When I was obsessing about the weather forecasts earlier this year, I read a lot on basic weather theory, hoping to better understand forecasting methodologies and, specifically, how all these intelligent people arrive at different results. Once this mystery was solved, I thought I’d tackle macroeconomic theory (ha ha!).
I went to my library and picked up some promising books:
And before you ask, no I didn’t read them all cover to cover. Do you think I’m some sort of freak? Wait, don’t answer that.
I’ve had weather theory drilled into me in sixth grade when I lived in the hurricane belt and again when I started flying. The basics haven’t changed:
I learned a lot from the books, and eventually found UIUC’s weather site which filled in the blanks. Here’s the deal: Weather is an inherently chaotic system. A popular saying related to chaos theory is “when pillow talk results in uproarius laughter, it can cause a hurricane on the other side of the world.” In other words, a small change, even a butterfly flapping its wings can have significant consequences. Thus, 100% accurate weather forecasting is impossible.
Weather forecasters are cagey on how accurate they are, which is somewhat ironic given they measure everything, but 85% is the number I’ve seen bantered around. This may seem low at first, but it includes freaky weather, which we seem to be having more of lately.
One obvious problem in forecasting is your local meteorologist is forecasting for a wide area. For example, I live in the suburbs east of Seattle. Most of the forecasts are for Seattle, Everett, or Bellevue. My home is more inland and tends to be warmer in the summer and cooler in the winter. The 700′ elevation difference and proximity to the foothills means we’re also more likely to see hail pellets or brief snow flurries.
Meteorologists also get a bum rap because people are predisposed to remembering only the bad forecasts, on days where they’re trying to do something fun. There are five basic methods used in forecasting.
- The Persistence Method is the most straightforward because it assumes patterns don’t change much. For example, if it’s sunny and 72°F today, the method places a high likelihood that it will be sunny and 80°F tomorrow.
Today Tomorrow The day after
In the movie L.A. Story, Steve Martin’s character, a TV Weatherman, wants to get out of work for the weekend. He pre-records the forecast and comedy happens: it rains. Normally, he would have been right because the weather in Southern California is relatively stable that time of the year. In other places, the persistence method wouldn’t have worked well.
Surprisingly, the persistence method is good for long-term forecasts for monthly and seasonal forecasts where the other methods lose precision.
Trends method: frontal system moving southwest
The Trends Method is also very straightforward, and uses speed and direction to estimate future weather. Most commonly, it’s used to predict movement of frontal systems and areas of clouds and precipitation. It works well when systems move at the same speed and direction for a long period of time.
For really short term conditions, known as “nowcasting,” this is used to monitor storm progress. For example, if a line of thunderstorms is currently in San Marcos and is moving northeast at 25 knots, it will likely hit Austin (30 miles away) in about an an hour.
Climatology data for Seattle, first week of February
The Climatology Method is another simple method, based on averaging weather conditions over several years to make a forecast. This was the control case
in the second weather rodeo. For normal weather patterns, this works okay.
- The Analog Method is a slightly more complicated method of forecasting than the climatology method. It examine’s today’s forecast scenario relative to a previous occurrence of similar conditions. For example, suppose today is very warm, but a cold front is approaching. In similar weather conditions
last week, a heavy thunderstorm developed in the afternoon as the cold front moved through. Using the analog method, you would predict thunderstorms in the afternoon when the cold front moves through.It’s extremely difficult to find a perfect analog: weather features rarely align themselves in the same location at the same time. A small difference in time or location can yield vastly different results. (Chaos theory, again.)
- Numerical Weather Prediction (NWP) uses supercomputers to make a forecast based on mathematical models. It sounds high tech, and it is, but the NWP method suffers from a couple of flaws. First, the equations used by the models aren’t precise. (As Edward Lorenz noted in 1961, a couple of digits of precision made a huge difference.)
As a simple mathematical illustration, suppose you are 99% precise in your calculations for weather one minute from now. Your model uses the previous data to predict the next. If you wanted to predict the weather 10 minutes from now, you’d multiply the percentages:
0.99 * 0.99 * 0.99 * 0.99 * 0.99 * 0.99 * 0.99 * 0.99 * 0.99 * 0.99 = 0.904
After the tenth iteration, you’re already down to 90%. Ten more minutes forward, you’re hitting 81%. And so on. NWP forecasts beyond five days are unreliable.
The second flaw is the initial data is incomplete. I asked someone at NOAA about this, and his ideal scenario was to have a weather recording station for each square mile, though this is unrealistic. We don’t typically gather recurring weather observations in mountainous areas or over the ocean, and these are usually estimated. But assuming we did, that’s a lot of data. If you’ve ever driven across the country, you’ll realize the US is big. That’s about 3.7 million weather stations recording up to eleven variables (time, temperature, dewpoint, atmospheric pressure, precipitation, cloud coverage, ceiling, wind direction, wind velocity, visibility and lightning) times once an hour is just under a billion data points a day, or the amount of spam I used to receive at my previous email address.
Overall, the NWP method is the best of the methods at forecasting day-to-day changes, but it isn’t very good for the long term.