More on tornadoes:

Records, the Fujita scale, and our observations


Chuck Doswell

Created: 13 March 2000 Updated: 23 September 2006 : minor revisions and moved this over from its former site.

As usual, this page contains my opinions only. It does not represent anything official. Any discussion can be sent to:


The following is a real e-mail "conversation" I had recently (edited lightly to protect the correspondent's identity and to remove some of my more intemperate statements):

Chuck, since you are an experienced stormchaser and scientist, I have a few questions. One, how close was the Oklahoma City tornado of May 3, 1999 to being rated an F-6? 

For my thoughts on the F-6 question, see item B.11 here.  

I know this this level of a tornado has never been officially documented but it seems this tornado was awfully close to that level, at least for a brief period. Homes being COMPLETELY removed and steel beams bent like coathangers - I couldn't believe it!!

Regrettably, such damage is what happens with violent tornadoes. Most people have never experienced a tornado, much less a violent tornado, and find the damage associated with one beyond their belief. However, this just reflects the lack of appreciation most folks have for such events, until they have seen it for themselves. The OKC tornado of 3 May 99 was certainly a violent tornado, but it probably is NOT the most violent ever - however, we really have no way of KNOWING what is the most violent tornado ever, for a lot of reasons. I may have to work up a Web essay on this topic, since it has become a common question since 3 May 99.  

I had watched >this week the special on the Learning Channel about the tornado. I know that the doppler-on-wheels at one point observed winds at 318 MPH and this is right on the threshhold of being that level of a storm. Has there been other storms near this intensity?  

At this point, the validity of the Doppler-On-Wheels "measurement" of 318 mph is open to some question. Even if it IS a valid measurement, however, the Doppler radar observations apply to winds somewhere well above the surface (which is where the Fujita scale winds are supposed to apply, because that is where the damage is being done).[5] Hence, even if 318 mph *is* a valid observation, it does NOT mean that the Oklahoma City tornado of 3 May 99 was very nearly an F-6. As noted at the Website above, there are numerous misinterpretations of the Fujita scale, and this is just the latest in a long series of such misinterpretations.  

I know the tornado at Jarrell Texas in 1998 was also a very powerful storm and had some unbelievable damage. Has there been storms in the past this strong but we didn't actually have the documentation (maybe due to extensive research) like we do on some of the storsm that are happening now.  

Bingo! We are just now beginning to obtain some "measurements" with the DOWs. It will take a long time to accumulate enough such measurements to begin to understand what is the real distribution of winds in tornadoes.  

I remember as a child (I am YY yo) reading books and how some of them said that tornado winds were estimated at 400 to 500 MPH. I know that due to modern technology that these myths have been disclaimed but is it POSSIBLE a tornado could ACTUALLY reach that strength of wind? Or is it out of the realm of possibility?  

At this point in our scientific understanding, it seems quite unlikely that winds of 400+ mph are possible. A rule in science is to never say "never" but we have no REASON to believe that such winds are ever created. In the absence of such a reason, it seems prudent to say that such winds are not likely.  

Another question is what is the widest single vortex (or at least not two separate tornadoes side by side) ever observed? I know that there have been some observed to be a mile or a mile and a half wide, but have there been any wider? I had heard that one in XX in 1984 (that is where I am from) was thought to be over two miles wide. I think it went through RR and MM.   The problem with this question is: Do you mean the width of the VORTEX, the FUNNEL CLOUD, or the DAMAGE?

It's a bit long-winded and perhaps technical in places, but consider reviewing the material here:    The tornado is the WIND, not the CLOUD - but wind is invisible unless it contains debris and/or cloud, and for most tornadoes, accurate measurements of ANYTHING are simply not obtained. Thus, when asking what was the widest ever, the short answer is - who knows? WE certainly don't. As for the widest ever DOCUMENTED, opinions vary, but something in excess of one mile for the damage path width has been observed on several occasions.

This correspondence is typical enough that, given the stimulus of the "observations" of a 318 mph windspeed in the 3 May 1999 tornado by the Doppler on Wheels (DOW) mobile Doppler radars, it seems the time is ripe for yet another discussion about tornadoes:

Our Scientific data

Let me begin by reviewing some essential aspects of our scientific data concerning tornadoes in the United States. The history of our data is long and complex, but it is clear that the accumulation of data concerning tornadoes was limited primarily to spectacular events that resulted in major damage and casualties for many years. This began to change in the 1950s, with the inception of tornado forecasting by the Severe Local Storms Unit (SELS) of what was then called the Weather Bureau (now, it's the Storm Prediction Center of the National Weather Service). There are many sources for information about our scientific database. I'm providing a selection of links here:

