Why does your long range forecast change?
Tuesday, January 15, 2019, 1:37 PM - We've all had the experience: It's Monday morning and you look at your forecast for next weekend -- 5 cm of snow. Check again on Tuesday morning and suddenly it says 2 cm, or 25 cm. It's a frustrating fact of forecasting, but why does it happen? Meteorologists at the National Weather Service came up with a pretty good explanation for the intricacies of long-range forecasting.
It all comes down to Plinko.
You might remember the game from The Price Is Right, but just in case it's been a while, Plinko is a pachinko-like game, where the contestant releases a disc at the top of the game board, and the disc bounces off pins as it drops toward different targets at the bottom. Which target the disc hits depends both on where the contestant releases it, which pins it hits on the way down, and how it bounces. The difference between a loss and a big win depends on a multitude of interactions, each difficult to predict in their own rights and even harder to predict in concert.
As odd as it may sound, an extended forecast goes through a similar process.
Image courtesy U.S. National Weather Service Kansas City Missouri
The image above, from the NWS office in Kansas City, gives a good depiction of the process. When meteorologists input data into a weather model, we're controlling the starting position of the ball. But after that, it's out of our hands (literally). The closer we get to an event, the 'lower on the board' we release the ball, and the more control we have over where it falls; that's why -- normally -- the forecast tends to converge toward one solution over time.
One data source meteorologists are using more and more these days is ensemble modeling -- essentially the same model run over and over with slight changes to the initial conditions (releasing the ball at different spots on the board, to use the Plinko analogy). You'll frequently hear our meteorologists here at The Weather Network talk about these ensemble models when we're anticipating a major storm, be it a winter storm or a hurricane.
FROM THE MET DESK: HOW THE WEATHER NETWORK HEADQUARTERS MEASURES SNOW
SO WHY BOTHER WITH LONG RANGE FORECASTS AT ALL?
Well, for one thing, not all long range forecasts are created equal. As you'd expect, the more complex the weather pattern -- say, when we're looking at a large storm -- the greater the number of uncertain items that go into the forecast (more pegs on the Plinko board, more possible outcomes). Quieter pattern? Fewer pegs, fewer targets. Obviously this isn't ideal when it comes to big, impactful storms; ones where everyone's eyes are on the forecast, wondering, for instance, if their area will see significant snowfall.
This is where the human forecaster comes in. As the meteorologists at the NWS say, "Just as no one can tell exactly where a ball is going to land in a pachinko game, it’s not a meteorologist’s job to tell you exactly how much snow you’ll receive 5 or 7 days in advance. However, a good meteorologist can look deeper into forecast models to get a sense of their reliability as well as what factors could cause that forecast to change." That's why we tend to focus on trends rather than specific beyond the first few days of the forecast, narrowing down the possibilities by what looks realistic as more and more data becomes available.
In the original example the NWS office was responding to, it was a case of long range model guidance suggesting a monster snowstorm with significant accumulation for an area that doesn't typically see very much snow. Those are the kinds of instances where a meteorologist will take the result with a grain of salt; it becomes a situation worth watching closely, even if you know the actual amount produced by the long range model is likely overdone. It then becomes a tricky balance between how much information to communicate to the public, to give adequate warning about the possibilities without 'crying wolf' -- kind of like getting it close, without going over.