Much like Danny before it, Tropical Storm Erika was no match for dry air, wind shear, and the higher terrain of islands in the Caribbean.
Fears for a landfalling Erika were unfounded for the United States, however, and Florida will continue to tack on days to its ten-year hurricane-free streak, for now. Unfortunately, there was still more than enough storm to cause major problems for some, including the island nation of Dominica (that’s pronounced dom-uh-NEE-kah, by the way). Erika killed 20 there, and Dominica’s prime minister says the storm set the county back some 20 years.
While I don’t think our collective communication about Erika set us back two decades, I did notice a few things that should give us weather communicators pause. Every storm — and every communication effort — should be a learning experience, and to that end, here are four items that caught my eye and some suggestions for making our next effort better.
Single Model Solutions
First, and perhaps of greatest concern to me, is the use of single-run, deterministic model solutions, especially by professionals. Perhaps the most egregious example of many was discussion about Thursday’s 12Z Canadian solution, which forecast a strong Hurricane Erika making a very Sandy-esque turn into the New York City metro area more than a week out:
The alarming irony is this particular model has consistently been among the poorest performers in track forecasts over the last few years. So of all the single-run-of-a-single-model solutions to draw people’s attention to, this might have been the worst one of the bunch! Sadly, it was far from the only example. I saw numerous model projections of a “Hurricane Erika” off or making landfall along the coasts of Florida, South Carolina, and North Carolina, thanks to various single runs of the GFS, European, and other models.
The meteorological hazard of highlighting a single run of a single forecast model should be obvious. Models are guidance, not forecasts. They are subject to a multitude of errors, including but not limited to systemic biases, poor initialization including incomplete or incorrect identification of a storm’s center, and inadequate representation of physical processes. As a result, a given model may be subject to wild swings from one run to the next, owing to the interplay of these and other factors. The old saying goes, “all models are wrong, but some models are useful”, and there certainly were a lot of “wrong models” with Erika:
In fact, when it comes to those above hazards, we saw all of them at various times with various models with Erika. Every single one! In fact, at one point, the center became so difficult to identify, many models couldn’t initialize a tropical cyclone at all, and of those that did, many located it incorrectly. The result is the first lesson in Computer Science 101: Garbage in, garbage out:
This is why nearly all of the weather enterprise is moving toward an ensemble approach toward forecasting — by understanding the range of solutions, including how those solutions cluster around different outcomes, and the models that produce those solutions, we can begin to wrangle with the uncertainty that is inherent with any single model run. The end result is better, more stable, and more reliable forecasts. The range of solutions from model to model and run to run practically screamed that we would be fools to put our faith in any one single solution.
And yet, the temptation of showing a single model run showing a major hurricane making landfall is great. It’s certainly easier to show a clear “what-if” scenario than it is to talk about chances and probabilities. But easier doesn’t mean better. Featuring a single model run, especially one that features a high-impact outcome, can drown out communication of other, far more likely scenarios, including the official National Hurricane Center forecast or even the consensus of better-performing models. At the very least, it can present a false sense of likelihood of these one-off model runs, not dissimilar to the notion of “false balance”. That’s where journalists, in a desire to present “two sides to every story”, give undue weight or airtime to a side of the story that either doesn’t exist or isn’t deserving of that weight.
This is especially important on social media, where the context of lengthier discussions is stripped away entirely or withheld from the user until they click on an article. Images and article previews should, to the extent possible, use self-explanatory graphics that do not depend on a user’s reading an article for proper context. Moreover, before sharing any single model forecast track, meteorologists and weather communicators would do well to remember the official National Hurricane Center forecasts (the solid black line marked “OFCL”, below) historically perform as well as or better than a blend of top-performing models, and those model blends tend to outperform any given model over time:
Spaghetti With a Side of Insight, Please
To address the hazards of showing a single model run, spaghetti plots of model forecast tracks have become rather commonplace. They first appeared online but made their way onto television quickly as the broadcast graphics vendors implemented software to automate their display. Today, even disturbances without names, numbers, or closed circulations get an entrée of spaghetti on TV and social media.
Among other things, spaghetti plots can show the spread of model solutions as well as how different solutions may be clustering toward particular outcomes. This can be very useful information, especially when coupled with the secret sauce: An understanding of how the various models handle tropical systems and their environments, the factors that drive a storm’s intensification and movement, and perhaps most importantly, the availability of reconnaissance and other key data sources for the models to process. All too often, however, the spaghetti is served by itself, an unfulfilling dish indeed.
Bryan Wood, a meteorologist in the insurance industry, showed the value of serving up the spaghetti with the sauce. In a single tweet, we see a cycle’s worth of model tracks annotated with three elements that both add value to the graphic and differentiates Wood from the noise of social media as someone with beneficial insight:
As I mentioned in my presentation at the NWA Annual Meeting in Salt Lake City, the Internet and social media are helping to advance an incredible democratization of data. What isn’t being so rapidly spread far and wide is the knowledge to process and interpret those data into useful information, or better yet, knowledge and understanding. A sure-fire way to rise above the crowd of people simply re-sharing weather data is to add value to the data you share and make it useful to people in some way. Go beyond a surface analysis and provide context or insight based on your experience and education.
Of course, if you’re going to do this, you need to ensure that what you’re adding is correct! Too often, spaghetti plots are built without care for which models are included. Sometimes, all the “model” tracks are turned on. I use quotation marks because one giveaway is the inclusion of “XTRP”, which is simply a forecast of motion over the last six hours extrapolated in time. This is rarely, if ever, useful. Other non-model “models” are frequently included in these plots, and while that isn’t in and of itself a bad thing, their inclusion warrants special attention and analysis.
