Weather Banter - View Single Post - Tornadoes: Accuracy of warnings vs actual touchdowns?? help

Harold Brooks · January 7th 06, 01:25 AM posted to sci.geo.meteorology

In article ,
says...
Harold Brooks wrote:
In article ,
says...

On Sun, 1 Jan 2006 16:24:01 -0600, Harold Brooks
wrote:

In article ,
says...

I have watched the refreshing screen the NWS has for all the watches
and warnings for severe storms and tornadoes. I have noticed that at
times they will display, over a period of many hours, many dozen
tornado warnings as red dots. However, when I check the pages with the
actual storm reports on the following days, it shows only a few. In
fact, in appears that of all those tornado warning dots, that perhaps
10 and at most 15% of those tornado warnings actually formed a
tornado.

Now, if this is in fact true, then what about those tornadoes that are
confirmed where a warning was _never_ even issued?

What is the accuracy rate of [issued tornado warning vs confirmed
tornadoes]?

Also, what is the percentage of times when an actual tornado hits the
ground _but_ was never associated with a warning?

Here are the annual tornado warning statistics for the US from 1986-
1994. POD is the probability that a tornado will have a warned county
associated with it. FAR is the false alarm ratio, the fraction of the
warned counties for which there is no tornado.

Year POD FAR
1986 0.338 0.813
1987 0.289 0.813
1988 0.290 0.813
1989 0.358 0.801
1990 0.438 0.715
1991 0.412 0.760
1992 0.453 0.695
1993 0.429 0.730
1994 0.459 0.743
1995 0.599 0.748
1996 0.593 0.770
1997 0.588 0.775
1998 0.652 0.802
1999 0.702 0.720
2000 0.648 0.762
2001 0.692 0.712
2002 0.757 0.750
2003 0.792 0.748
2004 0.759 0.743

If we limit attention to the strongest tornadoes (F2 or greater, roughly
10% of the tornadoes), the POD is a little over 0.8 for 2003 and 2004.

Let me ask you then: Does this mean 8% in 2004 were FAR and 7.59% were
warned counties?
Could you explain a little more in laymen terms. I am no scientist by
any stretch..

For the POD, start by looking at all tornadoes. Compute the fraction of
them with warnings. In 2004, it was 0.759 (75.9%.) For the strong
tornadoes, it was a litle over 0.8 (80%.)

For the FAR, start by looking at the warnings. Compute the fraction
with no tornado. In the 2004, it was 0.743 (74.3%.)

Is there any geographical bias to FAR or POD? It seems like
you'd be far more likely to have a POD where more tornadoes
are radar-detected -- which should be where there is little
population. I don't recall seeing a state-by-state mapping
of these two parameters, however. Is it online somewhere?

It's really complicated. Separating out the meteorological and non-
meteorological factors is almost impossible. Effectively, you can
improve both FAR and POD by getting smarter or by working harder at
collecting verification information. That's on top of detection,
population, and road network issues.

Harold
--
Harold Brooks
Head, Mesoscale Applications Group
NOAA/National Severe Storms Laboratory