Reports under-measure, the archive has a seam in it, and the radar product everyone checks against was built to over-forecast on purpose.

In this lesson you will

  • Explain size quantization around coin and object references.
  • State that Storm Data and warning sizes underestimate maximum hail diameter, citing the field evidence.
  • Explain the nocturnal and population reporting biases.
  • Explain what MESH is, why it is not ground truth, and how this platform uses it.

Everything snaps to a coin

Hail diameters are continuous. Reports are not. A stone that is genuinely 1.15 inches gets reported as a quarter, and one at 1.6 inches becomes a golf ball, because the reference objects are what people have to hand. The result is a record that clusters on a dozen values and is empty between them.

This is not a criticism of spotters; it is what the method does. It is also fixable in one move, which lesson 2 already gave you: measure it and say you measured it. A measured 1.15 inches is worth more than an estimated quarter precisely because it does not land on the grid.

And everything lands low

Blair and colleagues went and checked. Their field campaign sampled 73 severe thunderstorms between 2011 and 2015, measuring what actually fell, and compared it to the record.

Blair et al. 2017, verbatim
Hail sizes recorded in Storm Data as well as hail size forecasts in NWS warnings frequently underestimated the maximum diameter hailfall occurring at the surface.

Companion work put the underestimate of a storm's maximum size at close to a factor of two. Supercells produced the largest stones by a wide margin, which is lesson 1's physics showing up in the archive. The record is a floor, not a measurement.

Nobody is outside at three in the morning

Two biases decide whether hail enters the record at all, and neither is about the hail. Reports need a person, so they thin out at night and away from towns, in exactly the way module 1 lesson 7 described for tornadoes.

Wendt and Jirak quantified the night half by comparing human reports against MESH, the radar hail estimate. About 30% of significant hail reports occurred at night, against about 40% of MESH. Radar, which does not sleep, finds a larger share of its hail after dark than people do. Their plain-language version: "Human observers are increasingly less likely to be available to observe severe hail after sunset." Across the whole record MESH diagnoses severe hail hours two to four times more often than Storm Data has reports.

Read that carefully, because a draft of this course got it wrong
An earlier version of this material said reports "drop 30% at night while MESH drops 40%". That is not what the paper says and it points the wrong way. The numbers are shares of each dataset that occur at night, and the finding is that MESH assigns more of its hail to the night than reports do. Same two numbers, opposite conclusion. This is what a plausible-sounding paraphrase does to a real result, and it is why every figure in this course is traced to a page.

MESH is not a measurement

MESH comes from SHI, a thermally weighted integration of the reflectivity profile above the freezing level, with POSH derived against a threshold that scales with the wet-bulb-zero height. What matters for a spotter is not the machinery but the design intention, which Witt and colleagues stated when they built it.

Witt et al. 1998, on their own algorithm
The algorithm was designed to be an overforecast such that 75% of all observed hail will fall below the radar estimate.

So MESH is roughly the diameter that three quarters of stones fall short of. A report coming in under MESH is the expected case, not a contradiction. And the skill numbers are sobering:

QuestionBest productSkill
Was there hail at all?RALA at or above 55 dBZHSS 0.42
Was it severe?MESH at or above 30 mmHSS 0.38
Was it significant severe?nothingHSS under 0.2

Ortega 2018, against nearly 40,000 field-collected hail reports. Read the last row twice: radar cannot reliably size giant hail. That is exactly the range where a careful ground measurement is worth the most.

What this platform does with all of it
Checks your report for plausibility, not exactness, and never auto-rejects it. A hail report modestly above or below MESH is expected behavior from a product built to over-forecast. A large, well-documented measurement in the range where radar has no skill is the most valuable thing you can send. Module 10 has the mechanics; this lesson is the argument.

One more seam in the record

The severe threshold moved from 0.75 to 1.00 inch on 5 January 2010, and the NWS estimated around 40% of severe thunderstorm warnings had previously been issued for hail below the new mark. That makes the date a genuine discontinuity: hail climatology before 2010 is not directly comparable to after it, and any trend drawn straight through it is measuring a paperwork change.

Check yourself

MESH over your location reads 1.75 inches. You measure a stone at 1.25 inches. What does the difference mean?

  1. Nothing is wrong. MESH is built to over-forecast, so most observed stones fall below it
  2. Your measurement is wrong, since radar samples the whole storm
  3. The radar is miscalibrated and the reading should be discarded
  4. Both are wrong, since they disagree
Hint

The algorithm's authors said what fraction of hail should fall below their estimate.

Answer

Nothing is wrong. MESH is built to over-forecast, so most observed stones fall below it

MESH was designed so that about 75% of observed hail falls below the estimate. A stone under MESH is the expected case, not a conflict, which is why a report is checked for plausibility rather than matched against radar and thrown away when it differs.

Sources

  1. Blair et al. 2017, Wea. Forecasting 32(3), 1101–1119: high-resolution hail observations and their implications for NWS warning operations
  2. Wendt & Jirak 2021, Wea. Forecasting 36(2), 645–659: an hourly climatology of operational MRMS MESH-diagnosed severe and significant hail, with comparisons to Storm Data hail reports
  3. Witt et al. 1998, Wea. Forecasting 13(2), 286–303: an enhanced hail detection algorithm for the WSR-88D
  4. Ortega 2018, Electron. J. Severe Storms Meteor. 13(1), 1–36: evaluating multi-radar multi-sensor products for surface hail-fall diagnosis
  5. NWS Service Change Notice 09-52: the one-inch hail severe criterion, operational 5 January 2010
  6. Allen & Tippett 2015, Electron. J. Severe Storms Meteor. 10(1), 1–31: the characteristics of United States hail reports 1955–2014