While I find cold core dynamics much more interesting, the time has come to talk about warm cores. Undoubtedly you will see doomsday deterministic forecasts showing “a Cat 5 hurricane hitting <insert city here>” on social media, if you have not already. Tropical systems are unique in the actions taken to prepare or mitigate impacts, and their predictability at various lead times. One of the best ways to get the most out of a forecast is to ask a question. In other words, is the forecast going to change your behavior, or is any potential fear just going to bubble up, without any actions to control fear? Okay, enough on philosophy, let’s take a quick peek at some stuff. Everything I will post below will be free and publicly accessible, as I strive to do every time. We will focus mainly on the global models with the acknowledgement that yes, there are tropical system-specific models (though not many at this lead time).
The problem with deterministic forecasts
In the gallery above, are three deterministic forecasts for the (as of the time of this writing) still developing tropical system in the Caribbean from TropicalTidbits. The problem with each of these individually is they convey absolute certainty in the timing, placement, and strength of this system, when we know we cannot have that level of confidence in any of those elements at this time. Viewing them all together is not a real big improvement, as you get the impression they are all equally likely. If our intention in forecasting is protect life and property, and to get the right people to take the right action, at the right time, we have to consider fear and fear management in how we approach forecasts. Dr. Ann Gordon at Chapman University has done quite a bit of research on this topic. In nearly all senses, deterministic forecasts aren’t useful for messaging at this lead time. So, let’s move on.
Useful ensemble data has traditionally been hard to come by. Sure you might find a spaghetti plot or two online, but in this age of big data, many post-processed visualizations are available, for both a qualitative (eyeball) and quantitative (numbers) view of uncertainty. When I think of ensemble forecasts, the first place I think of is ECMWF. And rightfully so, the European Center for Medium Range Forecasts is by far the agency with both the resources, will, and vision in regards to ensemble forecasts (they do it very well). I often thought of “cost” when I think of ECMWF, as they help their budget out by licensing their data. However, they have made great strides in the last few years of opening up by providing either degraded-resolution or latent data for free. And if you have not wandered over to their free charts page, you are missing out on good stuff. First, let’s look at their ensemble tropical storm strike probability plot. I place high value on the EPS because not only is it the largest global ensemble at 50 members, it arguable has the best data assimilation and perturbation schemes.
So, in the plot above, we can see the ensemble does closely follow the deterministic ECMWF model with high probabilities of a Florida impact. However, this is not universally true within the ensemble, and probabilities extend all the way up the gulf coast to Louisiana (something closer to the GFS deterministic solution). At this forecast range, even a 10% probability should not be ignored or dismissed. Even with the center of the highest probabilities in this plot (60-70%), if this ensemble has good dispersion and is well-calibrated [yes, I know this is a raw ensemble], we should expect it to be outside that area 30-40% of the time (that is, 3-4 out of 10 times). Especially when human lives are on the line, if we think about things in a cost-loss relationship, the denominator gets really big really quick, so while the average John/Jane Doe might not care about a 10% probability, officials such as governors and emergency managers should, and probably can start taking action at such thresholds.
What about the other models/ensembles? I am not very aware of the kind of post-processing for tropical systems that ECMWF does on their ensemble being done at either NOAA/EMC or CMC, but luckily, a Ph.D student at OU, Tomer Burg, has developed some useful visualizations.
In the plot above, we can see what would have taken several plots to see before. The first is that the GFS begins diverging early in the forecast, and does indeed end up pretty far from the mean of the super-ensemble, but still well within the ensemble solution space. This gives a more complete view of the uncertainty than a few deterministic plots. We can also see that the bulk of the super-ensemble as a more eastward track in the gulf than something like the GFS deterministic would indicate.
So, what’s the point? It’s not time to freak out yet. But it could be a good time to start managing fear. It’s always a good idea to start doing this at the beginning of the season, but, especially, if you live anywhere on the Gulf Coast from southern Florida to New Orleans, it would be worth at least thinking of where you would go, if an evacuation order comes. Messaging what you know, when you know it is often better and more trustworthy than just using uncertainty as a blanket to not talk about. Deterministic graphics will be posted, uncertainty be damned. In any case, hopefully a talking head on the internet, or a scary deterministic model plot doesn’t drive decision, but rather resources from both local officials and sites like ready.gov and nhc.noaa.gov. Such messages should be paired with any forecast to maximize the value and impact of the forecast.