John's Forecast Results | News
No one hates a missed forecast more than a meteorologist. The challenge is to improve! Despite our profession's reputation, short term forecasts are reliable. But when they're not, they tend to be very memorable.
Here at 13WMAZ we value transparency in our news coverage. I wanted to apply that spirit to weather and give you a look at my overall forecast performance last month.
Meteorologists add value to the computer models you see on the air by giving context and meaning to that information. A skilled forecaster will also appreciate the strengths and weaknesses of each model and use it to create a superior forecast.
In January, I made forecasts for 16 shifts (both morning and evening). For simplicity's sake, the computer performance is an average of three frequently used models: FutureView (our branding for a proprietary model used by our graphics vendor) and government-run models called the GFS and the NAM. In short, the GFS has been a better short-range performer than the NAM or FutureView.
For this article I only evaluated temperature error: the difference between the forecast and the official observations at Middle Georgia Regional Airport. I took those errors and averaged them together to create one number for each time period. (There are much better statistical methods, but the point is to keep things simple).
The first number is my average error followed by the average model error, with the better value in bold.
Day 0 High (Today), 1.4°, 1.8°
Day 1 Low (Tonight), 1.9°, 2°
Day 1 High (Tomorrow), 1.9°, 2.4°
Day 2 High - 4.2°, 4.1°
Day 3 High - 2.8°, 3.1°
Day 4 High - 2.9°, 2.6°
Day 5 High - 4.8°, 4.1°
Day 6 High - 5.9°, 6.2°
Day 7 High - 7.9°, 8.8°
The results are generally similar because my forecasts are influenced by the models. One thing immediately stands out: the size of the errors increases over time. That's because the atmosphere is chaotic! Long term forecasts have more uncertainty than short term forecasts. That can also vary by season. During the stagnant summer days, one could forecast a high of 95 degrees two weeks in advance and probably be very close to the result!
My forecasts are slightly more skillful than the raw models in the short term. Sometimes, the models do not accurately represent or time small-scale cloud features. Any meteorologist will tell you that clouds are a huge culprit for busted temperature forecasts. It's back and forth two-to-five days out, but my results are a little better in the extended range.
My reason for doing this is to know where I stand and improve with time! One month is a very small sample for a comprehensive study so I hope to share more results in a few months!