As I will describe below, there were two problems with last week's forecasts:
- although the storm intensity was well forecast, the track was off by about 60 miles.
- meteorologists such as myself did not effectively communicate the changing threat of this storm and the uncertainty in the forecasts.
Saturday waves on Puget Sound
Picture courtesy: Seattle Times
Picture courtesy: Seattle Times
The Initial Warnings
The first hint of a major storm event was apparent about a week before Saturday, when a number of the global models began predicting that Typhoon Songda would travel across the Pacific, transition to an extratropical storm, and then strengthen rapidly as it approached the Pacific Northwest.
What was particularly threatening was that model run after model run (for forecasts initialized Tuesday and Wednesday) showed a huge, deep storm developing just offshore....a storm as deep as the famed Columbus Day Storm. Here is a sample. 955 hPa. Amazing. And of great concern.
84h forecast valid at 5 PM Saturday
We knew there was considerable uncertainty in these forecasts, since the storm had to undergo transition from a typhoon to a midlatitude system (switching energy source in the process) and would travel thousands of miles before reaching our shores. These projections were not only consistent in time, but since they were relatively short projections (120-84 hr) there should have been some skill. In other words, we could not ignore the threat inherent in this forecast and had to tell the public about the possibility.
Not surprisingly, the media went bonkers after they heard about it, with all kinds of stories about the Columbus Day Storm (see Seattle Times headline below)
Accompanying these stories were lots of scary pictures.
Let me stress, at this point (Tuesday or Wednesday) there was substantial uncertainty inherent in this this forecast, as confirmed by our ensemble systems (models run many times) and I don't think we communicated this strongly enough.
But everything changed on Thursday after the storm had gone through transition into a midlatitude system and was firmly embedded in the midlatitude flow: virtually all the modeling system indicated a very different evolution. There would be two storms, another midlatitude system late Thursday and Friday morning and the Saturday storm would be MUCH weaker (965-970 hPa) and very compact. Such a low pressure still indicated a powerful storm, in the class of the kinds of storms we get every 10 years (e.g., 2006 Chanukah Eve, 1993 Inauguration Day, etc.). Let me illustrate.
Here is the UW model sea level pressure forecast (driven by the National Weather Service GFS model) for for first storm (valid 11 AM Friday). 982 hPa low crossing the tip of the Olympic Peninsula. Typical looking winter storm around here. Yawn.
The forecast started a day later (5 AM Friday) and valid at the exact same time (5 PM). A bit weaker, faster up the coast, and very compact. Same track (across the NW tip of the Olympics).
Want to see the vast difference in horizontal scale between the Friday and Saturday storms? Here is a direct comparison. As shown by the scale of the clouds spiraling into the storm, the low center was enormously bigger on Friday. Even Trump would be impressed.
So on Friday, we knew we had a weaker and smaller storm, but the central pressure was still quite low (anything under roughly 985 hPa can produce strong winds around here). But there was something else, there was reasonable uncertainty regarding the track...and that was critical for such a small storm. There was variations of 50-100 miles among the major modeling systems and among the ensembles of specific models, like the GFS.
One way to show this uncertainty is from a graphic that presents the variation of the pressure forecasts for the UW high-resolution ensemble system, in which the UW WRF model is run many times with different initial conditions from varying Global modeling systems. Here is the variability of the sea level pressure forecasts for 8 PM Saturday, when the storm was making landfall. The yellow/orange colors show substantial differences among the forecasts, signifying uncertainty in the location/intensity of the low center.
The public needed to understand that the threat level of the Saturday storm was much less than earlier in the week. They needed to understand that this was a very compact system, and that if the actual track was only 50 or so miles west of the predicted track of the high-resolution prediction, Seattle and the interior would experience much lesser winds.
I tried to communicate this track uncertainty in my blog and the National Weather Service did as well. In fact, forecasters at the Seattle NWS forecast office did a masterful job on Thursday morning, a portion of which I will repeat:
The exact track of the low will make a huge difference in how badly this storm impacts western Washington. There is a 1 in 3 chance of the low center directly crossing some part of the central or north coast of western Washington. This would be a worst case scenario leading to a historical windstorm for nearly all of western Washington that would be long remembered. There is a 2 in 3 chance that the low center will pass further offshore... making landfall on Vancouver Island. This outcome would confine the most damaging winds to the coast and to the north interior (areas north of everett). Inland locations such as the Puget Sound region and the I-5 corridor of southwest Washington would experience the type of windstorm that would normally be expected a couple times each storm season. Power outages and tree damage over inland locations would be less widespread in the scenario.
We not only need to provide society of the worst case scenario, but also the most probable and the uncertainties. Unfortunately, we (meteorologists) were not as effective as possible in communicating our current knowledge of the storm and the uncertainties. How do we know that? Because some of the media was still hawking a Columbus Day Storm type event, and exaggeration and hype were spreading through social media (see a tweet below that was making the rounds). Folks were stripping stores bare and organizations were canceling activities even during times when no threat existed (e.g., Friday night/Sat morning).
In the end the actual storm track was about 60 miles west of the predicted location of most (but not all) of the forecat models (see graphic). Such an error is not large for a 24-h lead time.
Unlike the story presented by some, the storm was not weaker than forecast (in terms of central pressure), but a bit smaller in size. In fact, when the system passed Tatoosh Island(TTIw1) on the NW tip of the Olympic Peninsula, the pressure fell to 968.5 hPa (see graphic). Excellent intensity forecast.
There was some false information being spread about the storm, even by local meteorologists. For example, it was claimed that the storm was weakened by splitting into two storms. This interpretation was based on faulty understanding of satellite scatterometer data.
Better Technology Would Have Helped
There are two main technological enhancements that would have really helped.
First, we acutely need a radar on the Oregon Coast. The Langley Hill radar near Hoquiam picked up the storm when it was nearing landfall (see graphic at 5:10 PM Saturday), but that only gave us a few hours warning.
If we had a radar on the Oregon coast, it could have told us on Saturday morning that the storm was tracking farther to the west and greatly improved our short-term forecast. I am not sure if Senator Maria Cantwell can help us with the Oregon radar (as she did with the Washington one)--we need our friends in Oregon to step up to the plate on this one.
Second, our ability to forecast such events is undermined by the lack of a high-resolution (e.g., 4-km grid spacing) ensemble forecast system over the U.S. Such a system would help provide high quality probabilistic predictions of such events by making large numbers (e.g., 20-50) of forecasts. Report after report, committee after committee, user group after user group have pleaded with the National Weather Service to do this and they have dragged their heals. Even the National Center for Atmospheric Research (NCAR) is experimenting with a small high-resolution ensemble system (see below). Congress needs to push NWS management to finally build this acutely needed technology and to provide funding for the additional computer resources it requires.
Finally, my profession needs to work hard to find better approaches for communicating uncertainty in our forecasts. Methods that will be effective in this age of social media. On Thursday, some of us will meet at the NWS Seattle Forecast Office to talk about this important topic. We must find a way to tell folks about the worst case, the probable scenario, the uncertainties of the forecast, and eventually provide calibrated probabilities of parameters such as wind speed.
Please support I-732, the revenue-neutral carbon tax swap, which will help reduce Washington State's greenhouse gas emissions, make our tax system less regressive, and potentially serve as a potent bipartisan model for the rest of the nation. More information here. Some opponents of I-732 are spreading false information, suggesting that I-732 is not revenue neutral. This claim can be easily disproven as discussed here. I strongly support I-732 as do many UW climate scientists. We have an unprecedented opportunity to lead the nation in reducing carbon emissions and to establish a model that could spread around the country.