Tornadoes: Some Hard Realities


Chuck Doswell

Last update: 01 May 2014: more minor tweaks, plus some updates

No disclaimer needed ... this is my personal account. This is an essay, not a scientific document! Comments and whatever will reach me by e-mail at This document is copyrighted [©2009 C. Doswell] ... if you have any intention to reproduce it in any form, please contact me by e-mail ... use the link above or cut and paste after substituting '@' for '#'.

I. Historical Perspectives

When Europeans first came to the U.S., they began to encounter tornadoes, almost certainly for the first time in their lives. No doubt they were astounded by the ferocity and frequency of tornadoes here in North America. Pummeling their villages and farms without warning, they inflicted terrible death tolls. As settlers pushed west, they encountered tornadoes with higher frequency and at higher intensity than in the Eastern states, but of course the sparse populations offset that to some extent.[1] With the slow but steady growth of populations at risk, by the early parts of the 21st century, tornadoes were exacting a substantial price from the still largely agrarian Midwest.

If we look at the annual tornado fatality totals over the long haul (See Fig. 1), what we see is a pretty spotty record. Almost certainly, the death tolls prior to the 20th century are underestimates. As Grazulis (1993) has noted, the native Americans, slaves, and other non-caucasians killed in tornadoes were almost never included in the fatality figures - largely as a result of racial bigotry. What's somewhat intriguing is that when the figures are smoothed to minimize the impact of the interannual variations, some pretty clear trends emerge. Note the decline in tornado fatalities beginning sometime after 1925 and continuing (with some fluctuations) for many decades.

Fig. 1 Annual tornado fatalities 1880-2008 (source: Storm Data ) rate per million people; including the raw data (black dots connected by dashed lines), and the data filtered by using a 3-point median filter and then a 5-point moving average (solid red line).  

In 1917 and again in 1927, peaks in the fatality rate were associated with several events throughout the year having 20 or more deaths. In 1925, the Tri-State tornado dominates the fatality rate "spike." In 1936, the Tupelo, MS and Gainesville, GA tornadoes dominate that peak. For the 1953 peak, totals are dominated by three tornadoes (Waco, TX; Flint, MI, and Worcester, MA). In the 1965 peak, the fatalities are associated with an outbreak (the first "Palm Sunday Outbreak") - whereas the famous "Jumbo Outbreak" is responsible for the peak in 1974. It can be seen that fatalities from single tornadoes are decreasing (as noted by Grazulis [1993]); since 1953 it usually takes one or more major outbreaks in a year to produce a "spike" in the population-normalized fatality rate. If a tornado kills 20 or more people today, it's considered a serious disaster ... such disasters were relatively commonplace at the beginning of the 20th century.

Thus, it seems that the peak threat from tornadoes was reached sometime in the 1920s, with the growth of population into the main tornado-prone areas of the United States during an era when radio communication was still relatively new technology. Many of those in harm's way perhaps didn't have electricity in their homes until the 1930s or later. It's plausible to suggest that simply being aware of the possibility of tornadoes and (later) having radio to spread the word of them when they occurred may have been responsible for some of the decline in fatalities. In some parts of the U.S., tornado shelters and other preparations for protection from tornadoes were a regular element of Midwestern agricultural life. It seems that in the era before watches and warnings, except for a few major disasters, many people began to take steps to protect themselves, especially in tornado-prone parts of the country. Disasters happened when the occasional violent tornadoes swept into populated areas. Perhaps city dwellers were more prone to complacency and little preparation for tornadoes, but rural populations had only themselves to depend upon for their safety from tonradoes.

Harold Brooks (of NSSL) has created a slightly different way of looking at these data, based on a suggestion from Raul Lopez (formerly of NSSL).

Fig. 2 Annual tornado fatalities 1875-2008 (source: Storm Data ), normalized by the annual U.S. population and plotted on a logarithmic scale. Shown are: the raw data (black dots connected by the dashed black line), and the data filtered by using a 3-point median filter and then a 5-point moving average (solid red line). Linear trends (solid green lines) have been fitted to the filtered data (first for the period 1883-1925, and second for the period 1925-2008).

What's been done in Fig. 2 is to present the normalized figures on a logarithmic scale and to do some simple linear regression. The linear fits are truly fascinating in that they suggest (a) a period with a nearly constant trend very close to zero for the period (1875-1925) and (b) a long-term exponential decrease in the fatality rate for the period (1925-2008). Superimposed on these trends is considerable interannual variability, of course, so the 10th and 90th percentile lines are included for the latter period.

How do we interpret these results? There appears to be some influence of demographics, with a shift from a population that lived and worked in locations widely distributed over the plains to one concentrated mostly in cities. Public awareness may have played a role, as well. Harold has collected a lot of this material into a nice Web page about Tornado Climatology ... I recommend you visit it. This material has been incorporated in a scientific paper [Doswell, C.A. III, A.R. Moller, and H.E. Brooks, 1999: Storm spotting and public awareness since the first tornado forecasts of 1948. Wea. Forecasting, 14, 544-557]

It's noteworthy that tornado fatalities in a given year are not very closely related to the number of tornadoes in that year (or even to the number of F2 and greater tornadoes). Fatalities are most closely associated with bad luck, when long-track, strong/violent tornadoes strike in populated areas. There's a capriciousness about such events that seems to heighten the fear they can generate, but which also breeds complacency. When this essay was first posted in the mid-1990s, the most recent events wherein a violent tornado struck a major metropolitan area had been in (a) 1979 - the infamous Wichita Falls event on 10 April that year - the 44 deaths represent a remarkably low figure given the amount of damage, with thousands of homes affected, and (b) 1989, when an F4 tornado resulted in 23 fatalities in Huntsville, AL on 15 November 1989. Surely the excellent watch/warnings for these events prevented a disaster of early-century proportions. This situation was updated on 03 May 1999. The number of homes affected in the Oklahoma City Metropolitan area F5 tornado is even larger than in the Wichita Falls event - at least 7500 homes were heavily damaged or destroyed, and yet only about 40 fatalities. I'm sure that no one affected can appreciate this (and my sympathies go out to anyone affected), but in my somewhat objective perspective is that the citizens within the OKC metro area were "lucky": it could have been a lot worse. Had the tornado struck at 2 a.m., or marched down the Interstates during rush hour on a weekday afternoon, or any of a number of other scenarios, the event could have had a much higher toll. It's a tribute to the integrated warning system (IWS - see Doswell et al. 1999, mentioned above) and the acceptance of responsibility by the residents within this large, sprawling metroplex in "Tornado Alley" that more deaths were avoided. I should add that the Wichita, Kansas area (Haysville) F4 tornado also devastated a large number of homes with a resulting fatality count of only 5! Considerable credit to the IWS is due there, as well. I note that it's only since 1991 that the Wichita area was hit by a violent tornado (the "Andover" event of 26 April 1991), whereas the Oklahoma City area has been spared a major event for a longer interval.

