Saturday, July 30, 2011

Do we warn too much? Some thoughts on tornado warnings in this tragic year of tornado deaths...

It's been 2 months since the horrifying Joplin tornado (death toll now 159), and 3 months since the shockingly deadly tornado outbreak in the Dixie states that killed well over 300 people. It now looks like 2011 will be the 4th deadliest U.S. tornado year on record. I don't think most people have any idea how jaw-dropping that is... one has to go way back before a coordinated tornado warning system was in place to find annual tornado death tolls of that magnitude!

When big tornadoes hit cities, there will be deaths. And there have been many large long track tornadoes this year in populated places. But the staggering death tolls this year also raise some questions, including these: Are people getting complacent regarding severe weather preparedness? And do we now warn so much that warnings sometimes fall on deaf ears? A respected nurse at one of the hospitals in Joplin told me, "We get warnings and sirens all the time... we're, like, tone deaf... it's hard to take them seriously." I've heard many similar comments from other people. Severe weather education is a never-ending task that requires continued attention and priority, and getting people to listen and take action is deeply rooted in social science rather than meteorology.

It is noteworthy that, compared to the 1980s, the number of tornado warnings in some parts of the U.S. (such as the Southeastern states) has increased at least seven or eight-fold, according to a recent article in the Birmingham News, while false alarms are averaging close to 80%. I know of no concrete social studies that have examined "warning fatigue" regarding tornadoes and severe weather. Yet I do agree with respected Birmingham TV meteorologist James Spann's recent comments that have stirred some controversy.

I continue to see a number of tornado warnings issued in marginal settings where, based on years of data and increased operational knowledge about tornado environments, the atmosphere can only support, at best, weak or brief tornadoes. In those cases, the threat is considerably less compared to days like the Joplin tornado event this year, and the Dixie Super Outbreak. Yet it is often difficult to discern the relative degree of threat or risk in the text of many warnings, whether read verbatim or presented via media such as television. I know that the National Weather Service (NWS) and television/media meteorologists have a sincere desire to save lives and have the public know about all weather threats. Yet I wonder if we shouldn't reaffirm that we can't warn or cover all severe weather events. Instead, shouldn't we make a stronger effort to emphasize warnings in those weather settings that clearly have potential to be truly dangerous to many people in populated areas? I'm no expert in social science, but with my background in severe weather research, here are some comments from a meteorological perspective.

The first graphic above shows combinations of low-level wind shear (storm-relative helicity or SRH) and instability (CAPE, or convective available potential energy). On the most simplistic/basic level, these are the most important environment ingredients linked to supercell tornadoes based on research over the past 25 years. On this graphic, I've plotted estimated SRH/CAPE points (using nearby RUC model analysis soundings) for 44 tornadoes that were associated with 4 or more deaths over the past 11 years. I've also drawn in a curve suggesting a rough lower "limit" to SRH/CAPE combinations that generally support tornadoes that can kill larger numbers of people; note that all events but one (98%) fell above this curve.

Looking now at a much larger database (2214 supercell soundings I've collected over the past 11 years, 90% associated with NWS tornado warnings), the 2nd graphic above shows the percent and number of both non-tornadic and tornadic supercells (tornadoes shown by EF-scale intensity) falling near and above the red curve in the first graphic, suggesting more favorable SRH/CAPE combinations for strong/violent tornadoes. Notice that, using the discriminating curve suggested by the first diagram, only 17% of the non-tornadic supercells were "false alarms", falling near or above the curve but not producing any tornadoes. But moving over to significant tornadoes (EF2-EF5), notice that increasing majorities of these supercells fell into the more "favorable" area above the curve. Using only this simplistic SRH/CAPE scheme in supercell settings, this suggests workable false alarm and detection rates for supercell tornadoes, not even considering other ingredients such as deep-layer shear, cloud base height, amount of convective inhibition, storm motion relative to surface boundaries, etc., that have been found relevant in tornado forecasting research.

