May 21, 2013

The Lessons of the Moore Oklahoma Tornado

On Monday, an EF-5 tornado struck Moore, Oklahoma resulting in substantial loss of life, hundreds of injuries, and economic losses that will run into the billions of dollars.  The storm had winds reaching 200-210 mph, was on the ground for 17 miles, and was observed from 2:45-3:35 PM CDT (19:45-20:35 UTC).

How did the National Weather Service and the weather forecast enterprise do?  Can we do better?  This blog will provide some analysis and a few recommendations.

Let me say at the outset, a lot went right and this event demonstrates the substantial improvements in weather prediction technology during the past decades. 

For several days before, based on a range of numerical forecast models and substantial knowledge of severe thunderstorms, the National Weather Service forecasters at the Norman, Oklahoma NWS office and staff at the NOAA/NWS Storm Prediction Center (SPC) had predicted the threat of severe thunderstorms on Monday over Oklahoma.   Here is the graphical outlook for severe convection released on Sunday.  Central and eastern Oklahoma was clearly in a high risk area.
And the Storm Prediction Center's discussion highlighted the threat for Monday

One of my graduate students, Luke Madaus, happened to be in Oklahoma on Sunday and several people commented on the potential severe storm threat for the next day.

The next morning it was clear that the threat of severe weather was enhanced.  Many of the numerical models showed the development of strong convection, although their solutions differed considerably in strength and position.  The radiosonde (balloon-launched weather instruments) sounding at Norman, Oklahoma showed extraordinary instability (CAPE of roughly 5000 Joules per kilogram and plenty of vertical shear; CAPE stands for Convective Available Potential Energy.  Northwest locations rarely gets above a few hundred, a few thousand is very large, 5000 is extreme).  PLUS, there was a frontal boundary and a dry line that intersected near Norman, Oklahoma.  A very, very big threat. 
Our nation is lucky to have the best severe storm forecasters in the world, backed by world-leading research at the National Severe Storms Lab, the University of Oklahoma, the National Center for Atmospheric Research, and many others).   They proved themselves on Monday.  The 11 AM CDT forecast (communicated via YouTube among other ways), painted out the threat and even talked about dangers to schools. (see image, click on it to see the video).

You will notice that the warnings of severe thunderstorms was over an area.  Our current level of forecasting technology, coupled with the substantial uncertainty and the chaotic nature of convection, made it impossible to do better.  But an extraordinarily valuable forecast.  A major threat was communicated well. 

Then the next level of warming technology took over:  the U.S. Doppler radar network.  The U.S. has invested heavily in state-of-the-art Doppler radars across the country, radars that have recently been upgraded to dual-polarization (allows the radar to determine the type of precipitation or the nature of the "targets" it view).  

Around 2:30 PM the CDT the Doppler radar in Norman, OK observed the classic signs of a rotating, supercell tornado, including a hook echo and a mesocyclone (an area of rotation 5-10 km wide).  The image below (at 3:06 PM CDT) shows the hook echo, with a "debris ball" at the end.
With the radar image and reports from spotters, a tornado warning went out  at 2:40 PM, 36 minutes before the tornado hit Moore.  You may not think that is much, but 36 minutes is a far larger compared to the pre-radar days (average of 5 minutes in the 1980s).  (Thanks to Mike Smith's blog for an analysis of the lead time issue). This is enough time to run the tornado sirens, put out warnings in the media, and to give folks a chance to move to safe locations...if there ARE safe locations.

That is one big problem.  For an EF-4 or 5 storm the damage can be catastrophic, with buildings either being blown away or experiencing severe structural damage.  Safety can only be found in specially hardened rooms or enclosures.  And such protective spaces were not available to many resident of Moore and for several of the schools.  This needs to be changed.

In short, National Weather Service forecasters did a magnificent job for this event.  But could we do better?  I believe the answer is yes.

Because the atmosphere is chaotic (which means small errors in the initial state can have large negative impacts on the subsequent forecasts, impacts that increase in time) and the requirement of very detailed information to describe the initial environment for thunderstorm forecasts, it is virtually impossible to predict the details of severe thunderstorms a day or more ahead.   And this is not going to change soon. Yes, we can predict that  a major threat exists, but we can't get the exact locations or strength of the future storms correct.    So the day ahead forecast will have to be broad brushed.

But there IS the potential for major forecasting advances in the period from 1 to roughly 6 hours before the storm, if we can run models with enough resolution and can get enough information to describe the initial 3D atmosphere with lots of detail.  And we need to run many simulations (called ensemble forecasts) to get a handle on the uncertainties of the forecasts.

What kind of model resolution am I talking about?  Probably 1-2 km between the grid points, which requires huge computer resources.  We need to apply new ensemble-based data assimilation approaches (data assimilation is the technology of using data to describe the structure of the atmosphere).  And this modeling system needed to be frequently updated, at least once per hour.

