This is the full text of a WeatherZine Guest Editorial, appearing in abridged form in the Newsletter of 27 August 1999

Are warning lead times the most important issue in tornado events?

by Chuck Doswell

Having reviewed the testimony by Professors Bluestein and Wakimoto, and Dr. Weisman on 16 June 1999 for the House Committe on Science, I'm a bit distressed to find it somewhat misleading. Much of what they say is fine, of course, but their perspectives are limited to those of basic researchers in academia. Their testimony uniformly notes the value of research in increasing tornado warning lead times, which might well be true, but is not always relevant. There is no doubt that warning lead time is an important factor in life-saving, and I'll have more to say on that, but it certainly is not the only factor and, in some cases, is not even the most important factor. As noted by Dennis McCarthy's testimony in the same hearings, the outbreak of tornadoes in Oklahoma and Kansas on 3 May 1999 serves as an excellent example of forecasts and warnings at their best, with warnings having up to 32 minutes of lead time for the Moore and southern Oklahoma City area. It's hard to imagine any real need for warnings with lead times longer than 32 minutes! [In fact, it's possible that longer lead times would be less effective because they might not convey a strong enough sense of urgency. There is very little concrete information about what might be the most effective warning strategy, and I know of no effort underway to obtain such information.] Supercell storms producing significant tornadoes are not difficult to recognize and track with radar and/or spotters, and the National Weather Service (NWS) has a pretty good record in events of this magnitude. The fatality count associated with such events usually is not the result of a problem with lead times!

Instead, when large, long-track, violent tornadoes interact with populated areas, other factors influence the casualty counts: for example, the structural integrity of buildings, the climatological frequency of tornadoes (related to the level of prior participation in tornado preparedness programs), the availability of proper shelters, warning dissemination, and public confidence in the forecasts and warnings.

I'm not going to address the structural integrity issues more than superficially here. It turns out that there are many things citizens and businesses can do to mitigate damage from tornadoes. It's a myth that there is nothing that will reduce tornado-associated property losses. FEMA's Building Performance Assessment Team Program offers considerable information on this topic. Further, there's good evidence that reduction in property losses could help reduce casualties. FEMA's goal is to build disaster-resistant communities (not disaster-proof!), and that should be a goal of ours, too.

In central Oklahoma, most homes don't have basements and only a scattering of homeowners have built tornado shelters. Was it just a coincidence that another violent tornado on 3 May 1999, but in Kansas (where basements and shelters are much more common than in Oklahoma) resulted in significantly fewer fatalities? In less intense tornadoes, the ordinary precautions advocated in the absence of a basement or shelter generally will be effective. Unfortunately, for violent tornadoes, nothing short of a proper shelter offers a reasonable expectation of avoiding serious injury or death. Even in such a tornado-prone area as central Oklahoma, a certain amount of complacency tends to set in as the years since the last tornado go by. Compared with other regions, central Oklahoma has a relatively high level of severe weather awareness, but this has not led most citizens to build shelters for themselves. Perhaps complacency is an inevitable outcome of the relatively low probabilities of experiencing the violent winds in a violent tornado. Although the investment in a proper shelter is relatively modest, it can be argued that the odds favor a "do nothing" strategy as the most cost-effective one, even in central Oklahoma.

Unfortunately, the odds are not of much comfort if you're in the path of a violent tornado. In fact, the odds are a real problem in those parts of the country where tornadoes of violent proportions are sufficiently infrequent that most citizens believe "Tornadoes just don't happen here!" Such erroneous assumptions result in a pronounced lack of preparation. The Worcester, Massachussetts tornado of 9 June 1953 that killed 94 people is an example of why preparedness is important: even rare events happen from time to time. Whereas Worcester has a pretty low violent tornado probability, the probability is distinguishable from zero. A true zero probability is only an abstraction; even New England's violent tornado probability is high relative to, say, Antarctica. The frequency of violent New England tornadoes is low enough that we have only a fuzzy knowledge of those probabilities in comparison to violent tornado probabilities in Oklahoma, for instance. It's only a matter of time before another violent tornado creates a major disaster in New England, and it's quite possible to imagine even worse outcomes than on 9 June 1953. How do we sell citizens on the importance of avoiding complacency? More on this later.

Another factor to consider is the public response to warnings in view of the "Cry Wolf" or "Chicken Little" syndrome. It's possible that warnings are actually more effective in regions with relatively low tornado frequencies; hearing warnings regularly may lead to a lack of confidence in tornado warnings that can be dangerous.

At the moment, I see only two ways to decrease the false alarm rate and, at the same time, increase the detection frequency:

(1) an uprecedented commitment to forecaster training, or

(2) some currently unforeseen scientific breakthrough.

At present, the only way to decrease false alarms is to increase the number of tornadoes that strike literally without warning, which I don't believe anyone wants. Lead times for warnings are closely related to uncertainties about which storms, if any, are going to produce tornadoes. Uncertainty translates into angst over whether or not to "pull the trigger" on a tornado forecast or warning. In turn, agonizing over the decision can result in reduced lead times. There clearly are asymmetric penalties associated with the binary decision (to warn or not to warn); a false alarm is unlikely to kill anyone. Hence, false alarms are favored heavily by forecasters over failures to detect. Scientifically absurd goals set by high NWS officials notwithstanding, there just is no way to mandate a reduction in false alarms without inevitably increasing the detection failures, barring either (1) or (2) above.

Harold Brooks and I think the answer to the very correct public perception that most tornado forecasts and warnings are false alarms is to rethink our whole approach to forecasts and warnings in order to accomodate our uncertainty via some sort of probabilistic approach. Certainly, improvements in severe weather forecasting-related science can improve our track record, but unless some unforeseen breakthrough occurs, I don't look for the reality of mostly false alarms to change in my lifetime, if we persist in categorical forecasting and warning strategies. I think it's high time to rethink our relationship to our users.

A transition to a more honest and more truthful forecast product than we now produce, that provides users with what I think would be useful information about our uncertainties, will not be easy. Any serious consideration of this should include a careful study done in collaboration with psychologists and sociologists on how best to accomplish this transition. I do not claim to know precisely how to do this. We need to test our ideas with the help of people who really know how to gauge public perception and response, and get the benefit of their ideas, as well. The NWS has a tradition of making default assumptions about what "the public" (whoever that might be) wants and needs. This is a bad tradition. We weather folks need to overcome our prejudices about the value of contributions from outside of meteorology, and embrace a comprehensive, substantive, and empirically-validated look at how we can serve best the users of weather information. Let's replace speculation and opinion with facts. If we weather professionals make the choice to commit to a transition to probabilistic forecasts and warnings, then we will need to employ a wide range of skills outside of meteorology to test and evaluate how to make such a change with the least difficulty. It won't be quick or simple to do, but I believe that in the long run, it will be the right thing to have done.

***Comments on this essay can be sent to me at <>***