This isn't really new information; forecasters at SPC have used SRH and CAPE combinations along with forecasts of other ingredients for years now when generating outlooks and watches for tornadoes. However, tornado environment ingredients seem to be used less consistently when actual tornado warnings are considered. This is supported by the first column of the 2nd graphic above, where tornado warnings (not shown) were issued for 984 of 1160 non-tornadic supercells, yet the SRH/CAPE environment for 83% of these cases fell below the red curve in the first graphic above. Certainly, supercells in SRH/CAPE environments below the red curve do produce tornadoes, but when they occur, the large majority (80-85% according to my database) are weak (EF0-EF1 intensity). When thunderstorms form and become supercells, it is the environment area above the red curve where the probability of deadly tornadoes increases dramatically.

My point is this: Shouldn't we place a much stronger emphasis on radar-based tornado warnings issued in settings where SRH and CAPE are in the area near and above the red curve in the first graphic above, even prior to confirmed spotter sightings?

There are a variety of ways to work toward this, including better environment awareness by meteorologists (the SPC mesoanalysis page is a great tool in this regard) when issuing and presenting warnings, stronger wording and importance placed on warnings when environments are in the enhanced SRH/CAPE area, and making the public more aware that there are different levels of danger in weather settings.

I know that many of my colleagues in the NWS work hard to incorporate environment information into tornado warning decisions, and that there are many times when it is very difficult to know when to hold back from issuing a warning, or to go ahead and "push the red button". Also, I know that meteorologists in today's media intensive society don't usually get rewarded for not warning in a marginal situation and avoiding a false alarm, but they are ruthlessly scrutinized when a notable severe weather event is missed. I'm not suggesting that we don't warn in marginal situations where tornadoes can still produce isolated damage and threat. I am saying, let's make sure to put the strongest emphasis on warnings in environments where ingredients appear more optimum for stronger supercell tornadoes, and try to be more clear about that to the public.

A 2-tier warning system that would use an enhanced danger "red flag" in larger SRH/CAPE combination situations would certainly be a more radical solution, and require more public education. But this could help hospitals, factories, and public venues make more informed decisions regarding costly labor-intensive sheltering in the more dangerous situations, and possibly reduce apathy from "false alarms". It also might help emergency managers make more informed decisions regarding the use of sirens, maybe running them longer and more consistently in "red flag" situations, or using different siren tones to catch people's attention.

Several of my colleagues have told me the above would be unworkable, and even confusing to the general public. That may be. But I do feel strongly that we can do more to help the public recognize warning situations that clearly have greater danger affecting more people. I can't avoid a strong sense that, with all the information and knowledge available to meteorologists today, we can do a better job conveying true tornado threats to the public and reducing false alarm perceptions that can engender public apathy. I think this is worth at least a look in this tragic year of tornado deaths.

In a future post, I'll discuss some recent example cases that relate to the above discussion.

- Jon Davies 7/30/11


jimmyc said...

I like the idea's you presented and I think what you are saying is that probabilistic warnings are needed. A system that can convey both the threat and the probability of occurrence. A warning can have lower than typical probabilities of a tornado to convey the marginal threat. Whereas a tornado warning which has a tornado on the ground can reflect higher or the highest probability of occurrence.

There was an event in northeast OK one night this season where a powerful supercell went through town
but the damage was from the rear flank downdraft, not a tornado. I think we have to have both flexibility in the warning types to convey the severity and uncertainty, and an ability to cover a sufficient area to account for rapidly changing storm structures, motions, and threats. The latter also covering our technological abilities to be able to diagnose such features with accuracy at the level of scientific understanding and yet allow forecasters to make pattern recognition calls.

I also like that you used the terminology "false alarm perceptions". Assuming some people are fatigued, we have no idea WHY they are fatigued. If it didn't strike them while they were in the warning, do they count that as a "miss"? Or did a tornado actually occur, which by our standards is a hit?

rdale said...

Excellent post! I am a BIG fan of NOT warning 4 counties for extended periods of time for a possible QLCS EF0 spinup that lasts at most 5 minutes... I wonder if this can work into probabilistic warnings somehow?

