What happens after you report

Your report does not stop at the forecast office. It scores the warning, enters the national record, and becomes part of the climatology the next generation of forecasts is built on.

The path your report takes

1. You, with a location, an occurrence time, and whether the number was measured or estimated.
2. The WFO, which may act on it immediately: a new warning, an upgraded tag, an SVS update.
3. The Local Storm Report, the office's published record of what was reported and by whom.
4. SPC storm reports, which aggregate the day nationally.
5. Storm Data, the archival record, prepared under NWS Directive 10-1605.

Only the second step is about the storm you are standing under. Everything after it is about every storm afterwards.

How the office is scored

POD

Probability of detection. Of the events that happened, the fraction that were warned. Misses lower it.

FAR

False alarm ratio. Of the warnings issued, the fraction that did not verify. Unverified warnings raise it.

Lead time

How long between the warning being issued and the event reaching the location. Counted two ways, which is why you will see two very different numbers for it.

All three are computed from reports. To the verification system, an event nobody reported is an event that did not happen. An unreported tornado is not counted as a miss, and an unreported hit is not counted as a hit.

National tornado warning performance over thirty years looks like this. Brooks and Correia tracked all three metrics together, which is the only way any of them means anything.

The same care applies to the other two. A false alarm ratio near 0.75 is the number people reach for when they say warnings cannot be trusted, and it means three in four tornado warnings are not followed by a verified tornado. Read it next to the POD: the system is tuned to miss fewer events at the cost of warning more often, and reports are the only thing that shows whether the trade is set right.

Why a late report still counts

A report that arrives after the storm has passed cannot change the warning, so people skip it. Send it anyway. It still verifies the warning, still enters the Local Storm Report, and still reaches Storm Data, where it becomes part of the record used to set thresholds, tune guidance and argue for resources years later.

Known holes in the record

• Population and distance. Reports thin out with distance from people. Potvin et al. found tornado reporting falls off sharply away from populated areas, so the archive understates rural events.
• Nighttime. What cannot be seen is not reported, and a nocturnal tornado is both more dangerous and less likely to enter the record.
• Hail size quantization. Estimated sizes snap to familiar reference objects instead of spreading smoothly, so the record clusters on a dozen values and is thin between them. Module 5 lesson 3 takes this apart properly.
• Nocturnal under-collection, measured. Wendt and Jirak compared human reports against the radar hail estimate and found about 30% of significant hail reports occur at night against about 40% of MESH: radar, which does not sleep, assigns more of its hail to the dark than people do.

These are not reasons to distrust the record. They are the reason a careful report from an underrepresented place, at night, with a measured rather than estimated size, is worth more than three routine ones.

What WxAlerts does with it

A report submitted here is checked for plausibility against the radar data for that place and time, then published with a confidence tier rather than accepted silently or discarded. The tiering and the validation method are the subject of Module 10. The principle here: a disagreement between your report and a radar product is recorded, not resolved by deleting your report.

Check yourself

A forecast office issues 100 tornado warnings and 30 of them verify. What is the false alarm ratio, and what does it not tell you?

1. 70%, and it says nothing about how many tornadoes went unwarned
2. 30%, and it says nothing about lead time
3. 70%, and it means 70% of tornadoes were missed
4. It cannot be computed without knowing the lead times

Sources

1. NWS Directive 10-1605, Storm Data preparation
2. Brooks & Correia 2018, Weather and Forecasting: "Long-Term Performance Metrics for National Weather Service Tornado Warnings"
3. Potvin et al. 2019, Wea. Forecasting 34(1): tornado reporting falls off with distance from population
4. Wendt & Jirak 2021, Wea. Forecasting 36(2), 645–659: an hourly climatology of MRMS MESH-diagnosed hail against Storm Data reports
5. NWS Weather Spotter's Field Guide (YPA-201154, 2011), "Spotter Reporting Procedures"