Capstone: Hurricane Ivan, 15 to 16 September 2004
- View
Everything in this module, applied to one documented event where almost nothing was visible and the warnings went out anyway.
In this lesson you will
- Sequence rotation cues chronologically from imagery and radar for a real event.
- Make reportable and not-reportable calls carrying time, location and measured-or-estimated metadata.
- Justify each call against the NWS criteria stated earlier in the module.
The event
Ivan came ashore near the Alabama and Florida line early on 16 September 2004. The tornado outbreak in its rainbands had already begun the day before and ran across the south-east.
Approximately 117 tornadoes were reported associated with Ivan across the southeast United States. Eight people were killed and 17 were injured by tornadoes.
Two of those deaths and much of the damage in this office's area came from a pair of F2s on 15 September: Panama City Beach in Bay County, one death and seven injuries, and Blountstown in Calhoun County, four deaths and one injury. The Tallahassee forecast area counted 24 tornadoes in all: 14 F0, 8 F1 and 2 F2.
A note on the count, because you will meet a different one: NCDC catalogs the wider continuous outbreak at 120 tornadoes, passing Hurricane Beulah's 1967 record of 117. This lesson uses the NWS and AMS figure of approximately 117 throughout so the numbers here agree with each other. Both are real counts of slightly different things.
What the office was working with
This is the part worth studying. NWS Tallahassee monitored more than fifty identified mesocyclones over the event and issued 130 tornado warnings.
| NWS Tallahassee, Ivan | |
|---|---|
| Tornado warnings issued | 130 |
| Verified | 20 |
| Not verified | 110 |
| Tornado events | 23 (21 warned, 2 unwarned) |
| POD | 0.913 |
| FAR | 0.846 |
| CSI | 0.152 |
| Average lead time | 11.2 min |
| Events with lead time above zero | 78.3% |
Watson, Jamski, Turnage, Bowen & Kelley, NWS Tallahassee. These are the office's own published numbers for its own performance.
Module 1 lesson 7 gave national tornado warning performance: a POD around 0.60 to 0.64 and a FAR around 0.70 to 0.75. Ivan produced a POD of 0.913 and a FAR of 0.846. The office caught almost everything and was wrong most of the time, and both of those facts are the same decision. Faced with fifty shallow, rain-wrapped mesocyclones and no ground truth, they warned. Given the alternative, that is the right call, and the FAR is what it costs.
Why nobody could see anything
Every difficulty from lesson 8 was present at once. The storms were mini-supercells in tropical rainbands with shallow, small circulations. They were rain-wrapped. Much of the activity ran through the night. And the whole region was already inside a landfalling hurricane, so the conventional advice to find a safe vantage with a clear view to the south-west was not available to anybody.
A spotter's value in that event was not in confirming funnels. It was in damage reports, timings and locations after the fact, which is what turned 20 of those 130 warnings into verified events and left a record precise enough that we can read the office's performance twenty years later.
What a report needed to carry
WHO you are: trained spotter; WHAT you have witnessed; WHEN the event occurred: NOT when you make your report; WHERE the event occurred, using well known roads or landmarks.
The WHEN clause is the one that decides whether a report verifies a warning. A damage report timed to when you called it in rather than when the damage happened can fall outside the warning polygon's valid window and verify nothing. In a nocturnal, rain-wrapped outbreak where after-the-fact damage reports are most of the ground truth available, that detail is doing more work than any funnel identification.
Plains cues do not transfer to a humid climate unchanged. Smaller, shallower, rain-wrapped, often nocturnal, frequently embedded in a line or a rainband. What can you still report when you cannot see the storm? Everything in lesson 8's last section, and the four fields above, on time.
Check yourself
NWS Tallahassee finished the Ivan outbreak with a POD of 0.913 and a FAR of 0.846. What does that pair of numbers describe?
- An office that warned aggressively on rotation it could not confirm, catching almost every tornado at the cost of many unverified warnings
- An office that under-warned, missing most of the tornadoes
- A radar failure, since a high FAR indicates faulty detection
- A verification error, since POD and FAR cannot both be high
Hint
POD is measured against events. FAR is measured against warnings. They can move independently.
Answer
An office that warned aggressively on rotation it could not confirm, catching almost every tornado at the cost of many unverified warnings
POD 0.913 means they warned 21 of 23 tornado events. FAR 0.846 means 110 of 130 warnings did not verify. Both come from the same decision to warn on more than fifty shallow, rain-wrapped mesocyclones with almost no ground truth available. In an environment where a spotter cannot see the circulation, that trade is deliberate rather than accidental.
Sources
- Watson, Jamski, Turnage, Bowen & Kelley, NOAA/NWS Tallahassee: the tornado outbreak across the north Florida panhandle in association with Hurricane Ivan (AMS 32nd Radar Conference, paper 10R.3)
- NWS Weather Spotter's Field Guide (YPA-201154, June 2011)
- Edwards 2012, Electronic J. Severe Storms Meteor. 7(6), 1–61: tropical cyclone tornadoes, a review of knowledge in research and prediction
- NSSL Severe Weather 101, tornado basics and tornado types