Why Storm Data Differs Between Government Sources
By StormAuditor Editorial Team · Published 2026-08-26 · Updated 2026-08-26
Based on public NOAA and National Weather Service records, published under StormAuditor's methodology and data-source standards. Radar-derived hail sizes are estimates unless attributed to a ground report.
NWS offices, SPC preliminary reports, the NCEI Storm Events Database, and radar mosaics each answer a different question about the same storm. This guide explains why their numbers diverge and how to read them together for claim research.
The National Weather Service calls hail severe at 1.00 inch. A wind gust earns the same label at 58 mph. Those thresholds are stable across the country. The records built around them are not. Pull one storm from three federal databases and you can get three different pictures of it.
This is not sloppiness. Each source answers a different question, on a different clock, with different tools. Once you know which question each one answers, the disagreements start to make sense. They can even work in your favor during claim research.
The main government sources, briefly
Four public sources cover most storm questions a property owner will ever ask.
Local NWS forecast offices issue warnings and event summaries in real time through weather.gov. Their job is protecting life and property, not building a tidy archive.
SPC preliminary storm reports are same-day logs of hail, wind, and tornado reports. Spotters, emergency managers, and the public phone them in. Speed matters more than polish here.
The NCEI Storm Events Database is the reviewed record. NWS staff edit, merge, and finalize entries weeks or months after the storm. Researchers treat it as the closest thing to an official history.
Radar mosaics such as MRMS estimate conditions everywhere, not just where someone was watching. Hail sizes from radar are always radar-estimated values, inferred from the strength of the echo aloft.
The Iowa Environmental Mesonet archives much of this material and makes it searchable long after the fact.
Why the same storm reads differently
People measure points; radar estimates areas
A spotter report is one observation at one spot at one moment. Radar covers the whole map but senses the storm aloft, then estimates what reached the ground. A rural hailstorm may show a strong radar signature and zero human reports. Nobody stood under it with a ruler. Both records are honest. They just sample the world differently.
Preliminary and final records diverge on purpose
The SPC log is raw intake. Duplicate reports of one hail swath can appear as several entries. During review, NCEI staff may merge duplicates, adjust a size, shift a time, or drop a report entirely. So a storm can shrink, grow, or move slightly between the preliminary and final record. Neither version is wrong. They are snapshots taken at different stages of quality control.
Clocks and time zones split events
Government archives often log events in UTC. A hailstorm at 10 p.m. local time can land on the next calendar day in one database and the same day in another. Anyone searching a single date can miss the event entirely. Checking the day before and after is a cheap habit that catches this.
Thresholds filter what gets recorded
Databases built around the NWS severe criteria emphasize hail of 1.00 inch and gusts of 58 mph or more. Smaller hail and weaker gusts still occur, and can still matter to a roof, but they draw less reporting attention. Absence from a severe-weather log is not the same as absence of weather.
How StormAuditor estimates it
StormAuditor's SAWE-2 wind engine and SAHE-2 hail engine blend these government sources into one address-level estimate. Results are flagged against screening thresholds of 45 mph wind and radar-estimated 0.75 inch hail. Every output is an estimate of the public weather record, sourced and dated. The full approach lives at wind methodology and hail methodology.
Practical guidance for claim researchers
Treat disagreement between sources as information, not as a defect. A radar-estimated size with no ground report suggests a sparsely observed area. A ground report with a modest radar signature suggests a localized cell.
A few habits help:
- Search a window of dates, not a single date, to absorb UTC shifts.
- Read the preliminary and final records side by side and note which is which.
- Record the source and access date for every figure you keep.
- Anchor the timeline first; date-of-loss weather research walks through that process.
- Keep the underlying feeds straight with the data sources reference.
Adjusters and engineers see these same databases. Showing that you understand why they differ tends to make a conversation easier, not harder.
Limitations
No government database observes every storm, and no blend of them can. Radar-estimated hail sizes are inferences, not measurements. Human reports cluster where people live and thin out where they do not. Reviewed records change as corrections arrive. StormAuditor's estimates inherit all of these limits, and none of them establish what happened to a specific structure. A qualified inspection does that job. The honest boundaries are laid out at limitations.
FAQ
Why does one database show hail on my street and another show nothing?
The sources sample differently. Radar estimates cover every address; human reports exist only where someone reported. A gap in one is common and expected.
Which government storm database is the most reliable?
None is best for everything. SPC preliminary reports are fastest. The NCEI Storm Events Database is the reviewed record. Radar mosaics fill the spatial gaps between reports. Strong research cites more than one.
Why did the hail size change between the preliminary and final report?
Reviewers at NCEI can merge duplicate reports and adjust sizes during quality control. The final entry reflects that review, so differences from the same-day log are normal.
My storm happened at night — why is it filed under the wrong date?
Many archives log times in UTC. A late-evening storm in the United States often rolls into the next UTC day. Search adjacent dates before concluding nothing happened.
Can differing government records still support an insurance claim date?
They can support an estimated date of loss when read together and cited honestly. Present each figure with its source and label radar-estimated values as estimates. Coverage decisions rest with the insurer, not with any weather record.