Note added 01 May 2014:

There have been major tornado events in the OKC vicinity several times since 1999:  08 May 2003, 09 May 2003, 10 May 2010,  24 May 2011, 20 May 2013, 31 May 2013

Since the issue comes up from time to time, let me consider the infamous Tri-State tornado of 18 March 1925. There are some reasons to believe that it was not a single tornado, but for the moment, let me ignore such questions. We've not seen its like since that time! There probably have been about 1000 tornadoes a year (roughly), on the average, since that time - forget the actual record of events, which has a number of biases and problems. Thus, as of 2009, there have been something like 84,000 tornadoes without another event like that Tri-State tornado. How many more years will pass before we see a comparable event? No one can know, of course, since it hasn't been repeated. It's hard to calculate a recurrence interval for a singular event! When a tornado like that one happens again (and it's inevitable that it will), what would be the death toll? Interesting question - the population at risk certainly has grown, but the watch/warning program and growth in awareness surely would offset that, to a large extent. It's fun speculation, but basically an exercise without much value. However, asking the question about the likelihood of killing 600+ people in a single event is not without value, as I'll discuss later.

Tornado forecasting began in the U.S. with John Park Finley (Galway 1985) in the latter part of the 19th century, but this early effort simply was not followed up. Finley's efforts came to naught in part because the civilian Weather Bureau (now called the National Weather Service, or NWS) felt that using the word "tornado" in a forecast was likely to cause panic, resulting in greater harm than that of any tornado. This myth of panic is a pervasive one and it took many decades to overcome this mindset. Modern tornado forecasting began with the efforts of Ernest Fawbush and Robert Miller, Air Force officers then stationed at Tinker Air Force Base near Oklahoma City in the late 1940s. [In March of 1998, we celebrated the 50th anniversary of the first tornado forecast.] When word of the Air Force program leaked out, it was inevitable that the Weather Bureau, however reluctantly, would have to follow suit, which they did in 1952, with the formation of the Severe Local Storms (SELS) unit. As luck would have it, the first full year of operation of this Weather Bureau unit - 1953 - was a year filled with major tornado disasters: Waco, Texas 11 May (114 dead), Flint, Michigan on 08 June (115 dead) and Worcester, Massachusetts on 09 June (94 dead) being the most well-known. See here for the history of the unit, which has evolved into the Storm Prediction Center.

Tornado forecasts were issued in the form we know now as "watches" and the newly-developed tool of weather radar, combined with the development of citizen volunteer spotter programs soon permitted tornado "warnings" to be issued: short-fuse alerts to the presence of tornadoes. The idea of the developing program was a three-step process of increasing awareness. An outlook indicated the potential for tornadoes. a watch indicated a localized area where tornadoes were possible, and a warning was issued when tornadoes had been seen (or indicated on radar) and were headed toward the threatened area (usually a county-sized region).

As these tools have been refined, and as radars and their information have become more capable of providing advance notice of tornado formation, the public has become (a) very aware of tornadoes in certain regions where tornado frequencies are highest, and (b) dependent on the "system" for alerting them to the tornado threat. The NWS has taken a role in training volunteer spotters, as well as continuing to refine and improve their methods for forecasting and detecting tornadoes. It's virtually certain that the continued fall in the frequency of tornado disasters and in the steady reduction of their casualty figures can be attributed in some measure to the increasing effectiveness of the watch/warning program.

It's a fun speculation to imagine what the annual death tolls might be without the watch/warning program of today. The graphs shown earlier suggest an average normalized death rate through 1925 of around 2 deaths per million of population. This has now fallen to around 0.1 deaths per million. Thus, on the average, it can be argued that without development of public awareness and severe weather forecasting, we might still be experiencing an average of several hundred fatalities nationwide, per year - with big interannual fluctuations both above and below that average, depending on when and where violent tornadoes affect populated areas. Perhaps the rural populations of tornado-prone regions in the U.S. wouldn't be affected that much, but it's quite likely that cities would occasionally be taken by surprise, as they were in the pre-public awareness era. Of course, in regions where tornadoes are rare and populations largely complacent, occasional disasters still can surprise both rural and urban populations (as they do even with the system in place). I note that the South (east of the Mississippi) has several handicaps in dealing with their not-rare violent tornado events:

  1. Relatively poor construction practices, with a relatively high percentage of mobile homes and few homes with basements,
  2. Lots of trees and rolling hills to limit clear views,
  3. Tornadic storms often are obscured by low cloud bases and poor visibility,
  4. Major tornadic storms happen after dark with some regularity,
  5. Typically high tornado speeds along the track during the cool season tornado outbreaks
  6. The absence of a marked seasonal peak in tornado frequency.

This combination of factors (and others) can create big problems for the warning process. Thus, it's plausible to suggest that a big tornado disaster is waiting to happen in the South. Will the next big disaster strike there, first? Impossible to say, of course.

Update:  Note added 13 April 2012:

2011 was the year that our luck ran out (see Fig. UD1).  Some major tornado outbreaks across the southern plains and "Dixie Alley" (see Fig. UD2) produced a disturbing 553 fatalities for the year, including the 160 fatalities produced by a single tornado in Joplin, MO on 22 May.  I believe that the factors discussed above regarding tornado outbreaks in the South came into play in a year when tornadoes happened to strike a number of populated areas.  As bad as 2011 was, there's no reason to assume it represents the worst possible scenario!

As shown in the following figure, large fatality counts occur only sporadically, when violent tornadoes interact with urban/suburban areas. The number of reported tornadoes continues to increase, but the overall behavior of the fatalities remains the same.  Big numbers are when we are unlucky, which only happens occasionally.  The increasing tornado frequency is assumed to be mostly the result of reporting practices, but it's logically possible that there could be meteorological or climatological reasons included in this increase.  Unfortunately, it's not possible to separate the reporting practice from any meteorological factors at present.

Figure UD1.  The annual tornado count (black) and tornado fatality count (red), for the period 1950-2011.  Note the fatality "spikes" in 1953, 1965, 1974, and 2011.

Figure UD2.  My proposal for three "tornado alleys" in the distribution of violent tornadoes (distribution map by Dr. Harold Brooks, National Severe Storms Laboratory).  The "regular" alley is what is classically referred to by the term "tornado alley", the "Dixie" alley has been recognized for some time, and the "Ohio Valley" alley is heretofore unmentioned.  The data don't show it, but I believe the actual distribution of violent events in the classical tornado alley extends farther west onto the High Plains (not just in northeastern CO) and northwestward toward the Canadian Prairie Provinces, bounded on the west by the high plains in the lee of the Canadian Rocky Mountains.