Steve Nelson said...

Jon, yes the idea of revising warning thresholds based on threat is being looked at by the NWS. I would say questions such as "how do people respond to warnings/alerts?" and what information specifically do they respond to need to be asked before establishing a new warning system, such as warning vs red flag. However, your point about utilizing the storm environment in warning decision making is well taken.

I think many, including me, share your concern that the thresholds we use to issue tornado warnings has dropped some over time, this in spite of previous research that has established radar and environmental thresholds for discriminating tornadic storms. I have heard from some forecasters that this happens due to (perceived) emphasis on verification measures (which do not track magnitude, duration, intensity, or impact, only occurrence) by management. Others simply are afraid of the missed event. As a warning forecaster of 15 years, I can attest to that fear. Factual information and reassurement from management and end users are the only way that fear will be alleviated. The reassurement from end users may be difficult to come by due to the issues involved.

I also have some concern that if there was a multi-level warning system that it should be based more on radar parameters than environmental parameters. Higher resolution radar data (superres) does show mesocyclones more clearly and seems to have skill at estimating tornado intensity especially with larger tornadoes at close range to the radar. Environmental data gives potential strength, but I can provide several examples of low CAPE/high shear events that produced very significant damage with injuries and/or deaths. On the flip side of the argument for use of radar data, I believe superres radar has also made similar features look more threatening due solely to increased resolution. The features look "scary" but the magnitude of the rotation or velocity is the same or less than it would have been with legacy resolution.

Do you have specific cases of warnings being issued at low environmental thresholds you can share?

Steve Nelson
SOO WFO Peachtree City, GA

Jon Davies said...

Dr Correia (jimmyc):

Appreciate your comments... I agree that probabilistic warnings would be a good thing ideally, at least for meteorologists, EMs, and properly trained media information disseminators. As far as the public is concerned, I'm told by many that most people can't handle any more "complexity", as there are too many out there who still have trouble with concepts like warning vs. watch. It goes without saying that this is a very difficult issue.

Jon Davies said...

Rob (rdale):

I know QLCS/bow echo tornadoes have been a hot topic in recent years, and, make for interesting study. But I've slowly come to the conclusion that it isn't practical to warn on these as tornadoes... the duration is too short and the area affected typically too small. Before very many meteorologists knew what QLCS tornadoes were, the damage was most often attributed to channels of strong straight-line t-storm winds. On a practical level, that's how I think those events should be warned, with an emphasis on damage potential from 80-90+ mph winds. I know people occasionally get killed from QLCS tornadoes, but the same goes for strong squall line wind events.

Jon Davies said...


Excellent comments, well put. Thanks. There are clearly difficult issues and no clear answers when giving consideration to making any changes to the current warning system.

The verification problem is big... the current yes/no occurrence system is too simplistic, and there are major penalties for missing events, with no rewards for restraint when attempts are made to avoid false alarms. New better radar resolutions can make many "lesser" events look scary on radar without the benefit of long experience with new radar products.

The low CAPE/high shear events seem to be most common in the southeast where you are, and are a legitimate concern, as you mentioned. That's an area where I'm planning to do more work.

I also have numerous examples of low environment threshold tornado warning situations, and I'll share some notable cases for discussion in future posts.

Tom said...

Personally, I believe one of the big reasons people don't heed tornado warnings is not that they don't hear them but rather that the sirens go off so often, it is a bit like they are crying wolf. I live in Joplin (no damage from the storm) and I'm used to hearing the sirens go off numerous times each spring. Unfortunately, it would be a lawsuit waiting to happen if they didn't err on the side of caution and blow the sirens even when they may not be necessary. A few years ago some straight winds took out a convenience store in an area town. Following that I remember numerous discussions about lawsuits because the tornado sirens weren’t blown (umm… it was straight winds and it was over so fast nobody could have blown the sirens). I heard similar rumors around Joplin; there wasn’t enough warning and the sirens didn’t go off early enough or long enough, etc. Personally, I would rather have the sirens indicate a direct and immediate threat rather than the possibility that there MIGHT be a threat. I don't want the sirens 20 minutes early. When I hear the sirens, I want to KNOW that I am in imminent danger and I WILL be hit within the next couple of minutes. I have weather radios, TV, FM radio, and an eye on the sky to tell me there is a threat that might be deadly. I don’t know for a fact, but to me it “feels” like the sirens blow any time there is a tornado warning issued and/or as soon as someone sees something on Doppler that might look bad. While I appreciate the heads up, I cannot spend 1-2 hours per day 3 or 4 days per week hiding in my basement throughout the spring season. With that said, please blow the sirens when you know that a tornado is on the ground and it is going to hit me. That’s when I want to hear the sirens.

