Highest risk corridor?

[Dave Gallaher]

http://blog.al.com/breaking/2013/09/birmingham_huntsville_part_of.html#incart_river_default

 

[MClarkson]

Most of the previous studies have a fairly even distribution of large tornadoes, with maximum alternating between Kansas and Alabama and points nearby depending on the various dates and strengths chosen. Did they all neglect path length? And how skewed is the path length towards being longer in regions with higher population density?

 

[Mike Smith]

I believe the fact this study begins in 1973 has an effect on the findings. 1947, Woodward; 1953, Waco, Flint, Worcester; 1955, Blackwell-Udall; 1957, Dallas and Ruskin Heights; 1966 Topeka.

 

[rdale]

...and I am not certain that tornado tracks and path lengths in the 70's were done with the same rigor that tracks in the 2000's were recorded...

 

[Shawn Schuman]

...and I am not certain that tornado tracks and path lengths in the 70's were done with the same rigor that tracks in the 2000's were recorded...

Yeah, that's an understatement. Path lengths even as recently as the 90s are sometimes suspect, and path widths even moreso. Not to mention the change from median path width to maximum width. I've actually been working on something similar (and a lot less scientific) for my blog, later tonight or tomorrow I'll post some of the stuff I've got so far. I think it's pretty obvious that the southeast (particularly north/central Alabama and Mississippi) is the focal point for violent, long-track tornadoes, but there are several things at work there. First, because forward speeds tend to be higher in the southeast, path lengths are often longer even when duration may be similar. A tornado that's on the ground for two hours in Oklahoma, moving at 25 mph, is going to wind up with a much shorter path than a tornado that's on the ground for an hour and a half in Alabama, but moving at 60 mph. There's also the obvious fact that it's often easier to document the full path length (and width, for that matter) of a tornado tracking through the relatively populated and heavily forested areas of the southeast than in the sparsely populated and relatively barren areas of Texas, Oklahoma and elsewhere. These same factors contribute to tornado ratings in each area, with the southeast presumably having many more potential damage indicators in the way.

So, the short version is that there are any number of variables that prevent us from being able to draw many conclusions. One thing I think we can say is that Dixie Alley belongs on equal footing with Tornado Alley in terms of violent (and especially deadly) tornadoes, but I think most of us already knew that.

Actually, here's one thing I've been working on. I'll share some of the other stuff later. I did something similar to the study in this article, actually. I took the SPC tornado database (1950-2012) sorted by county, and then filtered it to include only significant (F2+) tornadoes that killed at least one person. Then I just used a simple equation that I called "Combined Area Intensity": [Fujita]*([Length]+[Width])/[Number of Tornadoes]. It's obviously not the best way to do things, but I thought the results were interesting nonetheless. You may need to click the direct link if you want to see it full-size.

http://i.imgur.com/FSOqgJZ.png

And in a similar vein, this is a graph showing the results by state. The color of the bars is correlated to the number of fatalities for each state, with darker red being more fatalities.

 

[Rob H]

I believe the fact this study begins in 1973 has an effect on the findings. 1947, Woodward; 1953, Waco, Flint, Worcester; 1955, Blackwell-Udall; 1957, Dallas and Ruskin Heights; 1966 Topeka.

Very good point. What is it about '73 that makes it relevant to the study? That it conveniently leaves out some big tubes in the Plains while including both Super Outbreaks?

edit: Ah, 1973 is the first year they included length/width in the data. So that makes it a good candidate, but it still has an inherent bias due to some "lucky" timing. I'm guessing '75-'10 would look rather different.

 

[Logan Karsten]

While I appreciate this study and agree that tornadoes that do occur in Dixie Alley tend to be more dangerous on average, I feel like this study is geared to make it sound like Dixie Alley is the real tornado alley of the U.S., when it's more complicated than that. Tornado statistics can be done in so many ways to make different parts of the country stand out, as was done in this study. However, I think it's fair to say that tornado alley is really where tornadoes frequently occur with more dangerous tornadoes occurring on a less frequent basis across Dixie Alley. Again, this study was done by UAH where they are trying to build up a severe weather research institution outside of Norman, so it makes sense that this study is trying to draw attention to Dixie Alley.

 

[rdale]

From what (little) I know of Dr Coleman, he would not do that. Let alone present it for peer review...

 

[Patrick Marsh]

Very good point. What is it about '73 that makes it relevant to the study? That it conveniently leaves out some big tubes in the Plains while including both Super Outbreaks?

edit: Ah, 1973 is the first year they included length/width in the data. So that makes it a good candidate, but it still has an inherent bias due to some "lucky" timing. I'm guessing '75-'10 would look rather different.

The authors claim that '73 is the first year in which the NWS used the F-scale in real-time. However, it was not adopted at every office until '75.

 

[mikedeason]

I think a lot of people are reading way too much into the study. I think 99.9% of most people in the traditional Tornado Alley as well as Dixie Alley would be more than happy to bestow such a dubious distinction to some other geographical location than their own.