Events in the Southeastern U.S. during late winter and early spring of 1998, 2008, and 2011 have underscored the region's vulnerability. This vulnerability is at least in part attributable to an "it doesn't happen here" attitude that is notably inconsistent with established facts. The factors I mentioned above about lack of adequate shelters in many homes (especially with respect to mobile homes), the occurrence of the storms after dark, and the tendency for some really violent storms to be produced during the Southeast's late winter/early spring tornado season almost certainly were important in the events of 1998. Also, see Fig 2, where 1998 is a year with a fatality rate "spike" above the 90th percentile. Although excellent watches and warnings were issued for most of the really bad events during this period, these fatality figures are a portent of what continues to be possible in the region. It was fortunate that no major city was dead-centered by a violent tornado - the violent (F5) 1998 tornado narrowly missed Birmingham, AL, and the tornado(es) that hit Nashville that year were not violent ones. Despite the devastation of 03 May 1999, that year wound up just below the 90th percentile mark. Major disasters are still possible, as 2011 has demonstrated.  2008 also has spiked above the 90th percentile, with the so-called "Super Tuesday" outbreak of 05 February in the South-central part of the US being the major contributor, as well as a number of other, smaller killer tornado events.

It may be that our luck is running out - we may already have "bottomed out" in the annual fatality counts, although the jury is still out in that regard. To some, it appears that the atmosphere has created more devastating situations in recent years than in the recent past. I don't believe that this is attributable to El Niño, La Niña, or global warming. Rather, I believe that we've been fortunate in the recent past, and in recent years, our luck has been relatively bad, resulting in large interannual variability - something that has been going on for decades (as seen above). Whether or not the long-term fatality rate has bottomed out is yet to be determined. As I've hinted at earlier in this essay, there's no reason to believe that we've seen the worst of what's still possible. There's no guarantee that events more devastating than those in 1998, 1999,  2008, and 2011 can't happen again. In fact, given enough time, the potentially large fatality-producing scenarios are inevitable. It's just a matter of time ...

The enhanced awareness maintained by the watch/warning program also has led to a vast increase in the reporting of these events by storm spotters and storm chasers. As can be seen in Fig. 3, there's been a considerable increase in the frequency at which tornadoes are being reported. Even when these data are smoothed to reduce the effect of interannual variability, we can see that there's been a remarkable upward trend since the inception of tornado forecasting and warning with the advent of the Severe Local Storms unit (SELS - originally in Washington, D.C., then in Kansas City, MO until 1997, when it moved to Norman, OK and became the Storm Prediction Center, or SPC) in 1952.

Fig. 3 Annual reported tornado totals 1916-2008 (source: Storm Data - black dots with dashed line), included a smoothed version (first median filterd and then smoothing with a moving average). Annual "tornado" day totals are also shown (solid green line - only available through 1997).

The late Prof. Ted Fujita of the University of Chicago originally developed the now relatively well-known "F-scale" for rating tornado intensity [or, as discussed in Doswell and Burgess (1988), perhaps more properly the intensity of tornado damage]. Overall, most (~68%) reported tornadoes are given "weak" ratings of F0-F1. Only about 30% of all tornadoes get a "strong" rating of F2-F3, and only 2% get a "violent" rating of F4-F5. These percentages are only estimates, but the actual distributions year to year will not look wildly different from these rough figures (see Fig. 4).

Fig. 4 Number of reported tornadoes in each F-scale category, by year, 1950-2008 (Source - Storm Prediction Center).

It should come as no surprise to anyone that the majority of the increase in reported tornado frequency is in the "weak" (F0-F1) category. Much subjectivity is inherent in the F-scale, and the relationship between damage (which is the real basis of F-scale ratings in actual practice - see Doswell and Burgess [1988]) and windspeed is an uncertain one, at best. See Item B.11 here. Recently, the F-scale has been superceded with the so-called "Enhanced Fujita scale" (or EF-scale) by the NWS - see Doswell et al. (2009) for a discussion of this revised rating system


II. Reports, Verification, and Climatology

Awareness of the existence of tornadoes is necessary (but not sufficient) for tornadoes to be reported. In places where tornadoes are considered to be rare, they're rarely reported. The latter amounts to a self-fulfilling prophecy. For instance, when I began my career in meteorology, it was generally accepted that northeastern Colorado had some pretty fierce hailstorms but relatively few tornadoes. Today, it's widely acknowledged that eastern Colorado is well within the confines of "Tornado Alley." The difference? Awareness - and a lot of storm-chasers who venture to pursue storms wherever they occur in northeastern Colorado. Europe is beginning to recognize the reality that they have tornadoes, too.

It's interesting to note that in places around the world where tornadoes are considered infrequent, the "system" there makes little effort to recognize the situations that produce them - no watch and warning program, no official record of tornado events, etc. This tends to perpetuate the perception of the rarity of tornadoes; if the tornado threat is minimal but not zero, the potential for a disaster has been created, where a rare tornado hitting a populated area will catch forecasters and citizens alike unprepared. The growth in storm chasing is beginning to reveal new areas of heretofore unsuspected high tornado frequency even here in the U.S., as are new policies within the National Weather Service.

As warnings for tornadoes have become a part of life's background in many areas of the United States, there's been pressure to reduce the number of false alarms and, at the same time, to have virtually every tornado warned for in advance. Given a certain amount of reality with respect to the forecasting process,[2] this has put forecasters under considerable pressure from within the system. If they put out warnings, those warnings are expected to verify. If they don't put out a warning, nothing is expected to happen. As noted elsewhere by Roger Edwards (with a new addition by me as of February 1998), this pressure has resulted in some good things and some bad things regarding day-to-day operations in the NWS offices. Since the forecast offices issue the warnings and then go out and collect the data that are used to verify the forecasts, then they are vulnerable (rightly or wrongly) to the accusation of a conflict of interest. In my experience, there are shades of grey here, rather than black guilt or white innocence.

I'm going to take the following position here. There are several steps involved in enunciating my stance:

  1. Forecasts (and warnings) need to be verified. A forecast that is not verified is basically an exercise of little or no substance.
  2. The events being forecast should coincide with the verifying events. That is, if we verify a forecast of a thunderstorm, then we have to see lightning or hear thunder. Nothing else will do. For a tornado forecast (or warning) to verify, there must be hard evidence that a tornado did occur.
  3. For verification to have any meaning, the occurrence of the event must be detectable with high enough probability that the exercise of verification is worthwhile. If you are unsure whether or not an event happened, you are correspondingly less confident in the verification process.
  4. The only value associated with verification is to permit the forecasters to see quantitative assessments of the forecast quality, in order to proceed to improve the forecasts (and to document that quality for the benefit of their users). Any other use of verification, including most comparative verification,[3] is an abuse and should not be permitted to influence forecaster behavior.
  5. Forecast quality is a multidimensional topic that can't be summarized with a single measure. (see Murphy 1991; 1993)
  6. Once verification is done, the forecasters should be able to pursue a systematic procedure for forecast improvement. The improved procedures should again be subjected to verification in order to validate that the changes in procedure do, in fact, produce measurable improvements in forecast quality without also degrading quality in some way.
  7. For a forecast (or warning) to have any value, there must be some action to be taken on the part of the user as a result of receiving weather information that will mitigate the event in some way. For tornadoes, mitigation of damage can be viewed as nearly impossible, so reduction in casualties is the goal. Presumably, the forecast (or warning) must be received with enough lead time to allow the user to respond successfully. There are ways to reduce damage to structures, but they must be pusued long before tornadoes threaten a populated area.