We also need much more detailed information about the structure of the atmosphere, using innovative new data sources.  For example, I have a graduate student, Luke Madaus, who is using pressures from smartphones to improve weather forecasts and he is planning on testing this approach with strong thunderstorms.  A potentially huge advance for convective storm forecasting.  Unfortunately, the NWS support for this work (through the NWS CSTAR program) was cancelled for lack of funds....a tremendous frustration.  One of the weaknesses of the NWS is its inability to support and take advantage of university research.

The NOAA and the National Weather Service has been developing an early version of an advanced short-term, high resolution prediction system (the 3-km grid spacing High Resolution Rapid Refresh System, HRRR), which is only run in research mode because of the lack of sufficiently powerful computers in the NWS.  Below is an example of the HRRR forecast 6 and 3 hour out for 2100 UTC (4 PM CDT)----not bad, but not perfect).

Skillful 1-6 hr forecasts are potentially achievable since the forecasts are short enough that the growth in forecast error is modest.  And a few hour warming of a major storm allows sufficient time to evacuate folks from areas which severe weather is probable.

To achieve better short-term predictions, more model  development is needed, including higher resolution, state-of-the-art data assimilation, and moving to an ensemble approach.   But with enough research and sufficient computer resources, we can do better.  But the NWS did very well in this case.


  1. Cliff, thank you for your excellent weather analysis. I've had an idea for a theoretical tornado defense system, but don't know if the energy required would come even close. Perhaps you have some insight into whether I'm even remotely close...

    In a nutshell, the idea would be to bury rows of huge tanks of compressed liquid air in strips of vacant land or farmland along the edge of town where a tornado might start to rip through a town.

    The system would have funnel-cloud detectors of some sort, and right when a funnel cloud passes over a section of tanks, the tanks would explosively release the liquid air into the funnel-cloud, hopefully helping to quench the thermal updrafts and disrupt the funnel formation.

    Any thoughts on this theoretical system? I'm guessing that the scales of tanks required would be far too huge, but perhaps you might have some additional insight.

    Thanks again for your always-excellent weather reporting and analysis!


  2. Cliff: Moore had 36 minutes or more of warning (depending on the part of town).

    Please see:

  3. Cliff,

    The initial warning for this tornado was originally issued at 240. Then the warning and tornado emergency that you cite at 3:01. I made a similar oversight yesterday as well. Thus, the warnings on this event were first given with very long lead time. More info at and


  4. hi Cliff - If I looked at HRRR solution I would think squall line and not super-cells .. although its a close call .. still a long way to go with a skillful 3-6 hour nowcast.. great write-up ..thank you.. Bruce

  5. Since the Gov't falls behind onroviding the computer power, how about doing the idea that the radio-astronomers did, and harnessing the power of networked private computers?

    The Gov't could even encourage this by offering tax relief to those willing to pungle up the extra bucks to upgrade their computer purchases to more powerful machines, and their networking throughput to higher levels. This system would have the additional advantage of being very fast to become operational.

  6. Nice analysis Cliff. There certainly have been significant science and technology improvements in recent decades that pay dividends with events such as this.

    That said, if the goal is to reduce the loss of life and injury, non-science/technology factors (i.e., human nature) are surely a significant part of the problem, and I would argue, the bulk of the problem.

    I believe tornado fatalities can be reduced to essentially zero in 20 years, much as U.S. Commercial Aviation fatalities are essentially zero now -- something unthinkable 20 years ago.

    To reach that goal will take much more than better observational and modeling systems. Those interested might find this as a good discussion starting point:

  7. The annual per captita deaths from tornadoes has been decreasing although that would seem to be due to things not related to prediction of storms.

    Overall the lifetime odds of someone in the U.S, dying from a tornado (as of 2005) are 1 in 60 thousand while death by motor vehicle is 1 in 100.

    If preventing deaths is the job of government you would think they might be concentrating on highway safety.

  8. The weather around the Seattle area is a lot like the Seattle Mariners' play. Just when one thinks it is getting much better, the bottom falls out...

  9. Great analysis! It's unfortunate that this storm had such a terrible impact that we are breaking it down this much, however. I hope the city gets back to normal life soon.

  10. Thank you very much for this interesting blog entry.

    I'm still wondering when the 5000 J/kg of CAPE occurred. Or do you refer to SBCAPE?

    I only see roughly 4000 J/kg the day before:

    and about 3000 J/kg few minutes before touchdown:

    Thanks and best regards from Vienna, Austria

    Felix Welzenbach


Please make sure your comments are civil. Name calling and personal attacks are not appropriate.

Lots of Lightning, A Dismal July 3rd, and a Hopeful Forecast for Fireworks

 As predicted by high-resolution weather forecasting models, there were many thunderstorms and lots of lightning in eastern WA, eastern OR, ...