James B. Elsner

Florida State University

Are tornadoes killing more people?

Are tornadoes getting stronger?

**http://rpubs.com/jelsner/3710**

How do we turn these relationships into wind speeds?

1. Proportional odds model

2. Inner product of the predicted probabilities and EF scale wind range.

We solve for theta f and the beta's using a

gradient-descent maximum likelihood algorithm from the clm() function in the ordinal package of R.

**Problem: EF scale is categorical**

Solution: Invent a way to estimate wind speeds from path length and path width

Solution: Invent a way to estimate wind speeds from path length and path width

**What is causing the upward trend?**

Literature: Shear + CAPE

Shear: Upper level troughs

CAPE: heat & moisture

Literature: Shear + CAPE

Shear: Upper level troughs

CAPE: heat & moisture

**Can others reproduce these results?**

**Can the occurrence of less violent**

tornadoes be used to predict the

occurrence of violent tornadoes?

tornadoes be used to predict the

occurrence of violent tornadoes?

Evidence points to growing frequency and intensity of extreme weather worldwide due to global warming, but an effort to detect changes in the intensity of tornadoes has yet to be made.

Here we show compelling evidence for a growing trend in the ferocity of strong tornadoes across the United States.

The upward trend in wind speed is physically and statistically ascribed to increases in SST across the western Caribbean Sea and decreases in the volume of Arctic ice.

The deadly Tuscaloosa, AL and Joplin, MO tornadoes of 2011 are part of this rising trend.

The average rate of non-violent tornadoes is 55 per 100 sq km per 62 years which compares with an average rate of only 1.5 violent tornadoes per 100 sq km over the same period (less than 3%).

Violent tornado report density peaks at 2.6 per 100 sq km (62 yr) in the city but only 0.7 per 100 sq km in the countryside.

The model for the occurrence rate of violent tornadoes indicates that rates are lower by 11 +/- 4% (s.e.) for every 1 km increase in distance from nearest non-violent tornado after controlling for distance from nearest city.

The model is useful for generating a catalog of touchdown points that can be used as a component to a tornado catastrophe model.

Thank you. Questions?

Is there a connection?

How well does the model do?

Next we multiply the model-generated probabilities (one for each EF category) by a wind speed value representing each EF category to obtain a per tornado wind speed estimate.

For instance, the model predicts a wind speed of 88 (71, 108) m/s (95% CI) for the Tuscaloosa, AL, tornado of April 27, 2011. The official peak wind speed estimate from damage surveys is 85 m/s.

The methodology provides a proxy for strength using the tornado's footprint that greatly enhances measurement precision.

Is the model adequate?

Only tornadoes that were predicted correctly by the model

The estimated peak wind speed from damage surveys for the Adairsville tornado in nearby Bartow County last week was 160 mph.

Our model estimates a wind speed of 154 (141, 169) mph (95% CI).

The 35 km path length puts it above the 95th percentile of all tornadoes since 1985 and the 822 m path width puts it above the 97th percentile of all tornadoes over the same period.

What is path length and path width?

Who keeps these data?

**Is there a relationship between**

path length/width and EF scale?

path length/width and EF scale?

Regress tornado wind speeds onto Caribbean and Gulf of Alaska SST

Problem: Population bias

Cumulative logistic regression of EF scale onto path length and path width