Having looked at the record of tornado reports for the better part of my career, I'm ready to conclude that we still have at best a very sketchy knowledge of tornado occurrence (see Doswell et al. 2005, available here, for a related discussion of the problems with our severe storm event data). There are many, many reasons for coming to this conclusion. Basically, the reports of tornadoes reveal many hints of things that are going on unseen. For example, as noted by Grazulis (1993), the same percentage of reported tornadoes in western Kansas has damaged buildings as in eastern Kansas, in spite of the much greater density of buildings to be hit in eastern Kansas. It must be concluded that many more tornadoes are going unreported in western Kansas than in eastern Kansas, unless you're willing to believe that tornadoes in western Kansas have a more malicious nature than those in Eastern Kansas. Similar ideas were reported upon by Rasmsussen and Crosbie (1996) for the Panhandle of Texas. As noted already here, it is also clear from the record of annual tornado totals that the inception of the watch/warning program has had a major impact on reporting. Further, it can be seen that the recent emphasis on verification beginning in the late 1980s has caused further increases in the reporting of tornadoes. As noted in Kelly et al. (1978), the vast majority of the increase has been in the "weak" tornado category, whereas the strong and violent tornado totals have increased much less dramatically.

Interestingly, as shown in Fig. 3, tornado totals have increased by around a factor of 5 (from around 200 per year up to more than 1000), and the annual number of tornado days (days with one or more reported tornadoes) has also increased substantially (from around 40 to around 180), roughly by a factor of 5. The tornado day total is probably close to its actual value (it's already attained about half the absolute maximum possible), whereas I believe the annual tornado count might well increase substantially (perhaps doubling) before the numbers correspond to the true (currently and for the indefinite future, the 'true' values remain unknown) occurrence totals.

Even if the will to do verification is present, it's questionable whether or not the data on reported tornadoes are good enough at present to be used to draw meaningful conclusions. Clearly, if a tornado is reported, there's at least some good reason to believe that a tornado indeed occurred - although the confidence in that is not 100 percent, either. It is when a tornado is not reported that's the main problem. How can we be certain that those watches and warnings that have proven to be "false alarms" according to the data are really false alarms? How many "non-events" are really reporting failures? It appears that we simply can't be sure unless no convective storm at all occurred within the forecast area. Conceivably, we could at least figure that much out, right? Wrong! We currently maintain virtually no official record of the existence or non-existence of convective storms. We might be able to use lightning, except that the present lightning event data only include cloud-to-ground flashes, and then apparently only about 70% of those. Besides, not all potentially tornadic storms produce lightning, anyway! Basically, we're stuck without much hope for a conclusive verification effort.

There are ways we might work around some of the limitations imposed by these very imperfect data. but I'll reserve that for another essay, and perhaps a scientific paper or two!

This is hardly an ideal situation. The public has come to expect tornado watches and especially tornado warnings to be issued for every tornado that occurs, and only for events (not for non-events, ever!). This sort of perfection is basically beyond our grasp, but without good data about the events that occur, the process of developing any systematic approach to forecasting improvement is handicapped to a great extent. How can we assess how well our forecasting methods work when we don't have accurate and comprehensive data regarding what happened for each forecast?

The task of collecting the information about tornado occurrences falls mainly on the Warning and Coordination Meteorologist (WCM) at each Weather Forecast Office (WFO), in the newly modernized and restructured NWS. The WCMs, of course, have a multitude of tasks (as do all NWS employees in the WFOs): running spotter training, filling in with forecast shifts when needed, PR interactions, etc. etc. There's not a great deal of time to spend on detailed storm surveys after each event in the WFO area of responsibility, nor are there many resources to bring to a survey. A reasonably thorough survey should include a complete aerial survey of the effected area within no more than 12-24 hours of the event[4] ... clean-up rapidly destroys evidence needed to ascertain such things as damage intensity, what things were left standing vs. what things were swept away, where objects came to rest and their orientation when found, etc. In addition to the aerial survey, a comprehensive ground survey is needed at the same time. Witnesses need to be interviewed, photographs and videos need to be found and collected. There are many components to a complete survey and, given the resources being made available to the NWS for this purpose (essentially, none!), surveys of all but the most devastating events are virtually unknown. With the retirement and then death of Prof. Ted Fujita, who used to have independent funding from the National Science Foundation or other sources for his surveys, this information is not being collected routinely. With Prof. Fujita's retirement, those funds dried up and NOAA/NWS has done very little to replace his invaluable assistance.

Even in those infrequent events for which NOAA/NWS does manage to do a post-event disaster "assessment", it typically isn't done with much scientific care. Generally speaking, the main goal of the NOAA/NWS surveys seems to be to avoid the potential for litigation, rather than to arrive at scientific conclusions. I've offered to provide a scientific review of those surveys, at least to prevent the most egregious scientific howlers from appearing in the final version, and have been given the cold shoulder. I have a standing offer with NOAA/NWS to participate in one of their service assessments (previously, disaster surveys), but the chances of my being asked to participate are vanishingly small, in view of my reputation for being an outspoken scientist who will not shrink from calling each situation as I see it. If the NWS actually performed poorly in the situation, they know I won't hesitate to say so, and am not easily "encouraged" to silence.

In the wake of the 03 May 1999 devastating, long-track F5 tornado in the Oklahoma City metroplex, no one from the National Severe Storms Laboratory was included on NOAA's "Service Assessment Team" survey of the event! It seems that the NWS is not interested in the scientific aspects of major events. They merely want to be sure that their a$$es are covered. In spite of having world-class severe weather scientists available in the immediate vicinity, whom it could employ without incurring any additional expense, it appears the NWS didn't want to find anything it couldn't control - hence, team members from all over the country except NSSL were flown in to do the study. I find this patently absurd and believe this to be a shameful example of what I've described in this essay, and elsewhere. Although NWS performance in major tornadic events can range from outstanding to pitiful, most NOAA/NWS "service assessments" end up coming to roughly the same wishy-washy conclusions, many of which reflect the current agenda of NWS management, rather than a reasonably objective presentation of the facts in each case. The most hard-hitting of recent surveys was that done for the Plainfield, IL disaster, and that one had Bob Maddox for a team leader, when Bob was the Director of NSSL. Although some of the most negative material never appeared, there still were lawsuits from the event - as far as I can tell, those lawsuits never amounted to anything, since it's tough to sue the Federal Government and win a judgment.