The Tornado Project

A paper about the history of storm spotting

The Storm Prediction Center

NSSL's Severe Weather Climatology Project

National Climate Data Center (publisher of Storm Data)

To provide a short summary from my personal viewpoint, the database regarding tornadoes in the United States is, by far, the best such database in the world, but from a scientific viewpoint, it sucks! The information we have reveals several unpleasant characteristics (in no particular order):

  1. It contains clear evidence of trends that are virtually certain not to be associated with real variations in the occurrence of tornadoes
  2. The quality and quantity of the observations varies substantially with time and space. This is the result of large variability in:
    • The level of knowledge and enthusiasm of the person recording the data
    • Population density
    • The level of tornado awareness in the areas hit
    • The politicsof tornadoes [1]
    • The level of knowledge and enthusiasm of the person(s) seeing the event, especially their willingness to report what they've observed
    • The "observability" of tornadoes as a function of specific, local factors (including such things as visibility, height of cloud base, time of day, terrain features, the presence/absence of highways, the presence/absence of trees, buildings, and other obstructions to sight). This also includes issues associated with recognizing that the event is a tornado. As discussed in my essay on "What is a tornado?" it is not always easy to know that what is being viewed is actually a tornado!
    • The amount of interaction with humans and their constructions
  3. The vast majority of the information in the database is drawn from reported observations (or "reports") by non-meteorologists ... the database stores no information about the source of the observations, so it is generally not possible to know what level of confidence can be assigned to the reports
  4. Scientists and/or engineers survey only an extremely tiny fraction of all tornado events, so most of the reports have not been subjected to a careful quantitative analysis. For the vast majority of the data, the quantitative information about the events is based on surveys for which the quality level can range from pretty good to superficial (at best). Some reports get "default" ratings because they've not been surveyed by anyone
  5. Our system of describing tornado events continues to be limited by the fact that the first digital tornado databases were on IBM punchcards, with an 80-character limit on the information that could be included on a card
  6. Our ability to describe the characteristics of a tornado is limited to a single number for: (a) the intensity of the damage, (b) the width of the damage path, and (c) the length of the damage path, even though it is well-known that the first two of these quantities vary considerably over the track of a tornado; the path length depends on precisely how the start and end points are determined (see below)
  7. Most reports are given as single point observations ... this is especially egregious for hail and wind reports, but is true for many tornadoes, as well (the locations given are for the "touchdown" points). Obviously, real events traverse finite areas and are not well-represented by points
  8. The persons currently responsible for our severe weather database are the same persons issuing warnings for severe weather [2]

Consider the following characteristics of the damage path length and width observations (note that width does not refer to the funnel cloud width!). The determination of the path width depends on how the observer defines the "damage" path; obviously, there is some threshold of damage needed to define the damage path characteristics at all. If there is variation of the structures and objects in the path, the definition of the "path" is easily seen to be more than a bit vague. The ambiguity associated with variations of objects in the path will be compounded by variations in the wind speed in space and time. These factors influence just where an observer sees the "path" to be, the lateral boundaries of which can be quite complex, with swaths of damage often seen to swirl in towards the center of the damage path from the periphery. Different observers will have different thresholds for how they define the path. Given that there is some vagueness about what specific number is used to describe the width (Is it the maximum width of the path? Is it the average width?), the total area within the path can't be determined with a lot of precision. A similar problem applies to the length of the path ... with path length, it can be difficult to know where one tornado ends and another begins. Is it all one tornado, with gaps in the damage, or is it a series of tornadoes? This has been discussed elsewhere. For tornadoes given only a cursory survey, or no survey at all ... the majority of tornadoes ... the numbers are derived from little, if any, hard data.

The upshot of all of this (as well as other factors I may have left out) is that the database is pretty abysmal when it comes to providing substantial information about the real distribution and character of historical tornado events. Any attempts to use the raw data for quantitative analysis are essentially doomed to be misleading or even grotesquely incorrect. In order to draw much useful from the database, a heavy filtering must be imposed. Furthermore ...

It's impossible to say with any certainty what has happened in the past. Given that most information about tornadoes in the past comes from non-scientists, we know very little about tornado characteristics of the past. In particular, we have no way of knowing what are the absolute largest values of such things as windspeed, damage path width and length, speed of movement, and so on. To say that a given event, like 3 May 1999 has the highest windspeed ever recorded isn't really saying much of substance ... for virtually all tornadoes in the past, we have no information about their windspeeds. [3]

The Fujita Scale revisited

I've already expressed my concerns about the Fujita Scale elsewhere (item B.11 in my Pet Peeves essay). However, let me review the essentials:

The scale is uncalibrated ... there have been no carefully-done scientific/engineering studies made to relate damage to windspeed. Hence, Fujita's windspeed categories are essentially arbitrary.

