30 January 2015
We Must Do A Better Job of Communicating Forecast Uncertainty
Posted by Dan Satterfield
My wife says that most of her friends have no idea about how I make a forecast, and I suspect that some believe I just get it from the NWS, without thinking how even they might do it. Many folks do understand we use numerical models but beyond that it’s hazy, and they think that if the forecast is wrong, it’s because the model was bad, not the interpretation of it. I prefer to call numerical weather models guidance because that’s what (and ALL) they are.
The info-graphic above does a great job trying to explain what forecasting really is, and in many ways synoptic forecasting is similar to the process your GP goes through when you make an office visit. It’s both an art and a science where data from observations of the atmosphere or your blood play an important role. That said, the blizzard forecast last Monday is a warning call that we meteorologists need to do a much better job of communicating forecast uncertainty.
Make no mistake, the forecasts were badly wrong in many areas, but nothing drives forecasters crazier than getting blamed for a bad forecast some armchair weather guesser posted on Facebook. Often, we get blamed for missing a forecast that was actually not bad, and while some of that is due to the fact that the complainer didn’t really pay attention to what you said, there is ALWAYS the opportunity to communicate it better. In science, any measurement or prediction is worthless without an accompanying statement of the uncertainty involved, and if you’ve ever measured a room you want to buy carpet for, you inherently understand this!

A graphic like this, combined with a forecast of possible snow totals is much better information to viewers than either one alone.
While we often use percentages in weather forecasts, there seems to be very little of this in forecasting snow, wind, heavy rain etc. The graphic below is one I used Monday night on air for my viewers in Delaware and Eastern Maryland. While my forecast was a bust in the Dover area, it showed only a 50% chance of an inch of snow over areas to the south. Combining this with a map showing accumulations is IMHO a better way of sharing my thinking about the forecast with my audience.
Now, this does not work with everyone, and if you are a young second grader named Elisha Jones (and hoping for a day off from school), then there is nothing that will alleviate your disappointment! I received a packet of letters from some cute young kids this week, and this is Elisha’s view.

Elisha, I totally understand! There is no such thing as too much snow, and I’ve been there my friend!
Sometimes a forecast is going to be wrong, but we actually do a much better job of forecasting significant weather events than many people think. Failing to communicate our forecasts properly is not the entire reason, but it’s a substantial one. Yes, there will always be some who will blame you for someone else’s forecast (or even for a forecast that was basically correct) and better communicating uncertainty will not make this disappear, but for the majority it will be a great help.
Something I often tell viewers this time of year:
In the summer, if my rainfall forecast is off by 0.2 inches, no one will notice, but in winter that error means I miss two inches of snow!
Interpretation of the models, especially ensemble model runs is crucial in the forecasting process. I like one approach to communicating forecast uncertainty that one outlet has impressed me with. The Capital Weather Gang gives their forecast, and easy to understand graphics showing “boom” or “bust” which show the probabilities of less likely, but still possible scenarios that could take place.
Concerning a bad model run being to blame, to some degree, it IS a factor. In chess, where in the last decade and change, the best programs run on large computers have become superior to the best chess players in the world. Likewise, the most sophisticated weather models run on world class computers have been far superior to human forecasters (having access to raw data, but no computer model guidance) for decades. The difference here is that it is questionable that a chess player, presented with the computer’s solutions could improve upon the solution. At this point in time, a trained meteorologist, presented with computer guidance, IS capable of improving on the guidance the majority of the time. That’s where a lot of the satisfaction in our job comes from to compliment our curiosity about how the atmosphere works, and watching it all unfold in real time!
Agreed! Communication will be big part of not only forecasting but of getting people to understand and do something about this information. It’s a sensitive business when it comes to communicating risks and balancing it with doomsday predictions.