This article mentions a multi-tiered warning system. I believe we already have that. I see tornado watches and then tornado warnings on my weather radio, TV, and FM radio. I think the sirens should be the third tier; the “you are going to die in the next 5 minutes if you don’t take cover NOW” tier.

Jon Davies said...


You bring up a big issue... how much lead time is best for warning the public? Many meteorologists think the goal should be longer lead times... say 30-40 minutes or more. But a recent Accu-weather poll (non-scientific) suggested that many people feel 10 minutes or so is plenty. People want sirens to tell them what the danger is _now_, and if sirens stop running, many think it has passed.

Unfortunately, we don't always know when a tornado is on the ground, and even with all our technology, we often can't be that accurate as to when a concrete threat is truly imminent. With the Joplin tornado, it formed fast on the southwest edge of town and became large and intense very quickly. It does sound like that deadly situation could have been handled better with the siren sounding longer and more consistently, but it must also be said that it was a difficult fast-evolving situation. The point I was trying to make in my post is that when the meteorological environment
falls into a certain "intensity" of parameters, there needs to be more of a heads up and urgency conveyed for the public. Even though there had been no tornadoes prior to Joplin that afternoon in southwest MO, the parameters that support strong tornadoes were increasing.

You also mention the finger-pointing of locals when something damaging (such as a strong wind event) gets missed in the warnings and sirens. That tends to up the false alarms, as meteorologists then warn on and the sirens sound for all kinds of marginal situations to avoid being blamed for missing something.

With all our technological "sophistication", we find ourselves struggling with a host of sociological issues for which there are no easy solutions.

Simon said...

Great points raised in this post. This year has had a historically significant number of fatalities, but it, also, has been a historic tornado year. How many years since the Super Outbreak of 1974 have there been 6 EF5 tornadoes? Also, significant tornadoes, which moved through the hearts of relatively large cities like Joplin, Springfield, and Tuscaloosa. I have to admit I'm a little surprised there weren't more fatalities this year. I was in Tuscaloosa when the tornado plowed through the city; on the northeast side of town is a neighborhood called Albert City with many poorly built structures and a relatively high population density. I think it's remarkable the death toll in that neighborhood alone was not higher than in Joplin, MO. This might raise the argument that people might not be very complacent.

I've found through speaking with people many are more likely not to heed a warning, because they have a preconceived notion tornadoes don't occur in their area, not because of false warnings. Once an elderly gentleman from Mt. Vernon, MO told me he doesn't worry about tornadoes, because they don't hit ridge lines, so they strike places further south in valleys near towns like Aurora, MO. And I've heard multiple people in Norman, OK tell me they didn't heed tornado warnings, because tornadoes don't hit valleys like Norman's location.

Perhaps over a larger region maybe it appears there are many warnings, but in one location/county over the course of a year there might be relatively few tornado warnings. To the best of my knowledge the sirens in Norman, OK have only sounded once, maybe twice, this year, and haven't sounded at all in over half the years of the past decade to the best of my knowledge. Also, it might be best that all people should be warned of potentially tornadic storms to avoid potential injuries even if deaths are less likely; mobile homes and weak structures can receive relatively significant damage from an EF1 tornado. And while an EF1 or EF0 tornado will likely not roll a car it can knock over trees and large branches onto cars possibly seriously injuring or killing the passengers inside. While some people are complacent with warnings there are, also, many who are not, and they should not be withheld from the privilege of warnings that have the potential to be useful. I feel it's better to have a tornado warning and not have a tornado than to have a tornado and not a warning, because in the latter case nobody will take cover.