For a time, in 2005, the NWS created a Quick Response Team (QRT) to provide at least some fast response survey, and I was included on that team. I've done one such survey, after the May 2003 tornadoes that hit in the Kansas City metropolitan area. It seems the main thrust of the QRT is to determine whether or not a violent tornado rating can be justified, rather than any extended scientific and/or engineering survey. See Speheger et al. (2002) - available here - for a discussion of the overall situation. Since 2005, the QRT has become increasingly irrelevant, as the NWS managers evidently are uninterested in having outsiders oversee even something as minor as the intensity ratings.

Tornado reports in the U.S. not only suffer from lack of committed resources and sloppy surveys even when the few surveys get done; it's worse than that. Many times, the WCMs are permitted to take the easiest path through the data, if they wish. Of course, some WCMs work at the highest possible level ... others take the easy route. For example, if a lot of distinctly separate paths can be collected into a single "skipping" tornado, this means a lot less work for the WCM. I've already written elsewhere that tornadoes seldom "skip" but this fiction makes it easy for some WCMs to get through the event with minimal effort. Storm chasers have seen a consistent under-reporting of tornadoes in tornado outbreaks. On one event I saw in 1981, my chase partner and I saw 12 tornadoes in one day, whereas the official record has only 6 or so. Other examples abound - Grazulis (1993) mentions several examples where nothing exists in the record, although clear evidence of an event has been found.

I could easily go on at great length on this topic. In spite of the great need for a high quality data base on tornadoes, there continues to be a dwindling commitment of resources and energy being devoted to the task. If the public expects to have improving forecasts of tornadoes, the absence of a good record of tornado occurrence is going to be a major impediment to any process aimed at achieving the needed improvements.


III. Forecasts, Warnings, and Responsibilities

a. Forecasting

Tornado forecasting (i.e., tornado watches) has shown some slow increase in quality since its beginnings in 1952 (see Doswell et al. 1993). As noted, the lack of quality tornado occurrence data makes it difficult to know in quantitative detail to what extent forecasting really has improved.

Basically, tornado watches capture something like 50% of reported tornadoes. Although this might seem like a low figure, it includes the vast majority of the important tornadoes (i.e., those rated F2 or greater) because it's the stronger events that account for the majority of the fatalities. Thus, most killer tornadoes occur in watches (see Galway 1975). How many tornado-associated deaths have been prevented as a result of the watches? There surely is no way to be certain, although perhaps it could be estimated by making a number of assumptions (as we attempted in Doswell et al. 1999). The discussion in Section I describes indirect evidence that the watch/warning program has been successful in reducing casualties, although the NWS cannot claim full responsibility, since the decline in fatalities began before the watch/warning process was first implemented. See Doswell et al. (1999) for a more extensive discussion.

Nevertheless, herein lies a major dilemma for the NWS. If they're totally successful in their mission of reducing tornado casualties (and it appears certain they've been partially successful), the tornado threat seems to diminish in proportion to their success. Tornadoes cause a huge amount of damage annually but their associated yearly fatalities have fallen below flood and lightning death rates. This makes the whole tornado program (including tornado research) vulnerable to budget-cutters! The NWS tornado program could become a victim of its very success.

As an interesting exercise, consider the following facts:

  1. The entire budget for the NWS in the late 1990s is roughly $700 million. Assuming there are roughly 100 million taxpayers, this averages out to about $7 per taxpayer (less than half of that per capita). Presumably, the watch/warning program is but a tiny part of that amount (see here for an estimate).
  2. The cost of a typical meal in a fast-food restaurant is about $7. The cost of going to see a movie (say, "Twister") is also about this price.
  3. The production costs for the movie "Twister" were about $70 million. The cost of the entire origina VORTEX project for two years (1994 & 1995) was about $2 million. Thus, for the price of producing "Twister" you could run VORTEX for 70 years! What does this tell you about where our national priorities are? How much are you willing to give up to save some fraction of the current annual tornado death toll? If the VORTEX project had access to the computing capabilities and programming support used to create the special effects in "Twister," how much faster could they process the data from the project?

Of course, it can't be said that the tornado watch program is without its problems and failures. As noted, the tornado watches aren't expected to catch every tornado, and if someone happens to get unlucky and their home is hit by a weak tornado that slips through without a watch (and/or a warning), all the statistics in the world are not much consolation. However, there's the principle of the greatest good for the greatest number. In order to catch the majority of tornadoes, given the present knowledge and resources available to the NWS, there would have to be more tornado watches issued, leading inevitably to more "false alarms" and more griping about needless concern.

b. Warnings

As for warnings specifically, this is increasingly a troublesome issue. Research has indicated that the newly-implemented WSR-88D Doppler radars do not represent a panacea for tornadoes, nor did any scientist expect that. Some tornadoes will occur without any obvious indication on the radar. As noted elsewhere, the signatures once thought to precede all tornadoes may not be as common as once thought. The "radar horizon" problem is an inescapable constraint on the instrument; a radar can't see beneath the horizon. Tornadoes only are important when they're actually present at the ground, often masked from view by the inescapable curvature of the earth's surface. Other potential problems exist (the true "cone of silence" [not the one from the movie!!], ground clutter, range and velocity folding, etc.) - however, lest this sound unduly negative, it's clear from the events of 01 March 1997 in Arkansas that the WSR-88Ds will be a tremendous boon in many of the most important meteorological events, with classic supercells and outbreaks of strong and violent tornadoes. This is precisely where it can do the most good, so the picture is not so bleak as it might appear from all the nattering nabobs of negativism (including the NWS union bureaucracy) regarding the radar.

Curiously, in considering how NWS offices perform in tornado outbreak situations, it's clear historically that some offices have simply been unable to deal successfully with such events. In some cases, it's been particularly inappropriate office policies (e.g., don't issue tornado warnings based only on radar, or don't issue warnings for tornadoes at all, since they always are over before the warning gets out, or don't issue a tornado warning until a particular radar signature is present). Hopefully, after the event, such policies can be swept away - what other office policies are still in force that have yet to experience the acid test?