The transitions between F-scales are not really "hard" boundaries, but that is what Figure 1 seems to imply, does it not? A 157 mph wind does not do substantially different damage than a 158 mph wind, as the transition between F-2 and F-3 seems to imply. Rather, there is a range of windspeeds associated with a particular kind of damage. Hence, rather than looking like Figure 1, the real situation might resemble something like Figure 2.

Figure 1

Figure 2

Figure 2 suggests that a given amount of damage can be caused by windspeeds within some distribution (shown here as roughly bell-shaped curves). Generally speaking, damage is the outcome of an interaction between a tornado's winds and physical objects (like trees, buildings of various sorts, signs, etc.). Please note that Figure 2 is only a schematic ... the curves should not be accepted literally. We do not know how to draw such curves or what the precise distributions of windspeeds might be to create a given level of damage. The figure is simply to suggest the general idea that, for example, "F-3" damage could be caused by winds that might be much lower (or much higher) windspeeds than those indicated by the traditional interpretation of the F-scale. Note that the curves tend to converge at the upper and lower ends, suggesting that there is some minimum speed required to do damage (albeit an ill-defined minimum value), as well as some maximum windspeed in a tornado (also ill-defined).

For a particular fixed value of the windspeed, the damage depends on the characteristics of the object being hit by those winds. The duration of the wind, the gustiness of the wind, the objects (if any) that are being carried by the wind ... all these have an effect on the damage produced by that particular wind. A given windspeed does not always produce equal damage.

Conversely, if we hold the damage fixed, the windspeeds producing that damage can vary considerably from case to case, for the same reasons. Any attempt to relate damage to windspeed that does not account for this sort of variability is doomed from the outset to yield only ambiguous results.

When engineers look at damage, they focus on the structural factors that cause variations in the damage, and tend to assume that the windspeeds are constant. When meteorologists look at damage, they focus on the windspeed variations that cause variations in the damage, and tend to ignore the structural variations. This is why both engineers and meteorologists should collaborate on damage surveys! The engineering issues are complicated, and it's difficult to be confident in an engineering assessment when structures are destroyed ... it becomes hard to know what the structural details were when the evidence has been scattered in the wind. However, the meteorology is also complicated; there is close to no detailed knowledge of the temporal and spatial variations in the wind field of real tornadoes. When two complex issues (windfields and structures) interact, the interaction generally becomes more than twice as complex as each factor by itself! It's basically impossible to explain everything that is observed.

The consequence of all this complexity is the lack of precision in estimating windspeeds (the usual measure of tornado intensity). The F-scale estimates are basically damage estimates and any relationship to windspeed is pretty shaky for any given case. This is not to say there is no relationship between windspeeds and damage ... just that it's not obvious that the F-scale equates to windspeed in any unambiguous way (especially for the strong-to-violent events).[4]

Record-breaking events?

Without detailed knowledge of the history of tornadoes and without detailed knowledge of the windspeeds associated with tornadoes, the achievement of any measurement holds the potential for a record. As recently stated by Prof. Wurman of the Oklahoma University School of Meteorology, in reference to the devastating tornado of 3 May 1999 in the Oklahoma City area,

"We don't know this was the strongest tornado ever, just that no other had ever been measured with faster winds."

If, in fact, there were hundreds or thousands of tornadic windspeed measurements, and a particular one was the highest value ever recorded, an observation of the highest ever recorded windspeed might say something very special about that particular event. However, this is not the case ... of the thousand or so tornadoes that are reported in the United States over a given year, only a handful of them involve quantitative measurements of windspeed, and those tend to be limited to relatively brief segments of the total lifetime of the events. Who knows what went on in the unmeasured events, or those parts of tornado events that were not measured quantitatively? All we have to go on is the damage, for the most part and, as I've been trying to suggest, damage is a notoriously unreliable measure to use in gauging tornado intensity (i.e., windspeeds).

It's impossible to go back and determine much about historical tornado events (like the 1925 Tri-State tornado, the 1947 "Tri-State" outbreak that included the Woodward, OK event, the "Palm Sunday" Outbreak of 11 April 1965, the Super Outbreak of 3-4 April 1974, etc.). We have to depend on newspaper accounts and such, attempting to interpret what happened in terms that relate to events as we now see them. This is an unreliable exercise that we engage in so that we can have some semblance of a large database with which to expound on the significance of current events. Even the number of tornadoes on a given day is likely to be determined differently today than it would have been in years past. How many tornadoes would have been recorded in historical outbreaks of the past? There is no way to know, regrettably. All of this makes it a pointless exercise to talk seriously about "record-breaking" tornado events ... such hyperbole is common in media presentations but there is little scientific basis for talking about tornado records.