverifiable 9.99 in entry (Whitley, IN, 1961, no narrative) is disclosed, not deleted.
Big hail concentrates twice. Top 10% of counties hold 42.2% of ≥2 in reports; the ≥2 in share of severe reports runs 15.2% (TX) → 4.2% (PA) — 3.6× — so the Plains get disproportionately bigger hail, not just more.
The season marches north. May 23.9% + June 21.2% = 45.1% of all reports; peak month May in TX/OK/KS/GA/FL, June in NE/CO/MN. Severe hail falls somewhere in the US ~220 days a year (192–250, 2010–2025).
Meade County, SD: 30 of 30. The only county with a golf-ball report in every year 1996–2025. Nine counties scored 29 of 30 (Cherry and Lincoln NE; El Paso, Kit Carson, Weld, Logan CO; Pennington SD; Laramie WY; Gray TX) — together they draw the High Plains corridor with a statistic designed to ignore observer density. Median county with any golf-ball year: 8 of 30.
The 2010 criterion change is in the data. Sub-1-inch share of reports: 52.2% (2000s) → 26.1% (2010s) → 16.8% (2020s). All trend statements are made at ≥1 in, where the criterion never moved.

Reproducibility

scripts/extract.py — reads the live hail Eventdex spine through monitor.dex_reader → data/hail.parquet + provenance inventory.
scripts/analyze.py — every quoted number → data/results.json.
scripts/visualize.py — Figure 1 (four panels) → paper/figure1.png, www/where-hail-falls.{html,png}.
scripts/verify_draft.py — independently recomputes all 76 quoted claims from the parquet: 76/76 pass.

Data: NOAA/NCEI Storm Events (US government work, public domain), via the TerraPulse hail Eventdex (420,025 slots, 1955–2026).