Chuck Doswell said...

There's very little hard evidence to suggest that environmental conditions allow an accurate forecast of future tornado intensity. All of this uproar over the "overwarning" problem makes it all the more evident that we need to provide uncertainty information (probability) in our warnings. It has been shown that warnings with graded threat levels are "reliable" in a statistical sense - that is, as the forecast threat increases, the probability of the warned event increases. If we have a national overwarning problem, as opposed to one or more local ones, then the reason for overwarning across the nation is the basic problem that unwarned-for tornadoes can kill, whereas false alarm non-tornadoes kill no one, directly. Short duration relatively weak tornadoes can kill people!

Jon Davies said...


Straight-line t-storm winds kill people, too (e.g., trees falling on mobile homes), just as often as weak short-lived tornadoes. Does it really make sense to give the public the impression that we can detect all this and sort it out accurately via the warnings (tornado vs severe) that we issue?

I'm not suggesting we can predict tornado intensities (did I say that in the post? ... _no_). But independent environment studies over the past 20 years do definitely show that stronger tornadoes are more likely when storms are ongoing in environments where parameters such as SRH and CAPE are larger. Why not enhance the wording in warnings when radar rotation is detected in such settings? That way, maybe the public could get a better sense that a Joplin-type scenario (MLCAPE > 4000 J/kg, SRH1 > 200 m2/s2, 0-6km shr > 40 kts ) truly has the _potential_ to be more dangerous than the tornado-warned Nebraska cell last evening in a very marginal setting (SRH1 < 50 m2/s2 and 0-6km shr < 30 kts). Compare the actual warning text for both storms (5/22/11 vs. 8/6/11), and you can't tell any difference at all in the threat levels, which aren't even remotely comparable.

I'm all for looking at giving probability in warnings suggesting level of threat, NOT directly for the general public, but for EMs (sirens, etc.), media mets (dissemination of info) and other weather professionals who would (in theory) know how to interpret them.

Simon said...

Can we predict significant tornado invironments with relative accuracy? If warnings (outside of relatively recent tornado emergency warnings) were based on probability due to environment then an argument must be raised on whether we can accurately predict or detect a significant tornado environment.

Given a substantial lack of surface observations across the country I don't think most forecasters can consistently predict and most importantly detect significant tornado environments with relatively good accuracy. Outflow boundaries between surface obs, significant errors in model analysis and forecasts, etc.

jimmyc said...

I guess I disagree that people cant handle more complexity. Actually they want more complexity provided they know how, when, and why to act on that additional information. They are good at perceiving risk with familiar subjects, we just need to guide the public to come to an understanding of weather risk (hazard, exposure, frequency of occurrence).
Take the tornado myth by a ridge. That person has some knowledge and is willing to apply it, but they have only partial information. They assessed the risk by noting the frequency of occurrence for that hazard to their knowledge. So we just have to communicate the other information so that their personal risk model can be upgraded. This will always require more information.

Dave Longley said...

Jon, great post. I'm growing frustrated on the increasing number of marginal event warnings. I have been an on air met in the SYR area for nearly 20 years. It used to be a major event when we had a tornado warning in our area. Now it seems that every time a line of storms comes through, there's a TOR warning.

I don't know if the public needs any more probabilistic information, they will likely be even more confused. I know in year's past these marginal events might have had a severe t-storm issued and if there was a weak spin up, so be it. Granted, in this area we don't get anywhere near the severity of weather that the south/Plains gets, but I'm afraid that if these tornado warnings keep coming, and a barn in a field is leveled or a cluster of trees is blown down after the majority of the warned area headed to the basement, we're going to run into reliability issues.

This is indeed an area of concern coming from someone who deals directly with the public. I think all of this is further complicated by the 24/7 facebook, twitter world we live in, where information is thrown at people all the time.