Enhanced Fujita Ratings Debate Thread

[Grand Poo Bah]

If we stick with Empirical Approach to Evaluating the Tornado Fragility of Residential Structures for now we can examine some interesting implications.

The assumptions behind this theoretical that the treefall method used to independently estimate the wind in the paper is accurate (different treefall models produce different results) and that the distribution shown by the fragility curves in the paper are representative (whereas the low-bound biased approach of EF-scale assessments effectively assumes they're more left skewed).

What I've done is draw a dashed line for a theoretical 70 m/s (157 mph) tornado, and placed arrows over 4 DoDs (6, 7, 8 and 9) showing their ranges in the Enhanced Fujita Scale, produced by expert elicitation mainly based on engineering analysis. The vertical axis is the probability of a DoD for the given windspeed:

View attachment 54580

It can be seen that the DoDs have roughly the following probabilities:
DoD 6: 29%
DoD 7: 18%
DoD 8: 17%
DoD 9: 22%
Below DoD 6: 14%

The most likely DoD to be observed is 6, even though the windspeed lies outside the upper bound for that DoD. Only two DoDs are capable of producing an estimate that includes the true windspeed (8 and 9). Therefore, if our theoretical tornado was to encounter a single house where we otherwise know nothing about the structure except that it conforms to these fragility curves, an EF-scale analysis has a better than even chance of definitely underestimating the windspeed, as windspeed lies outside the upper bound for DoDs 7 or lower. Even the remaining two DoDs have enough room to produce underestimates.

Excellent work here. This is a really useful visual. I'd love to see it paired with empirical research that shows the average wind speed assigned to each DOD. If we looked at this year alone we'd see about 85% of all DOD 9s below your 157 mph line, with Enid being the only exception. Is that why you chose this windspeed to represent your example?

 

[Grand Poo Bah]

If we stick with Empirical Approach to Evaluating the Tornado Fragility of Residential Structures for now we can examine some interesting implications.

The assumptions behind this theoretical that the treefall method used to independently estimate the wind in the paper is accurate (different treefall models produce different results) and that the distribution shown by the fragility curves in the paper are representative (whereas the low-bound biased approach of EF-scale assessments effectively assumes they're more left skewed).

What I've done is draw a dashed line for a theoretical 70 m/s (157 mph) tornado, and placed arrows over 4 DoDs (6, 7, 8 and 9) showing their ranges in the Enhanced Fujita Scale, produced by expert elicitation mainly based on engineering analysis. The vertical axis is the probability of a DoD for the given windspeed:

View attachment 54580

It can be seen that the DoDs have roughly the following probabilities:
DoD 6: 29%
DoD 7: 18%
DoD 8: 17%
DoD 9: 22%
Below DoD 6: 14%

The most likely DoD to be observed is 6, even though the windspeed lies outside the upper bound for that DoD. Only two DoDs are capable of producing an estimate that includes the true windspeed (8 and 9). Therefore, if our theoretical tornado was to encounter a single house where we otherwise know nothing about the structure except that it conforms to these fragility curves, an EF-scale analysis has a better than even chance of definitely underestimating the windspeed, as windspeed lies outside the upper bound for DoDs 7 or lower. Even the remaining two DoDs have enough room to produce underestimates.

It's also worth noting DOD 10 would have a 0% chance of producing this wind speed estimate, but a statistically significant amount of homes are given those ratings anyways. (Ie. Lake City and Matador). Unfortunately, a lot of DOD 10s are given the label of DOD 9 so lower wind speeds can be assigned, which drastically increases the difficulty of empirical analysis in that range.

 

[Grand Poo Bah]

I’ve read this paper over a few times now and finally it’s clearer to me. It was a bit hard for me to decipher, but basically it seems like scientific evidence of something I have been saying for a long time, which is:

“The windspeed at which something COULD potentially begin to happen, is very different from the windspeed at which something TENDS TO happen”

For example, tree damage. The EF scale currently has total stubbing and debarking occurring largely in the EF2 to EF3 range. In real life though, actual case studies that compare tree damage to nearby structural damage, computer based physics modeling, and general statistical correlation from my personal observations over the years all suggest that total stubbing and debarking has a much stronger correlation with the EF4+ range. Now can really severe stubbing and debarking happen in the EF2 to EF3 range? Yes, but it’s quite rare and the only examples are cherry picked outliers. Thankfully the concept of severe debarking/stubbing as an EF4 indicator is slowly catching on, but you get my point, and I’m sure the same sentiment can be applied to other DIs as well. The scale and its application operates on a “windspeed at which this degree of damage could begin to happen” basis rather than a “windspeed at which this usually happens” basis. If the objective is finding the most likely actual max windspeed of the tornado based on damage, this approach isn’t favorable.

But with this said, at this stage I don’t really care about the wind speed estimates handed out in actual surveys themselves, because I’m not at all confident that they are genuinely accurate. They are simply a safe, conservative estimate rather than an accurate reading of the tornado’s actual intensity. Instead, I more prefer to rate and discuss tornadoes via a spectrum of EF-scale rankings rather than trying to assign an accurate windspeed estimate. For example, I prefer to simply say “low-end EF4 tornado” rather than “170 MPH tornado” because I’m not so sure that 170 MPH tends to be the most likely maximum real life windspeed for any given tornado that produces that kind of damage. When I do use numbers and wind speeds, I’m actually more just using them as placeholders to convey where the tornado was within the spectrum of an EF scale ranking (high-end, low-end, etc) rather than actually believing that this was the actual maximum windspeed.

But yeah overall, the paper seems to back up my thinking on this topic: EF scale windspeed estimates that get handed out are the lowest conceivable failure point of the DIs, but are conveyed to the public as the tornado’s maximum windspeed, when in reality these can be and often are two very different things.

I'm one that cares far more about accurate windspeeds than damage classification, but I'm mostly resigned to the fact that until we get consistent classification across WFOs, accurate wind speeds are much more difficult to achieve across the entire tornado database.