Some offices simply are managed poorly, by managers with no clear vision of what it takes to handle truly hazardous weather situations. Since the offices tend to reflect the competence or incompetence of their managers, this tends to create mediocre offices. There are places scattered about the country where the offices have had a long history of uninspired managers - a situation that reflects poorly on the Regional and National management who are the prime players in selecting these local managers. When their disasters happen (as they inevitably will), will the truly responsible officials come in for their share of the heat? Almost certainly not! Perhaps they'll have retired, or moved on to some other "service" in the system. The heat probably won't even fall on the local managers - rather, it's likely to fall on some person relatively low in the pecking order who'll have been unlucky enough to be on duty that fateful day - a sacrificial lamb to prevent the heat from spreading up the management stairwell. There are several offices like this in the country, ticking away like time bombs, waiting for the combination of circumstances that will result in their disaster. Some of these are offices that have already had one disaster in recent history and yet they remain primed for another. Lessons go unheeded and the potential for disasters remains high.

Curiously, the hammer sometimes falls on good offices, with good managers and good staff, doing mostly the right things. It seems that when that big day arrives, even though some quality people are doing nearly everything right, they just can't bring themselves to believe that this is the biggest day of their forecasting careers! On such days, the SELS (now the SPC) severe weather outlook typically is excellent and the watches are issued properly and in a timely way. It's just that the local forecasters just can't seem to "pull the trigger." I have no ready explanation for this. I do know forecasters who have experienced this, and it's clearly a burden for them to realize after the fact they blew their chance at the event of their careers.

Thankfully, it appears that virtually all aspects of the NWS system (SPC Outlooks and Watches, local warnings, media dissemination, and prior preparedness efforts) performed very well during the 03 May 1999 tornado that hit Oklahoma City. The area is very weather conscious, and it appears that public awareness was a major factor in keeping the fatality toll to around 40. It's possible that had this event struck with the IWS as inadequate as it had been in the early 1950s, the death toll easily could have been 500-1000. Unfortunately, as discussed elsewhere, less evident events still can slip through the cracks in the IWS.

The conclusion that should be reached here regarding warnings is that they're damned difficult to do well even for the well-prepared, especially in the heat of the battle under a lot of time pressure and the stress of knowing the consequences for a bad job. It's no great stretch of the imagination to picture how a violent tornado interacting with a populated area can be exacerbated when the office is one of low quality. It seems to me that the bureaucracy is not doing everything it should to prevent problems in these occasional big situations, and yet it holds its forecasting staff to be totally responsible when the big days happen. I've written at length regarding forecaster training and the human element in weather forecasting, so I won't rehash that here. The NWS (and NOAA) bureaucrats seem to escape most of the heat for bad NWS performance under fire, and they don't seem eager to take on any of that responsibility, especially when the sh!t hits the fan.

There are some important issues that confront local forecasters when they have to make warning decisions. They're encouraged to reduce their false alarms, but it's clear that they had best have a tornado warning out before the big disaster happens. Given all that we don't know about tornadoes and tornadic storms, this is a pretty difficult situation for a forecaster. One way to be as sure as possible to have the warning out for the big event is to put out warnings even in questionable situations. But then there are going to be many, many false alarms, which leads to the "cry wolf" syndrome. To reduce false alarms is almost certainly going to mean that more tornadoes will occur without a warning; or perhaps the warning will be for severe thunderstorms instead of a tornado warning. It's unreasonable to expect warnings for every tornado but with no false alarms.

Now the NWS severe thunderstorm warning verbiage goes to some pains to point out that tornadoes can and do happen in severe thunderstorms with little or no indication - this is sort of a hedge. But in many places of the country, the local reactions by spotters and emergency managers are quite different for severe thunderstorm vs. tornado watches and warnings. There's no scientific basis for guaranteeing that a given event will or will not produce tornadoes, although we can indicate situations that are more likely to be tornadic than others. But the watches and warnings seemingly either call for a threat of tornadoes or they do not. It's my belief that local emergency managers should not be making their decisions based on the NWS's severe thunderstorm vs. tornado decision. Meteorology makes this very uncertain. Of course, there are other factors that mitigate against implementing spotter deployment with every severe thunderstorm watch - spotters are mostly volunteers who have other jobs to do, etc. I don't mean to belittle those factors, but we meteorologists just don't know enough to make this distinction with high certainty, so local emergency managers should factor in that uncertainty when they make their decisions.

Although the odds of experiencing the violent winds in a violent tornado in any one year are about 1 in 10 million, the odds of experiencing a tornado of any intensity within any particular square mile in any given year are about 1 in 1000 (at least 1000 times higher than being in the worst possible situation), even in "Tornado Alley" (assuming this means the most tornado-prone part of the U.S.) It turns out that the NWS has a tornado warning out about 50% of the time, when tornadoes occur (this means a probability of detection [POD] around 50%). On the other hand, only about 25% of tornado warnings actually experience a tornado somewhere within the warned area. That is, the false alarm ratio [FAR] is about 75% [These numbers vary somewhat from year to year and from place to place]. If you hear a tornado warning, then the area within the NWS warning will actually experience a tornado about 1 time out of 4. That is roughly 250 times the likelihood when the climatology is as high as 1 in 1000! In most places around the country, being in a tornado warning makes you at least 1000 times more likely than your local climatological frequency. Sure, the warnings aren't perfect and everyone wants to lower the false alarm rate.

Unfortunately, if we lower the FAR, we also will surely lower the odds that a warning will be issued when a tornado does occur (the POD). As it now stands, individual forecasters who don't have a warning out when a disastrous tornado hits are in a lot more personal trouble than forecasters who issue false alarms. The NWS as a whole gets blamed for the FAR problem, not individual forecasters. Just how would you react if you were a forecaster under these circumstances? It seems obvious to most individual forecasters that you'd best "cover your a$$" with a warning if there is even a low probability of a tornado. Survey teams don't investigate false alarms, looking for someone to blame; they only go out after disastrous events occur. Given the politics, I certainly can't fault forecasters for their choices.

Of course, there are some forecasters out there who are doing everything they can to improve their products. Sadly, they make these efforts mostly on their own. The NWS commitment to training, in my opinion, is far less than it should be if management wants to make training an important contributor to forecast quality. Ideally, what we would like would be a severe storms scientist sitting behind every NWS radar. In order to be that sort of radar expert, individuals have to devote virtually all their waking moments to the subject for decades, before they begin to be really competent at running a radar and interpreting what they see. What the NWS gives their forecasters is a few paltry weeks of non-rigorous training and expects them to pick up the rest with "distance learning" modules done in their spare time. Balderdash! That's nowhere near what it takes, but that's all the NWS is going to do, barring a miracle. Of course, even with the best of the best behind the radar (say, Don Burgess, or Les Lemon, or Jim Wilson), they won't be perfect, either. We don't know what such a high level of knowledge and experience could do in operational performance - because it's never been tried - but it certainly still would be short of perfection. How much better than the existing staff, who must spend time with many other issues than severe weather? I'm confident it would be a notable improvement, but I can only speculate. My guess is that numbers like POD = 75% and FAR = 40% are attainable with the best of the best, but these are only guesses. It might be possible to refine these guesses, but that's not my point, here. See Brooks (2004) - available here - for a useful discussion of this issue.

