28 January 2015

## Predicting the Weather is Tricky Work

Posted by Dan Satterfield

During Tuesday’s 4 p.m. and 6 p.m. news, I talked about ensembles, how they outperformed the deterministic models for New York City last night and that they appear to be performing better for the local area with regard to a storm arriving on Sunday.

First, what are ensembles and why are they so important? Let’s start with what we call a deterministic model. It is a single computer model, with mathematical equations that describe all the laws of physics and thermodynamics, along with some algorithms (estimates) of processes that cannot be modeled explicitly. We call them deterministic models because they, in effect, are saying to us “this is the way the atmosphere will change over the next week, period! End of discussion!”

Well, no computer model can do that, and one of the biggest reasons why is our inability to observe the atmosphere perfectly. The initial conditions we feed into these models are critical, and even very small errors in the analysis can cause big problems just a few days out.

Now that computing power is so much more robust, we can take those initial conditions and tweak them ever so slightly, to 20, 30 or even 50 different extents. Then you run all those models to see how much variance those small initial changes cause. If the solutions are tightly packed, you have high confidence in the forecast. But if the solutions look like a finger painting exercise, then the message is this: The atmosphere is in a highly volatile state, so do not hang your hat on any one solution. Here are the two or three most likely solutions, and you need to prepare for all three until the situation resolves itself.

Do you see this as being wishy-washy? I can see how you would. After all, you never heard about levels of uncertainty 10, 20, 30, or even 40 years ago.

Why now? The truth is, that level of uncertainty was even higher back then than it is now. It’s just that nobody wanted to talk about it. A meteorologist’s job is to take the best science available to him or her, combining real observations with computer model guidance to produce the best forecast possible. And yes, there are times when we have low confidence. It is our duty to be honest with you and tell you that, just like it’s your duty to realize that weather forecasting is not an exact science.

Nobody likes uncertainty, but it is a fact of life, and we all need to deal with it.

During the past week, the European deterministic model produced one forecast Sunday night indicating nearly a foot of snow for Raleigh this coming weekend. That forecast was circulated all over the internet, even though every forecast before and since has shown next to no snow, and the ensembles have never indicated a big snow. The above graph shows snowfall forecasts from the European Ensemble Mean (top) and the European deterministic model (bottom). The times along the bottom represent the time the forecast was made for the upcoming weekend. As you can see, the ensemble mean is consistently low while the deterministic has a spike of 9 inches.

There are a lot of folks out there that want to be heroes, and they take that one forecast, circulate it as their own forecast, hoping that it turns out to be right, promoting them to an iconic status. It rarely works, but when the blind squirrel finally finds the nut, all is forgiven.

My folks here in the Weather Center would rather be right than first. We will never be the first ones to call for a huge snowstorm or a massive tornado outbreak, unless of course no one else talks about it.

Yes Virginia, there is uncertainty, and it’s not going away anytime soon. In the meantime, honest communication is our only hope.