Monday, November 18, 2013

Improving Weather Forecasts

How you can improve weather forecasts ...

By far the biggest source of error in 1-2 day weather forecasting is in the communication of the forecast from the meteorologist to the user. That is, the computer models could get it right but the information gets garbled in the communication process resulting in error.

The communication errors stem from two sources. One is that a significant percentage of meteorologists do a very poor job of presenting the forecast. On a typical 3 minute weather segment, the meteorologist may spend 2:50 of that time talking about local temperatures, past weather, current weather, frontal boundaries, and pressure systems. Then the last 10 seconds they flash the 7-day forecast filled with stupid icons that don't accurately convey the actual forecast, and they expect the user to digest the entire forecast in those closing seconds of the segment.

And then there's the National Weather Service forecasters who absolutely butcher the meaning of precipitation probability. Every day in the summer the forecast includes an "XX% chance of showers and thunderstorms". Well the XX% is not the probability of having at least two showers and two thunderstorms at the user's location. It's the probability that 0.01" of rain will fall on your head. Quite the difference.

However, the meteorologist isn't the only one to blame in the communication process. Oh no, you're likely at fault too. Why? Because most people hear the weather forecast in passing or from someone else, and have a hard time remembering the actual forecast if called upon. Ask 10 people on the street what the weather forecast says for the next few days you'll likely get 10 different responses. When I taught an intro meteorology class at Penn State, one day I played the recording of the campus weather service forecast at the beginning of class. At the end of class I asked them to write down the forecast. Almost 90% couldn't. People generally have selective memories.

So what can you do to improve weather forecasting? Three things. One is if tomorrow's weather is very important to you for planning, then don't listen to what your friend says what the weather will be. Information will only get garbled in the extra transmission process. Only listen to the forecast directly from a meteorologists mouth. Second, be sure to listen to the forecast carefully (and maybe listen a couple of times) so that you can recall the forecast an hour or two later if you need it. Third, if forecast accuracy is important, then listen to the forecasts of several meteorologists and take the average. Numerous studies have shown that the consensus opinion will outperform the individual forecasters nearly every time.

As for improving the National Weather Service in this regard, well, I pointed out their probability problem over 15 years ago and it fell on deaf ears. Sad to say they're kind of stuck in their ways. As for the TV side, the door is wide open for a meteorologist to break the mold and present weather in a different, more intuitive manner that reduces the potential for miscommunication error.

No comments:

Post a Comment