Wednesday, September 12, 2018

Changes in Uncertainty for Hurricane Florence: Are We Communicating this Well Enough?

Hurricane Florence has been a study in contrasts.   For the last several days its track has been one of the most predictable on record.  Most major models have been spot on, the various ensembles have been tightly clustered, and the uncertainty in the forecast tracks is low.


Then as the storm approaches the Carolina coast everything changes:  the predictability of the storm is greatly reduced with all kinds of possible forecasts, including stagnation along the coast.

Why did this happen and how can my profession communicate such changes in predictability better?

To see the issue, here is an ensemble forecast from the European Center system showing the 51 forecasts started at 5 AM PDT on Monday.  Remember, that in an ensemble system a forecast model is run many times varying the initial conditions and model physics--showing the range of potential events.  This figure shows the probabilities of the storm being at any location, based on the ensembles.  Offshore there was little uncertainty in track but the the tracks really diverge near the coast.


A similar situation is true for the NWS global model ensemble (GEFS)


In contrast, the tradition way of showing uncertainty by the National Hurricane Center shows a progressive increase of uncertainty (see below).  This is because they simply use historical errors over time based on many forecasts.  The current figure communicates a loss of skill over time regarding the track forecasts, but there is a huge loss of information found in the ensembles...and such uncertainty information is very valuable.  Ensemble-based diagrams are far better.


Why were the forecast tracks reliable and tightly clustered out over the Atlantic, but all over the place near land?   It has to do with large-scale (or synoptic) steering flows.   To first order, you can consider hurricanes as huge tops that are pushed around by the large scale flow.  During the past few days, Florence has been steered westward by high pressure to the north of the storm, as illustrated by the sea level pressure map for Monday at 5 PM PDT.  The "L" due east of southern Florida is Florence.


The upper level (500 hPa) map for  11 AM on Monday shows high pressure north of the storm that was helping steer Florence to the northeast.


Similarly for 11 AM PDT on Tuesday.  A high pressure area to the north was effectively steering the hurricane towards the coast.


 But the situation on Thursday at 11 AM PDT is  very different.  A large ridge of high pressure is over the eastern US, without much flow over the SE U.S.  And weak high pressure surrounds the storm--thus there is little steering flow to push the storm in any direction. 


The dangers of a stagnating storm are substantial, particularly the potential for heavy precipitation, with some models going for 20-30 inches in some locations. 

Our models are now capable of providing useful information of how hurricane track uncertainty will change in time...the challenge is to find ways of better communicating the information.






13 comments:

TW B said...

This looks eerily similar to Harvey last year. Two years in a row seems very unlikely given the rarity of this pattern.

Wayne said...

The Washington post has a nice page/graphics
https://www.washingtonpost.com/graphics/2018/national/tracking-hurricane-florence/?utm_term=.7c867a355f79

Scott Souchock said...

I appreciate your continuing mission to figure out how to communicate weather information more clearly. I've been pondering this post and I think one of the problems is that storms are categorized. Is it only by wind speed? Or are there other factors. It seems that the categorization is to precise. And then the very nature of the variability of storms to change lends people to think that oh well, maybe the forecasts were wrong, and it won't be as bad, big, scary.

There also seems to be too much attention to where it makes landfall: that's nice for a track but that's somewhat irrelevant in the grander scheme. I was thinking that maybe rather than a "Cone of Uncertainty" maybe there needs to be a "Ring of Probability" or "Ring of Impact", concentric rings, perhaps color coded as to potential impact. I'm trying to think of new, visual ways to communicate possible impacts, areas affected, and so forth.

The more complex and nuanced story you tell in your blog posts — the explainers – should become part of the forecast story I believe.

Of course this is America/human beings in general and their bias to ok, it's going to be this, then do this, this, and that, and all will be well. There is a bias towards specificity and certainty. How forecasters and meteorologists navigate that tempest is, indeed, challenging.

Thanks for the blog and explanations.

Don Healy said...

It's early yet, but at this stage it appears that the new FV-3 model is closer to reality. Any thoughts?

John K. said...

Scott - I think you are on the right track. The "average Joe" just wants one or two quick facts to go on, such as "where does it make landfall". But of course as you point out, "where it makes landfall" isn't even the most important thing to know.

If one thinks about how most people go about trying to understand the natural earth verses how the earth actuality functions in terms of uncertainty, these two are in complete opposition to one another. I'm afraid Cliff and his compatriots have what amounts to an impossible job explaining things, to the public at large at least.

Eric Blair said...

Regardless of which forecast is correct, the authorities have done all they can do in order to make sure anyone within range of the winds, storm surge and widespread flooding zones are evacuated. Sadly, there will always be a few who insist on staying, and the tragedy is that lives will be risked needlessly in order to save them when things inevitably turn disastrous.

Unknown said...

If you look at the archive of tropical products from 5 days ago, it looks like the NHC will be within 50 miles of where landfall was predicted back on Saturday night, or at least brush the shoreline close enough to Wilmington to essentially count.

We need to seriously consider revising the Saffir Simpson Scale to a composite index of wind, rain and storm surge. While it might be a cat 2 by wind, it could be argued to be a cat 3 or 4 by storm surge and cat 5 for rain. I'm hearing reports of people considering of going back to Wilmington because it is "only a cat 2" when the forecasts of 30 or so inches of rain hasn't changed.

We revised the Fujita Scale, so why not improve this?

James said...

"high pressure north of the storm that was helping steer Florence to the northeast."

Do you mean northwest? Looking at the path of the storm, it doesn't appear it made any northeastern movements. Or do you just mean it was being pushed to the northeast relative to where it would have been without the high pressure system?

TW B said...

It seems like these cones of uncertainty give equal weight to all the models without taking into account their track record. Giving the better model tracks more weighting should cause less outliers that cause the cones to spread as much.
Another thing that would be helpful would be to plot the european model (or whichever model has been the most accurate} track seperately. on top of the cone of uncertainty.

Ellen Baker said...

I have family who live at the Space Coast - really familiar with hurricanes. They recommend this site with multiple models: http://spaghettimodels.com

Just sharing!

jimijr said...

High pressure on a constant-pressure chart? High heights more like it. But really, if you have never experienced a hurricane it is impossible to envision what is about to happen no matter what is said.

Rebecca Timson said...

Good proposal. I'm also interested in size and speed of the storm, but maybe your three factors get at that indirectly.

Anonymous said...

A similar technique is used in determining the probability distribution of nuclear power plant accident consequences.