Sure, everyone wants to issue only perfect warnings, but it's an inescapable reality (for the time being) that the only way (barring a miracle) to reduce the false alarms is to increase the frequency of tornadoes with little or no warning at all. We're probably doing the best we can at the moment, given the technology, the state of training within the NWS (abysmal!) and the politics of how we deal with hazardous weather. All of hand-wringing about false alarms at an official level doesn't mean diddly-squat when individual forecasters face an asymmetric penalty function - that is, they're punished much more for not having a tornado warning out when one hits than when they issue false alarm warnings. Barring an unforeseen scientific or technological breakthrough somewhere, nothing much is going to change with respect to the current statistics.

Rather than griping about it, citizens should thank their lucky stars that someone is doing the best they can under some far less than ideal circumstances.

c. Personal responsibilities on the part of "the public"

To consider your personal responsibilities, ponder the following information:

  1. If there are 1000 tornadoes in the U.S. every year,[5] then about 690 of them are weak, 300 of them are strong, and perhaps 10 or fewer of them are violent.
  2. In a violent tornado, less than 10% of the path experiences the strongest windspeeds, capable of doing the damage that results in the violent rating. Anything less than violent windspeeds will leave interior walls standing in a reasonably well-constructed frame home,[6] making the home a reasonably safe place to be except when struck by the peak winds in a violent tornado.
  3. The area actually affected by a violent tornado is typically less than 5 square miles. For weaker tornadoes, the area affected is normally much smaller. The contiguous 48 states encompass 3, 539,341 square miles.
  4. A typical county in the midwestern U.S. has an area of about 500-1000 square miles, although considerable variability exists in this figure.
  5. Even in the most tornado-prone parts of the U.S., the chance of a particular square mile of land being struck by a tornado of any intensity is about once in 1000 years (according to current tornado occurrence data). Being struck by a violent tornado takes roughly 100 times that long.
  6. Current typical tornado watch sizes are about 25, 000 square miles, valid for about 6 hr. Current typical tornado warning sizes are about the size of a county, valid for about 30-60 min.
  7. Even in the most tornado-prone parts of the U.S., it's unlikely that you would be included in a tornado warning more than a few times per year. You might be included in a tornado watch as many as 20 times per year in the most tornado-prone parts of the U.S.
  8. Forecasters in the NWS do not always put into practice all that is currently known about tornado forecasts and warnings, for a variety of reasons. However, even if the forecasters in the NWS did everything that the science of meteorology tells them to do to forecast (and warn for) tornadoes, their forecasts still will not be perfect. A lot remains to be known about tornadoes, as I've discussed elsewhere. Basically, it's scientifically impossible to be very precise about when and where a tornado will hit, even when it's already on the ground doing harm and its exact location is known. Prior to a tornado actually occurring and being seen, it's correspondingly more difficult to forecast where a tornado will strike, or even if a tornado will strike.

From this information, you can derive a number of interesting results. I'll focus on a few, but others can be inferred; feel free to indulge yourself. If your county is under a tornado warning, and a violent tornado strikes somewhere in the county, the chances of your being anywhere within the damage path are about 5 divided by 500, or 1% (more or less, depending on the county size). The chances of your actually experiencing the violent windspeeds in such a tornado are about 0.01% or less, even given that a violent tornado actually strikes your county. You may not even see the tornado. What's your reaction? Was this a "false alarm" for you?

Since NWS tornado warnings are generally issued for counties or large fractions thereof, the foregoing suggests that even if the warnings were perfect [i.e., one or more tornadoes occurred in every warned area and no tornadoes ever occurred outside of warned areas - an unlikely situation for the foreseeable future], the odds of any particular person being hit are still pretty small. In a sense, then, even perfect NWS warnings (of the current format) would contain mostly false alarms! For some people, this fact means that they feel the NWS is "crying wolf" - in other situations, other people are saying "It struck without warning!" because they paid no attention to the warnings. Can the NWS "win"? I'll leave that up to you, my readers.

For the most tornado-prone parts of the U.S., if you simply went immediately to shelter every time you heard about an NWS tornado warning for your county and spent the next 30 min in that sheltering location, you'd "waste" something on the order of a few h per year, on the average. In ten years, that "wasted" time would constitute approximately one whole day out of that time span. How much is your life (and the lives of your family) worth to you in terms of time? What do you do when you hear a tornado warning? Do you know what to do? What about the rest of your family? What plan do you have in place. If you have no plan, does that mean you're putting 100 percent confidence in the NWS?

In the most tornado-prone areas of the U.S., being in 20 tornado watches would mean that you would spend at most 120 hours per year (out of a total of 8760 hours per year in non-leap years, or about 1.5% of that year) being careful to stay close to a NOAA Weather Radio,[7] and to watch the skies for threatening weather, etc. Realistically, given that most watch areas are canceled before the full valid time of the watch is done, the total is likely to be less than 100 hours.

There can be no doubt of two relevant facts. First, you're likely to be in watches and warnings where you're not hit by a tornado. Presumably, this should be a cause for a certain amount of gratitude, not anger.[8] Second, some unlucky people will experience tornadoes with neither a watch nor a warning (or perhaps only a watch, but no warning). Although the odds of this happening are relatively low, they're not zero. To a great extent, this is a matter of bad luck - a wise citizen could be prepared for such a situation.

If a violent tornado moves through a populated region (as in the case of a particular tornado in Arkansas on 01 March 1997), even with the optimum combination of watches and warning (which is about what the NWS actually accomplished on that day), people almost certainly will be killed. For some such deaths, it was just a matter of bad luck - being in one of the few spots where the worst winds happened, or some other unlucky chain of events. For others (and I think by far the majority of fatalities in such instances), their deaths are the result of one or more bad decisions, often associated with ignorance of the facts concerning tornadoes.

In the 1997 Jarrell, Texas event, people's home were swept away. They experienced the worst possible luck - the highest windspeeds in an F5 tornado. With no underground shelter, the normal advice to seek interior rooms was of no use to them. When a home is swept away, there's no shelter anywhere above ground level. Does this mean that this advice from the NWS is bad? That advice is, to paraphrase - seek shelter below ground; lacking that, seek an interior room. In other words, seeking shelter within the above-ground part of a frame home is at best, only an inferior alternative to underground shelter (or a safe room). That advice works for most of the people, most of the time - except when they're unlucky enough to experience the worst winds in the worst tornado events (very low probabilities, even given a tornado in your county - see above). If you don't have underground shelter and you live in a tornado-prone part of the world, whose responsibility is it if you're swept away in an F5 event? The odds of this are low, of course, but have you considered the cost of gambling that a low-probability risk will never occur? People are notoriously bad at realistic risk assessment.

