Tuesday, March 17, 2015

A Six-Hour Tornado Forecast

Springfield News-Leader
In 2011, 15 were killed at Joplin’s St. John’s Mercy Medical Center by the F-5 tornado that moved across the city. In my interviews with hospital personnel, they said they were not aware the tornado was approaching until just six minutes before the tornado struck. There wasn’t sufficient time to move patients into safe areas. Surgery was in progress when the tornado hit. The outdoor generator was destroyed, so the operating room was plunged into total darkness.

Two years later, three were killed at the Moore, Oklahoma, hospital by the May 20, 2013, tornado.
[Rick Smith of the National Weather Service in Norman says the information I relied on was incorrect and that no one died at the hospital. While he did not provide a link, he is an expert on this event. I am happy to correct the record.]

These tragic examples demonstrate that hospitals and certain other industries need more notice of a tornado than the 10 to 15 minutes that works so well for the public-at-large. AccuWeather Enterprise Solutions has been searching for a solution to this challenge and we believe we have found it in a high-resolution meteorological model developed by Hazardous Weather Notifications, LLC (HazWx). AccuWeather purchased the company's assets in January. This was topic of a story in today’s Wichita Eagle business blog. 

In the story Eagle’s story, I mention hospitals and discussed how the model made an accurate a six-hour forecast of a tornado. A meteorologist with the Storm Prediction Center in Norman, Oklahoma, expressed skepticism via Twitter that our new model could do what I indicated. So, I replied that I would post the pertinent case history on my blog. BTW: An important part of science is exchanging information and ideas and seeing if those hold up to scrutiny. 

On December 23, 2014, the model was run at 7am 6am [forgot we were on standard time in December] and produced a forecast of  high helicity (turning in a thunderstorm's updraft) from north of Lake Ponchartrain across southeast Mississippi. This got our attention because the values exceeded the thresholds that our experience indicated are associated with tornadoes. Plus, the general weather conditions were favorable for tornadoes across the region.

The model was again run at 1pm noon with an extraordinary result. Extremely high helicity values were forecast along an intermittent path that (based on rules we have developed) overlapped the morning forecast. This, combined with an atmosphere favorable for tornadoes, would have prompted our meteorologists to issue a tornado forecast (not a “warning” as we generally think of them) for hospitals or other clients in the projected path area that needed extra “lead time.” 
Annotated copy of my storm notes
Lead time is defined as the interval from the receipt of the forecast or warning to the time of arrival of the storm. For a hospital, this might mean that extra people should be ready to move patients into safe areas and elective surgery should be postponed.

As the afternoon unfolded, tornadoes and damaging winds occurred along the path of the bright red arrow. Five people lost their lives in the town of Columbia, MS. The National Weather Service offices in New Orleans and Jackson did a post-storm survey and they were gracious enough to provide photos of the damage at various points in that path. Those are the thumbnail-sized photos.
The National Weather Service has a model known as the HRRR (left) that was run at the same time and also produces a helicity forecast. The HRRR’s helicity forecast was well to the west of the path of the tornado.

The HazWx model does not work in all situations. Fortunately, the times it is likely to perform poorly are fairly consistent and known to our meteorologists. In those situations, we would not issue a tornado forecast. That stated, some of the forecasts made by the model have been nothing short of astounding.

We continue to test the HazWx model to get more experience with it. We will be providing these forecasts to our clients this spring when nature creates the necessity. All of us at AccuWeather Enterprise Solutions are excited by this opportunity to provide this new level of actionable information to our clients with the need for extended lead time tornado forecasts.


  1. That is exceptional!

    As the model continues to be used, is there a margin of error that you've come up with (+/- 10, or the like) for this? Or is it so new that it is still in the experimental phase?

  2. I don't wish to comment until we have a little more experience. There are some tricks to using it which we are learning.

  3. If proven, the HazWx model could be a storm chaser's dream. On the flip side, hopefully a potential tornado area won't become flooded with amateur chasers...

  4. As someone who's looked a lot at these types of output fields from high-res models, I'm quite skeptical that they hold the value in terms of tornado prediction that you have implied here. They certainly do provide a lot of value to forecasters and other knowledgeable users about storm development, intensity, convective mode, etc., but the tornado prediction problem has certain aspects that we are unable to address with these types of models. I hope that before you sell the capabilities of such a system to customers and the public, you take a thorough examination at the underlying forecast skill, and not rely on subjective interpretations from a handful of cases. Also, a deterministic forecast is really not appropriate for convective-scale forecasting, since there needs to be some representation of the, often large, uncertainty associated with forecasts on these scales. These uncertainties are something your customers could use to make better decisions. I've summarized some of my thoughts in a blog post here: https://medium.com/@sobphd/thoughts-on-a-6-hour-tornado-forecast-f9a6860ec5b4

  5. Hi Ryan, thank you for alerting me to your blog posting. It is too bad it does not allow comments as far as I can tell. I would appreciate the opportunity to correct the record for your readers.

    While I welcome critical comments, you've made a number of assumptions that are not correct. As I have consistently said, this is "a" (singular) forecast.

    I never said or implied that the model works every time. Quite the opposite. In my description of the forecast above wrote, "The HazWx model does not work in all situations. Fortunately, the times it is likely to perform poorly are fairly consistent and known to our meteorologists. In those situations, we would not issue a tornado forecast."

    The third incorrect assumption is this is a singular case. The model has made numerous excellent forecasts over the past year.

    The fourth incorrect assumption is that we are going to sell these to the public. We have no plans to do so.

    Finally, your statement, "the notion that a tornado was successfully predicted... is just not true." is incorrect. I imagine the families of the five people that died in this storm would have loved to have had six-hours of notice they were at enhanced risk of a tornado.

    I realize the people at the 'National' Weather Center and OU are used to looking at severe weather forecasts in a certain way and that is fine. However, that is not the only way to look at them nor is it the only way that is scientifically correct. In future years, the private sector is going to, more and more often, lead the way in meteorologically innovation. Being open-minded to different approaches will be a great way to advance your career.


Note: Only a member of this blog may post a comment.