When AccuWeather announced they would begin issuing 90-day forecasts in April, science writers, and the meteorologists they interviewed, were notably dubious. Sort of like how they were when AccuWeather debuted a 45-day forecast in 2013.
At the time, Gizmodo’s Maddie Stone spoke to Jon Porter, a VP at AccuWeather who basically evaded the issue of how impossible these kinds of predictions are. As she wrote:
There’s no nice way to say it: this forecast is nonsense. And I’m pretty sure AccuWeather knows it — the company itself told me that the 90-day forecast tool is “more about long-term trends than individual days”. How on Earth the average layperson is supposed to glean that from a tool that literally presents individual, day-by-day forecasts out to 90 days, with no obvious disclaimers, is beyond me.
AccuWeather founder Joel Myers spewed gibberish in an interview with NPR’s Robert Siegal on the new long-range forecasts:
SIEGEL: And do you attach a level of confidence to the forecast 90 days from now in Washington? It’s obviously less confident than you are in next Monday’s weather in Washington...
MYERS: Well, exactly, and that’s the important trend. But we don’t give a confidence number. I mean, I wouldn’t take the 88 day and say, hey, there’s a ballgame from 3 to 6 o’clock, and it’s going to be such and such. But if I’m going on a vacation to Paris and I’m going to be there for five days between the 83rd and 88th day, this forecast will give you information that will be better than you can have figured out in any other way.
SIEGEL: Well, yeah, I didn’t introduce the idea of the baseball game. That came from AccuWeather saying that would be one reason why you would consult the forecast.
All this seems damning enough, so you would be forgiven for simply forgetting about the audacity of AccuWeather’s impossible claims. It’s probably bullshit, right? Sure. But industrious Deadspin reader AH did not forget, and set out to measure just how bullshit it is. As he wrote us recently:
On the morning of April 20, at approximately 11 a.m., I took 5 screen shots of AccuWeather’s proposed outlook for May. For the entire month of May, I matched the actual results compared to what was forecast, and my results show just how far off they were. There’s no way in good conscience that AccuWeather should be able to advise people of a forecast multiple weeks in advance.
Although he currently works on Broadway (our readers are a multi-talented bunch) AH claims credentials to back up his skepticism, including a master’s in meteorology. Either way, the data don’t lie. Tracking Accuweather’s forecasts for New York in May against the actual temperature and precipitation reported day-of by AccuWeather got the following results:
AccuWeather’s forecast was off by as much as 26 degrees one particular day. The average discrepancy between the predicted temperature and the actual temperature was 9.7 degrees for highs and 7.2 degrees for lows. Predicting a steady increase in temperature from the beginning of the month through the end of the month would have been more reliable. Of the 11 days it actually rained, the AccuWeather forecast only called for any precipitation on two of them. In other words, it’s total bullshit. And that’s not even measuring the accuracy all the way out to 90 days.
When we showed AccuWeather AH’s data, Porter gave an answer similar to the one referenced above, basically saying that people should know better than to take long-range predictions seriously:
Though we’re happy to hear from the user community, a single study for one location carried over a relatively short period of time isn’t a reliable measure of accuracy.
More importantly, it’s not reflective of how people use our long range forecasts. People understand that forecasts are going to evolve. Our audience uses the 90-Day forecast to get a flavor of what the weather will be like with greater detail and accuracy. As a particular date gets closer and users start making plans, they will be continually monitoring, up to the day of and even up to the minute, with the unique AccuWeather MinuteCast®.
AH praises outlets like the Weather Channel and the National Weather Service for providing long-range generalized comparisons to statistical averages.For other dedicated meteorology hobbyists, AH recommends the Weather Nerd app for the most up-to-date information. And as a good rule of thumb: remain skeptical of anyone trying to tell you the exact percentage chance of precipitation on Labor Day before the Fourth of July.