Perception is in the icon of the beholder
Posted February 2, 2011
I woke up this morning and, since we had forecast today to be unusually warm, I checked my iPhone weather app to see what the temperature was outside. In addition to that current temperature, the app also provides an "icon-and-numbers" forecast for each of the next six days. I have a couple of cities programmed into the app for easy access to their forecasts, and upon checking, I noticed the forecast for Apex and Raleigh — about 10 miles apart, as the crow flies — were significantly different. One featured all rain and temperatures well above freezing, another featured sleet and temperatures near freezing.
This brings up a very interesting point about uncertainty in weather forecasts and how that information is communicated (or not communicated) that with the end users of those forecasts. I will not make a comment about the accuracy of these iPhone weather app forecasts; it is sufficient for this discussion to say that such a change in weather conditions over a small distance is not completely unheard of around here, but it would be rare. Further, it is beyond the state of our science to be able to state conclusively that the line between "all rain" and "sleet and freezing temperatures" will be between two points that lie only ten miles apart.
Icons and Context
It is common practice for weather forecasters — from newspaper to TV to the web and on mobile devices and beyond — to condense a day's forecast down to a single icon. We have been doing this on WRAL-TV for decades; although, when we started, those extended forecast graphics were only used after a full weathercast and within the context of what the meteorologist had been talking about. So, if there was some question about whether we would see rain or snow on a given day, someone seeing that extended forecast graphic would have heard that and could interpret the icon accordingly, as the most likely of scenarios but not the only scenario.
That context is important! Taken out of the context of the broader discussion, a 7-day forecast treats all seven days equally: Each gets an icon and a couple of numbers. That is a problem since these icons-and-numbers forecasts make us look as sure about day 7 as we do about day 1. As any meteorologist worth their salt will tell you, our ability to forecast the weather accurately decreases the further out in time we go. Depending on the situation, we may feel confident about a week's worth of numbers; more often, our confidence drops much more quickly. People who watch a weathercast on TV or who read the text descriptions we include with our forecasts on WRAL.com will get a sense of that "big picture." Looking at just the graphic or another icons-and-numbers forecast will not benefit from that context.
This is exactly why we provide a text discussion with our 7-day forecast on WRAL.com. In talking with frequent users of WRAL.com's weather content, we hear time and again how popular that text is. That's great, as it means these folks (and hopefully you!) are getting the "big picture."
Picking the Icon
This opens up a bigger discussion of how we choose the icon for a given day and how people interpret that icon choice. Should we choose the icon that matches the conditions we'll see for the longest part of the day? Or should we choose the icon showing the most significant or impactful conditions, even if they will only last a short time? For example, if it is going to be overcast with rain in the morning but sunny and dry in the afternoon, which icon do you choose? Cloudy with a raindrop? Sunny? Partly cloudy?
How about another scenario: What if the weather for a given day, be it tomorrow or a week from now, is uncertain? We have plenty of days around here that we have to choose between two equally likely scenarios that have very different implications. Take for example, a situation where we have a stationary front separating sunny skies and warm air from cloudy and cold conditions. That front may shift a few miles in either direction, bringing a dramatic change in the weather for those in the middle, and our science isn't quite good enough to say conclusively where exactly that front will set up on a given day.
Consider this: Assume you're making a forecast for a football game in an open-air stadium in that "battle zone". Two outcomes are possible — sunny and warm or cloudy and cold — and they are equally likely. What icon do you pick? (Sound far fetched? Something like this actually happened with an 80,000-person football stadium in the middle. Read what happened; scroll to the sidebar at the bottom of the link.)
Friday's Forecast and Moving Forward
In the case of the iPhone weather app forecast for Friday, there is clearly uncertainty, especially since it involves the chance for wintry precipitation in central North Carolina. However, because of the format of the forecast, that uncertainty — and what it might mean for how you can plan around it — is not communicated at all. The fact that the forecasts for Apex and Raleigh were so vastly different is evidence of that uncertainty, but without the benefit of the bigger picture, the person checking the forecast has to plan their day based only on an icon and some numbers.
A large portion of the 91st Annual Meeting of the American Meteorological Society was dedicated to addressing just this issue. How can we quantify or characterize the uncertainty involved with a forecast? And once we've done that, how can we best communicate it to the users of weather forecasts, on TV, on the web, on mobile devices, and elsewhere? While we are trying to figure out the best "hows", the "whys" and "ifs" are clear. We should be sharing this information with you, and whenever possible, we do. Further, as a consumer of weather forecasts, you should always look beyond the forecast icon and numbers.