How Did Those "Maxed-Out" Probability Forecasts Work Out?

For sixteen years, the National Weather Service has been producing severe weather forecasts in a probabilistic form. Only twice in all that time have they maxed out (forecast the highest value) their numbers the day before the event (known as "Day 2"). So, with the help of the NWS's Greg Carbin, I wanted to see how well those forecasts worked out.

The Friday Day 2 severe weather forecast was for all types of severe weather (tornado, large hail, damaging thunderstorm winds). The highest the forecast (reproduced below) probability can go is 60%. Here is the forecast issued early Friday morning for Saturday:

Through a technique called "hindcasting," Greg re-ran the forecast using the actual reports of severe weather. In other words, to produce a "perfect hindcast" for comparison. Here it is:

With the exception of west central Texas, the forecast is excellent. The 60% maxed out probabilities were warranted.


Now, let's take a look at the tornado (only) probabilities issued 11:30am Saturday morning before the first tornado occurred. The 45% is very high and rarely attained, but they can go as high as 60%.

Here is the perfect hindcast:

Turns out they could have maxed the probabilities out at 60% in Kansas! The forecast of tornadoes also went too far south. Still, this was a difficult-to-forecast extraordinary event and these extraordinarily high numbers attracted the attention of the media, bloggers, and even a newspaper. They, in turn, got the word to the general public of the exceptionally high risk in the central Plains.

I would give the Day 2 forecast an A and the Day 1 tornado forecast an A-. 

Nice work National Weather Service Storm Prediction Center!