Saturday, July 21, 2012

Probability of Tornado Warnings: A Truly Bad Idea

National Center for Atmospheric Research (NCAR). Click to enlarge.

For about five years, a group of research meteorologists in Norman, OK along with a few others in Boulder, CO and Washington, DC have been pressing to change the tornado warning system into one that is based on probabilities as depicted above.

Now, tornado warnings are a yes/no system where "yes" means take protective action.

Is a completely changed tornado warning system a good idea? I think not.

Let's deconstruct NCAR's example above:

Imagine your cellphone going off at midnight...

I don't want my cellphone going off unless someone is giving me emergency information. Too many non-emergency messages leads to the cellphone being completely turned off rather than left in sleep mode.

We have already seen this played out with NOAA Weather Radio. Lots and lots of people have purchased the radios only to junk them because they are constantly going off in the middle of the night with non-desired messages.

... at midnight...  Between 1:00 and 1:30 a.m. there is a 50% chance...

Yes, and there is a non-zero chance the thunderstorm will dissipate all together and there is an even better chance the threat will change direction.  By doing this an hour in advance we greatly increase the chance of ending up with egg on our face.

... a 50% chance that a tornado will pass within two miles of your house.

OK, you are awakened from a sound sleep. What, exactly, are you supposed to do with that message??!! What does 50% mean in this context? I contend, when given something they aren't sure what to do with, most would just go back to sleep, "Wake me up in an hour when it time to do something."

Later in this same document, the head of NOAA (the parent agency of the National Weather Service), Jane Lubchenco, is quoted as saying, "Do people hear and understand the information we think we are providing?" That is an excellent question. There is certainly a role for the social sciences to make our existing system clearer and more actionable.

The existing watch/warning system has been in place for a half-century and there are still people who do not understand it. Trying to educate people on a probability-based system (when do you go to the basement, 5%, 15%, 50%, 78%?) will take a century or more.

There may well be a place for "probability of tornado" for governmental emergency managers and other specialized users. For the public? It will do far more harm than good.


  1. Agreed 100%. We probably ought to create a public consensus on what a 50% chance of rain means before we try to explain to the public what a 50% tornado risk means. As it stands now, if you ask 10 laymen what it means if they hear "50% chance of rain" in the forecast, you'll get 10 different answers. It's been my experience that the public, in general, is very poor at understanding the statistical concept of risk (or any concept presented mathematically, for that matter.)

  2. Mike...110% right on!

    I want to discuss with you some more.

  3. Mike, I was once at a meeting where a researcher was talking about probabilities in hurricanes. He stated that the public understands probabilities because they play the lottery. I responded that they play the lottery because they expect to win, not that they understand probabilities.

    This goes beyond just tornado warnings. The NWS is also looking to extend it to the forecast (and not just precipitation). Researchers are looking to extend this to all forecast measurements.

    In a focus group I explained that this was information overload. I told them that I had spent a career trying to be certain about a forecast, not uncertain.

    I have a great example of the misuse of probabilities in a tornado event that I will pass along to you later.

  4. Instead of putting all of that into the warning itself, let NWS put up explanatory text on its website to define "Tornado Warning". Education happens when we are NOT having an emergency. We educate the high-order brain; we drill to teach the "reptile brain" how to respond when the alarm is sounded. Once the reptile takes over, it tends to not allow the kind of thinking necessary to understand complicated probabilities.

    And at midnight, about the only thing still working is the reptile.

  5. "He stated that the public understands probabilities because they play the lottery."

    It doesn't mean that at all. It simply means that the potential for a huge payoff (a multi-million-dollar jackpot) appears to be well worth the relatively tiny risk/investment involved (a few bucks for a lottery ticket that can be easily purchased at a convienience store, gas station, etc.).

    It doesn't take an advanced understanding of statistics or probability to decide that dropping $2 bucks in spare change for the chance to win, say, $200 million or more is worth it.

    Also, the "payoff" in a lottery is far more certain -- someone HAS to win at some point -- than the "payoff" in the tornado warning scenario (escaping death or injury from a tornado strike). The payoff in the second case doesn't happen if the tornado dissipates or doesn't hit anyone's house.


    1. I have no doubt that some play the lottery as you suggest. However, you may be interested in a 2008 study by Carnegie Mellon University and published in July issue of Journal of Behavioral Decision Making. The study sheds light on the reasons why low-income lottery players eagerly invest in a product that provides poor returns.

      "Some poor people see playing the lottery as their best opportunity for improving their financial situations, albeit wrongly so," said the study's lead author Emily Haisley... ..."The hope of getting out of poverty encourages people to continue to buy tickets, even though their chances of stumbling upon a life-changing windfall are nearly impossibly slim and buying lottery tickets in fact exacerbates the very poverty that purchasers are hoping to escape."

      My point is that they do not or have chosen to ignore the probabilities in hopes of winning. They do not understand.

      Back to Mike's point. What information in the probabilities is needed by the public? The fact that warnings are now polygons has greatly reduced the area warned. My view is that the warning is issued to make people aware of the danger and do something proactively to protect themselves. Once I know the tornado is in the vicinity, why do I need to know the areal probability?

  6. Agreed Mike. If the public still cannot sort out what a watch or warning is, and some still expect to hear sirens when threatening weather is afoot, how can we expect them to understand that given a set of atmosphere dynamics, we expect a 50 percent chance of (pick your hazard) to occur within X miles of a point?

    Point: we still get snookered today with the quick spin-up twisters that fall out of a QLCS or an MCS, and by the time the a warning is out, it's long gone and we're chasing ghosts for 30 minutes or more on TV/radio/web. Or, a storm decides to track differently than expected, so your warning fan just went out the window, and you're scrambling to get a handle on it again and draw a new one.

    The 2011 Super Outbreak was definitely a wake-up call that SOMETHING needs done, but probability-based warning isn't the way to go. Seems like that just totally disregards a "course of least regret". Make folks understand that although imperfect and at times a hassle our current warning system is now, all the wolf cries are irrelevant when a Moore F5 or Tuscaloosa F4 really shows up to ruin your day.

    Forecaster Mike - Jackson, MI

  7. Thanks for all of the excellent comments!


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