For instance, it seems that many people believe that their experience includes just about anything that could ever happen.[9] In spite of the considerable amount of objective evidence that this viewpoint is patently false, it seems that since most people have never experienced a tornado up close and personal, and they believe it will never happen to them. And of course, for most of them, this will be true. But folks who have experienced a tornado tend to take it all rather more seriously. A low probability doesn't mean it can't happen - odds are you're going to skate by without a care, but there are unlucky folks every year. Just how much do you value your life and those of your loved ones? What's your reaction to tornado warnings?

The 03 May 1999 event in the Oklahoma City metroplex underscores this point. In a community that's probably as well-prepared as possible, with excellent forecasts and warnings, the tornado still produced about 40 fatalities. I believe this to be close to the current minimum possible in such a violent, devastating event. A big lingering issue at the moment is the availability of adequate shelter. Most new construction in this area doesn't include a basement, so like the unlucky people in Jarrell, those who experience the violent winds in a violent event (only a small fraction of the total area affected) have little or no guarantee of survival above ground. Many people in fact managed to survive despite not having adequate shelter. They were lucky. In my opinion, we don't need more technological gimmicks to provide advance warning of such tornadoes (at least in "Tornado Alley") - we need to consider how to provide adequate shelter from the extreme events if we want to reduce the fatality counts when violent tornadoes strike populated areas.

One by-product of modern life in the U.S. is that many folks seem to have concluded that it's the government's job to protect them from such things. Invariably, after a devastating tornado, the media seem to be able to find someone who is willing to say "The tornado struck without warning!" even when the watches and warnings were indeed issued promptly. For such folks, they either didn't hear the warnings because they took no steps to be able to hear the warnings, or they ignored the warnings, perhaps feeling that they've heard warnings dozens of times before and nothing ever happened. What they seem to require in the way of a "warning" is that someone in authority needs literally to drag them out of their complacency and into a place of shelter. Anything less means that "the tornado struck without warning"!

Everyone needs to understand that all people ultimately are responsible for their own safety. They cannot and should not depend on the government (Federal, state, or local) to do that for them. Surely, governments at all levels are doing many things to protect their citizens,[10] but in the end, if people aren't willing to accept some of the responsibility for their own safety, there's nothing anyone can do to prevent needless casualties. Stupidity, repeated often enough, can become a capital offense, and there's no way to prevent some people from behaving stupidly. How many times is the barn door shut after the horses have escaped? Sirens, tornado safety drills in schools, workplaces, and home, spotter programs - all these take resources and the tornado threat seems so remote at the same time that the budget is always stressed. Why not cut out those programs and spend the money somewhere else? Just ask the citizens of those towns that have experienced a tornado!! Afterwards, the sirens go up, the drills are followed religiously, the spotters are re-equipped - but then the wait begins for the next event. Complacency settles in remarkably quickly, it seems. One generation is sufficient for the majority of metropolitan area residents to have forgotten past events - even the rare major events. Part of the task of protecting oneself from weather-related hazards, like many other hazards we face, is learning what to do and when to do it. Vigilance for a rare event tends to slide perceptibly into complacency, and complacency is the bane of the people trying to create meaningful self-help programs for tornado preparedness. Citizens who experienced the tornado die, move away, forget. New citizens arrive for whom the tornado is something that happened a long time ago to someone else. In five years, the tornado is a fading memory. The sirens fall into disrepair and eventually are removed. The drills become meaningless rote and perhaps are dropped. The spotters get tired of false alarms, and lose interest in the training. "Why should I care? It's never going to happen while I'm here!"

The atmosphere is notoriously intolerant of ignorance on the part of humans. If some people refuse to learn and implement proper safety procedures, then there's (again) not much anyone can do to help those folks. It's important for the NWS to do its best to disseminate information about hazards like tornadoes, of course. The National Weather Service can still pursue many untried avenues in disseminating information about hazardous weather, but not all citizens avail themselves of the warnings that are issued. Many citizens who receive timely warnings still take no action, or their actions are inappropriate. Preventing tornado casualties is most assuredly a two-way street.


NOTE: Papers marked with an asterisk are available here.

Brooks, H. E., 2004: Tornado warning performance in the past and future: A perspective from signal detection theory. Bull. Amer. Meteor. Soc., 85, 837-843

*Brooks, H.E., and C.A. Doswell III, 2001: Normalized damage from major tornadoes in the United States: 1890-1999. Wea. Forecasting, 16, 168-176.

*Doswell, C.A. III, and D.W. Burgess, 1988: Some issues of United States tornado climatology. Mon. Wea. Rev ., 116, 495-501.

*Doswell, C.A. III, S.J. Weiss, and R.H. Johns, 1993: Tornado forecasting: A review. The Tornado: Its Structure, Dynamics, Prediction, and Hazards (C. Church et al., Eds), Geophys. Monogr. 79, Amer. Geophys. Union, 557-571.

*Doswell, C.A. III, A.R. Moller, and H.E. Brooks, 1999: Storm spotting and public awareness since the first tornado forecasts of 1948. Wea. Forecasting, 14, 544-557.

*Doswell, C.A. III, 2003: A Guide to F-Scale Damage Assessment. U.S. Dept. of Commerce, NOAA/NWS, 94 pp.

*Doswell, C.A. III, H.E. Brooks, and M.P. Kay, 2005: Climatological estimates of daily nontornadic severe thunderstorm probability for the United States. Wea. Forecasting, [in press].

Galway, J.G., 1985: J.P. Finley: The first severe storms forecaster. Bull. Amer. Meteor. Soc ., 66, 1389-1395.

Galway, J.G., 1975: Relationship of tornado deaths to severe weather watch areas. Mon. Wea. Rev ., 103, 737-741.

Grazulis, T.P., 1993: Significant Tornadoes 1680-1991 . Environmental Films, St. Johnsbury, VT, 1326 pp.

*Kelly, D.L., J.T. Schaefer, R.P. McNulty, C.A. Doswell III, and R.F. Abbey, Jr., 1978: An augmented tornado climatology. Mon. Wea. Rev ., 106, 1172-1183.

Murphy, A.H., 1991: Forecast verification: Its complexity and dimensionality. Mon. Wea. Rev ., 119, 1590-1601.

Murphy, A.H., 1993: What is a good forecast? An essay on the nature of goodness in weather forecasting. Wea. Forecasting , 8, 281-293.

Rasmussen, E.N., and C. Crosbie, 1996: Tornado damage assessment in VORTEX-95. Preprints, 18th Conf. Severe Local Storms (San Francisco, CA), Amer. Meteor. Soc., 153- 157.

*Speheger, D.A., C.A. Doswell III, and G.J. Stumpf, 2002: The tornadoes of 3 May 1999: Event verification in central Oklahoma and related issues. Wea. Forecasting. 17, 362-381.