Meteorologists have gotten the message: They messed up.
Before Saturday’s storm, forecasters warned that Puget Sound could be lashed by winds reaching tropical storm levels. After the storm – which wasn’t anywhere near that windy – the mea culpas were falling like a soft Seattle rain.
“Yes, our forecast did not turn out as predicted,” the National Weather Service’s Seattle office said in a Facebook post. “We are not pleased about it either.”
“Mother Nature proved once again we still have a lot to learn and that she still has some tricks up her sleeve,” KOMONews.com digital meteorologist Scott Sistek confessed in a posting that was headlined “What the Heck Went Wrong With the Storm Forecast?!?”
The National Weather Service’s Portland office said there were signs that the low-pressure system off the coast actually had two centers of circulation, which would make “the central pressure and winds associated with the system … lower than they would be if it were one center of circulation.”
But the main factor behind the fouled-up forecast had to do with the low-pressure system’s track: Most computer prediction models showed the low passing over the Olympic Peninsula, whipping up winds that would have wowed Seattle. Instead, the low stayed well offshore as it swept northward along Washington’s coast, finally making landfall over Vancouver Island and mainland British Columbia.
Cliff Mass, an atmospheric science professor at the University of Washington, was among those who warned last week about the potential for a “historic storm.” Now he acknowledges the shifting track was a problem. “As a result of the track uncertainty, there was considerable uncertainty in the wind forecasts in the Puget Sound interior, but we clearly failed to communicate that,” Mass said.
In Mass’ view, that failure to communicate – compounded by “media hype” – is as big a factor as any shortcomings of the computer models. Today he discussed his perspective on what went wrong and how to make it better. Here’s an edited transcript of the Q&A.
GeekWire: In this age where we have so much capability to predict the weather, such as computer modeling and machine learning, why was this forecast so wildly inaccurate?
Cliff Mass: “It depends on what you mean by ‘inaccurate.’ The storm prediction – getting the cyclone that developed, the right intensity – was actually quite good. A week before, we knew there was a threat of some kind of storm. It’s amazing that we could do that. Days before, the models predicted a storm was moving up the coast with a certain intensity. That was quite accurate. The main issue was that the track was about 100 kilometers farther to the west than forecast.”
Q: Any thoughts on why the track was so far west? Why didn’t it show up on the models?
A: “There’s uncertainty. That’s what we have to get at. We have ensembles that give us a range of possibiliies. There were close-in tracks, and there were further-out tracks. This took the further-out track.
“We were telling people the worst-case scenario: If it did come in close, these would be the impacts. That’s important for people to know what the worst case is. But our highest-resolution models ended up taking the storm closer in than what actually occurred. That was the issue.”
Q: The National Weather Service showed three potential tracks for the storm, but the actual track was even more westerly. Was the actual track even shown in the computer models?
A: “Yeah. There’s not just three tracks. We have ensembles that give us dozens of tracks, and we can pump that up. Just imagine: The storm is moving 1,000 kilometers, and we are trying to get the end position within 10 or 20 kilometers. It’s hard to do that. We ended up with an error of about 90 kilometers after the thing has traveled 1,000. That’s really not bad, right?”
Q: When did you realize that this weather system was not going to be anywhere near catastrophic?
A: “By the time we got to Thursday, we knew that we weren’t dealing with a Columbus Day storm any more. The models were all coming together … suggesting a much weaker, much smaller storm. A lot of the media didn’t understand that, or we didn’t communicate that well enough. Then the whole issue became the track.
“The weather service talked about that, and I know I did. ‘This is what the high-resolution model is showing, but if the track’s off, this forecast will be off.’ People didn’t understand that well. We failed to communicate that well. On Saturday morning, it was pretty obvious that there was a lot of uncertainty, and by the afternoon, I said, ‘OK, it’s clear, it’s going farther offshore.’ That’s the best we can do, to update stuff, right? But we need an approach to get people to understand this uncertainty issue.”
Q: How can you improve communication on that?
A: “We’re going to have a meeting on Thursday with the weather service to talk this thing through, and have a psychologist with us to work on this.
“First, I think it is important for people to know the worst case. Worst case tells people whether they should take it seriously, whether they should pay attention, and how particularly sensitive people should prepare. But I think we also have to give the most probable case, based on our ensembles. We have to give people information about the uncertainty in the forecast, what the range of possibilities could be.
“Once we get far enough along with the calibrated ensembles, we need to give probabilities. Say, ‘OK, at this location, the probability of getting a zero-to-10 mph wind is this, 10-to-20, this’ … give the probabilities of various wind ranges. We’re not there yet, but in the end, that’s what we need to do. And then let people decide how they want to protect.”
Q: What do you see as the frontier for computer modeling, and for the science of storm prediction?
A: “There are two critical things that have to be done: First, we need a high-resolution ensemble system for the United States, with a large number of members – which we don’t have. The weather service has been dragging its heels on this. We need 3- to 4-kilometer resolution for the whole United States. That’s how you explore the uncertainty.
“Then we need very sophisticated statistical post-processing at the end of it, to make sure the probabilities are calibrated. We need to invest more in the science and technology of high-resolution ensemble forecasting.”
Q: Does it seem ironic that so many people are upset about not facing massive inconvenience, or even missing out on what could have been a life-threatening storm?
A: “It’s interesting, the reactions of some people. You’ve got to realize that this weather and meteorology business is on the borders of a religion. People like big storms. The hit rate this time was extraordinary. I got a million hits a day. The weather service was getting a million hits a day. Social media has blossomed. Facebook was crazy. All of a sudden now, we have this extraordinarily powerful tool for communication, and we have to learn to use this. That’s another big lesson from this thing.”
Q: So it’s as if people missed out on a Stonehenge-like experience?
A: “It is. It’s a very primal need, to feel in awe of the natural environment. People get into it. I was at QFC, and the checker was laughing and saying, ‘It was amazing what happened here. People just came in and piled on the food, and they were just so happy doing it.’
“There’s a pleasure over stocking up for the storm, and being ready to experience it, and being prepared. It’s something that people really enjoy, in a strange kind of way.”