If you are building spaghetti plots for dissemination or simply sharing those created by others, consider reviewing a number of resources. For starters, NWS meteorologist Alex Lamers provided a few basic points on some of the most-often used spaghetti tracks. Once you’ve whetted your appetite, I would strongly recommend reviewing three NHC resources to ensure you have a solid foundation in what the models are, which have been the best performers, and under what circumstances each can over- or under-achieve:
- A summary of what models are available and used for what purposes
- Overall model guidance verification statistics
- Model performance sections of NHC’s excellent annual verification reports.
The “IT” Factor
I have to admit being taken aback by the number of individuals and outlets referring to Erika as a “she” rather than an “it”. And in one case, I heard the anchors for one outlet insisting on using “she” while the meteorologists consistently used “it”. Regardless, I honestly thought we had gotten over the hump with this one.
Apparently, I wasn’t the only one who noticed the uptick in “she” references, as the debate came up online in a couple of places last week. A number of broadcasting colleagues expressed support for using gendered pronouns on the grounds that they sound more conversational than referring to a storm with gendered name with the gender-neutral “it”. Others saw an ulterior motive: On Twitter, Weather Network anchor Kim MacDonald made the point that storms are “it” and was promptly castigated for “politically correct nonsense”.
As a reminder, even though the names given to tropical systems are gendered (they alternate female, male, female, male… this year; the order swaps next year), the National Hurricane Center refers to them using non-gendered pronouns. In other words, a hurricane is an “it”, never a “he” or a “she”. There was a time when this was not so. For more than two decades, hurricanes were given female names, exclusively. And as you might imagine, when writers wrote about those supposedly all-female storms, the results reinforced unfair stereotypes at best and were outright sexist at worst. So, a change was made to include an equal number of male and female names in the list and to alternate them from storm to storm.
In a recent WeatherBrains appearance, NHC Branch Chief James Franklin answered the “he/she/it” question quite succinctly:
Further, the Associated Press Stylebook, the go-to text for answering questions like this in news coverage, is equally clear:
hurricane – Capitalize hurricane when it is part of the name that weather forecasters assign to a storm: Hurricane Hazel. But use it and its not she, her or hers or he, him or his in pronoun references. And do not use the presence of a woman’s name as an excuse to attribute sexist images of women’s behavior to a storm. Avoid, for example, such sentences as: The fickle Hazel teased the Louisiana coast.
Yes, “he”/“she” sounds more conversational. No doubt. But here’s the news flash: So does much of how we talk when we’re not in front of the camera or what we write in emails instead of in news stories for publication! While whether something sounds conversational is a key factor in style decisions like this, it is far from the final arbiter. Correctness (NHC’s viewpoint) and appropriateness (avoiding sexism and sexist stereotypes, even if we think we’re not being sexist) also play key roles. That’s why this warrants an entry in AP’s style manual in the first place.
But the AP Stylebook is just that — a style guide. Many newsrooms and organizations use it as a starting point but modify it for their use. If you or your organization have and opt to use the gendered pronouns, consider this: One copy editor did some digging and what he found suggests that people are more likely to use gendered pronouns for storms with female rather than male names.
There’s a another, more insidious dimension to this, says Dr. Susan Jasko from California University of Pennsylvania:
…when we anthropomorphize natural non-human, inanimate phenomena, we tend to unconsciously, and therefore uncritically, make a host of assumptions and draw implications about current and future “behavior”. Because storms do not have human attributes and capacities, such as motivation, planning, anger, or mercy, guessing or believing you can anticipate a storm’s likely path or damage based on such unconscious, biased, and all-too-human assumptions can easily lead you astray. This can be disastrous.
This notion of “it” vs. “he”/“she” isn’t an idle musing from an ivory tower, nor is it politically motivated. As Dr. Jasko says, these kinds of things are subtle and usually, they’re not intentional at all. That’s why they can slip in without our noticing in places like this. But they can open the door to more damaging or demeaning attitudes. While it’s a bit less conversational, sticking with “it” will help us all avoid falling into a trap of sexism and gender stereotypes and will help you and your organization be more consistent overall. (And if you absolutely must break from the NHC convention and AP style, at least ensure you are consistent within your organization and from storm to storm!)
After The Storm
Finally, I want to talk about our final words about Erika.
Not long after Erika lost its closed circulation and was declassified by the National Hurricane Center, more than a few folks I follow on social media shared comments like “RIP #Erika” and “Fare well, Erika”.
No, I don’t think folks on Twitter were pronouncing a final blessing on a storm that killed upwards of 20 of our brothers and sisters in the Caribbean or wishing a violent storm that provided nothing but turmoil for hundreds of thousands to “fare well”. However, do our non-meteorologists friends know that? Before you answer, remember these are the same non-meteorologist friends who hear us talking about how “impressive” a storm is or who watch us or our colleagues actively chase bad weather for sport. Can we be so sure? (Plus, when we casually use phrases like “RIP” or “fare well” in referring to storms, we risk cheapening those words and sentiments for when we really need them.)
Look, I get it. We meteorologists and weather fans appreciate a good storm. We’re awed by the power and beauty of the atmosphere. And when we slap a name on a storm, even a forecasting-pain-in the-rear like Erika, it can take on a life of its own. But these storms are frequently killers! Now that Erika is gone, my sentiment is nothing short of a hearty do-not-pass-go, do-not-collect-$200, “good riddance”.
Let’s save our kind words for those afflicted by Erika and its ilk. In the words of Alabama meteorologist Alan Sealls, “root for people, not storms.”