 

[Rob H]

I think a lot of people are reading way too much into the study. I think 99.9% of most people in the traditional Tornado Alley as well as Dixie Alley would be more than happy to bestow such a dubious distinction to some other geographical location than their own.

The article itself states that this could perhaps be used to secure more funding for UAH to study tornadoes in Alabama. Hope they invest in some chainsaws!

 

[Shawn Schuman]

I've been tweaking this for a while and I still need to change a few things (red extends too far southeast, for one thing, and the green needs to be trimmed back in a few spots), but I figured I'd share it since it's relevant to the thread. It's based on the same sort of data I've been working with and that I posted earlier, with strong/long-track tornadoes weighted more than sheer number of tornadoes, and some consideration given to fatalities as well. Just my rough interpretation of a tornado threat map. Orange is sort of a moderate risk, while red is what I would consider the "real" Tornado Alley. White is the highest threat area, and the pink areas must have done something to make mother nature very, very angry over the last 60-ish years.

I think the back-and-forth over which area is "worse" is a bit silly, but it's clear to me that "Tornado Alley" ought to include portions of the southeast. I have another version of this map overlaid on a population map, I can post that later after I'm done with it. I'm sure it's no surprise to see that more populated areas tend to have a somewhat higher (apparent) tornado risk.

 

[Jeff Duda]

That's some great work, Shawn! However, I would suggest making a few changes which would probably make your results look more realistic (and also go along with what some other posters have said). This also applies to the study itself.

Normalize your data. That means you should divide your result by a quantity that affects how the result was obtained. For example, as MClarkson brought up in his post, population density. More people will die in tornadoes if more people are there to get hit by tornadoes. Population density is clearly higher in MS and AL than KS and OK. Also, considering the recent debate over the El Reno tornado of 31 May, weighting by (E)F-scale rating probably won't unmask the trend either for the same reason as not normalizing by population density. Tornadoes get rated very low if they don't hit anything. Manmade structure density is much higher in MS and AL than KS and OK.

Finally, I noticed in your first post that you used the sum of path length and width. Why not compute area by multiplying them instead? Then your formula would have meaningful units.

If you read partway down, they say that they re-ran their experiment leaving out 27 April 2011. The result - the two highest contours disappeared. This says a lot about how much individual events weighted the overall results in this study. I would like to see the results of this study without 24 April 2010 in there, too (Yazoo City 148 mi long tornado).

 

[Shawn Schuman]

That's some great work, Shawn! However, I would suggest making a few changes which would probably make your results look more realistic (and also go along with what some other posters have said). This also applies to the study itself.

Normalize your data. That means you should divide your result by a quantity that affects how the result was obtained. For example, as MClarkson brought up in his post, population density. More people will die in tornadoes if more people are there to get hit by tornadoes. Population density is clearly higher in MS and AL than KS and OK. Also, considering the recent debate over the El Reno tornado of 31 May, weighting by (E)F-scale rating probably won't unmask the trend either for the same reason as not normalizing by population density. Tornadoes get rated very low if they don't hit anything. Manmade structure density is much higher in MS and AL than KS and OK.

Finally, I noticed in your first post that you used the sum of path length and width. Why not compute area by multiplying them instead? Then your formula would have meaningful units.

If you read partway down, they say that they re-ran their experiment leaving out 27 April 2011. The result - the two highest contours disappeared. This says a lot about how much individual events weighted the overall results in this study. I would like to see the results of this study without 24 April 2010 in there, too (Yazoo City 148 mi long tornado).

Thanks for the suggestions, Jeff. That's actually what I've been working on, it's just taking a while to get the data and get everything sorted out. As for the other points, I think there are always going to be those kinds of inconsistencies (which is why I'm so confused about the El Reno debacle). I suspect that, if we were somehow able to accurately record the intensity of all tornadoes, not only would we find a higher proportion of strong tornadoes overall, we'd also find that many more of them occur in the Plains than we're currently recording. I considered using the pink contour only for Central Oklahoma because I think it's probably the "hottest" of the tornado hotspots, but I think Northern Alabama also stands out from the rest of the country. The difference seems to be that places like Oklahoma and Kansas tend to see strong/violent tornadoes fairly consistently from year to year, while the southeast gets the majority of their violent tornadoes during relatively few massive outbreaks.

The Yazoo City tornado certainly skews the data a bit. I suppose a path length of 149 miles and width of 1.75 miles tends to do that. Its effect is kinda similar to Joplin's when you're working with fatality data. Also, I did use the area, I just made a mistake in my post. Anyhow, here's a county map of the "Combined Area Intensity" that doesn't include 4/24/10 or 4/27/11. The difference is noticeable to be sure, but not especially significant.

 

[MClarkson]

I don't know if this data is easily accessible to anyone, but insured losses (adjusted for inflation) per person would be an interesting and perhaps easy way of comparing 2 different areas? I would think it would show a similar trend, with the most insured losses per capita occurring in the areas in question, from say OKC through Alabama.