When a Forecast Is Less than No Forecast at All

Can we all agree that mixed messages are bad? And they are especially bad when we want people to act on the messages and make good decisions in the face of a potential extreme weather event? As long as I can remember, it has been conventional emergency-communications wisdom that confusion impedes comprehension and delays action.

Last week I opined that the word “post” in Post-Tropical Cyclone Hermine was detrimental to clarity (not to mention “cyclone”). “Post”, in my mind, conveys the sense the storm is over, as in done with. Meanwhile Hermine was forecast to strengthen and be an increasing threat — a kind of mixed message.

But another more insidious kind of message mixing has been part of modern weather-forecast presentation for the last decade. And it is so systemic that almost everybody does it — including the National Weather Service, The Weather Channel, every weather app and website, and most TV stations.

In my opinion there are pluses and minuses to the modern forecast-presentation format for average-day weather, but it is bad in principle and in practice during extreme weather events.

There was a time, beginning before Hurricane Andrew, when we would not show a forecast for any days in a 5-day forecast (a long-range forecast at that time) if a tropical storm or hurricane might threaten South Florida in that time frame. The thinking was, the weather is either going to fine or life-threatening or something in between, so pinning it down with an icon and a temperature was misleading and possibly dangerous. A version of that philosophy lives on with a few on-air meteologists today.

In Hermine and every hurricane that has threatened recently, however, not only do we get specific daily forecasts for nearby coastal cities out to 7, 10, or 15 days, we get them for every hour. The winds in New York City will be exactly 28 mph at 2:00 p.m., 31 mph at 3:00 p.m., and 25 mph at 4:00 p.m. tomorrow. Look out, it is going to be especially windy at 3:00 p.m.! Craziness. And this with a Tropical Storm Warning in effect.

The full and correct message, of course, is that the winds will be light if the storm leans right, away from the coast and does not strengthen, but could be well above 40 mph if things go the other way. And both possibilities are reasonable forecasts. Modern science does not allow a more detailed prediction, but we make one anyway. Every day, every city, every storm.

This practice comes from database-driven forecasts. A computer-generated number — sometimes human tweaked, sometimes not — is dutifully entered into every slot in the database for each aspect of the weather forecast multiple times a day. There’s a number for the temperature, the wind, the humidity, and everything else, whether the numbers are misleading to the public or not. The computer doesn’t care. It’s just reading the model or the model average, depending on the scheme.

Private companies have extremely sophisticated algorithms that update their databases continuously as new data pours in. This is how you can find out when the rain is going to start and stop, or what the temperature is going to be at 5:00 p.m. at your home. The information you get is reasonably right reasonably often because the models are reasonably good. But not good enough to be reasonably right essentially every time when being wrong can have dire results.

Dangerous weather requires a different philosophy. You might take a chance and leave your umbrella at home with a 50% chance of rain, but a thinking person would not ignore a 50% chance of a life-threatening storm. For extreme events, forecasts must lean toward the extreme, or be presented as a range of outcomes, as in the forecast cone for hurricanes. A reasonable-worst-case forecast philosophy, in theory, gets most people who might be affected ready for impact in time to mitigate damage or save their lives.

These days, when a storm is threatening, television weathercasts, websites, and pretty much every weather app present two very different messages about the coming weather at coastal locations, and they often do it concurrently or one after the other. They show the cone and sometimes the watch/warning area generated by the National Hurricane Center. These and other NHC products take into account the intrinsic uncertainty in the forecasting process. But simultaneously we get specific definitive forecasts for the weather in coastal cities presented as if there were no uncertainty at all.

The same vertical forecast bars with numbers and icons we see every day are determinedly and deterministically used to depict the coming weather. The case for presenting the definitive, deterministic forecast when a tropical storm or hurricane is threatening usually goes something like, “it is the best information we have”. But we know that is not true. The probability of something very different happening is about the same — within the noise.

This problem does not just present itself with storms that come from the tropics, it comes up anytime the weather conditions vary dramatically over a short distance. A slight, unforecastable difference in the position, size, or strength of the storm can make a tremendous difference in the weather experienced at a given coastal location. But our trusty human and digital weather forecasters give us specific, detailed forecasts anyway allowing for just one possibility — generally not the reasonable worst case.

On an average day, this forecast-presentation system works, more-or-less. When people see a thunderstorm icon, a percentage, and a temperature for a week from Saturday, they understand that there is leeway built in, and are not hugely surprised if it is sunny. If they see cloudy-with-showers forecast for 3 or 4 days from now, however, and they get hit by a hurricane, they have been dangerously misled.

Before database builders and graphics designers started dictating how forecasts are presented, individual forecasters and their communications philosophies controlled what they were going to convey each day and how they were going to convey it. The content and tone of the message was crafted to reflect the significance of the uncertainty in the forecast, and to characterize the changing risk as a threat was developing. On the other hand, computers by their cold, soulless nature are tone deaf, and therein lies a small part of the problem. The rest of the problem is with their human masters who force a boiled- and dumbed-down forecast presentation format onto potentially dangerous, uncertain weather situations.

This concern was a point of discussion when the National Weather Service’s NDFD (National Digital Forecast Database) was proposed in 2003. My idea was to put an asterisk in the data field for any time period when the odds of an extreme event were above a certain threshold. A second extreme-weather-communication framework would be developed for those starred-out time periods so there could be no confusion whether the reasonable range of possibilities on those days included dangerous conditions.

Obviously, that idea didn’t get traction, but the National Weather Service has dipped its toe into these waters by enveloping the weather icons on its city-forecast pages with bright-colored wrappers. Some private vendors make modest graphical modifications as well when severe weather watches or warnings are in effect, but without fundamentally revising their presentation formats.

More than 50 years ago, Marshall McLuhan coined the concept that “the medium is the message”. It is unlikely he was talking about weather-forecast communications, but the idea is apt. If you present the forecast the same way for an extreme-weather event as you do every other day, the format sameness counteracts the idea that there is a special level of threat.

Add that to the inherent confusion in communicating specificity in one breath and intrinsic uncertainty in the other, and the task of assembling a clear message is transferred to the average guy trying to figure out what to do — certainly not where it belongs.