From Meteorologist Steve Glazier,
My forecasts are getting more inaccurate, and I don't like it! Every time I make a forecast, I go back through the data to see what actually happens. I mark down the observations and compare those numbers to my projected numbers. My error is rising.
However I believe it's a seasonal thing. We're coming off the summer months which, statistically, show I am best at forecasting. Here's what I do.
I forecast exact numbers for the Burlington International Airport in South Burlington. I write down a high and low number. Then the next day I write down the actual high and low and then calculate the error. It's a very simple verification which I can learn from now, and in the future.
So how am I doing?
Not too bad. I also jot down the model numbers from MOS outputs, which generate highs and lows for the airports too. I have a better accuracy than these model numbers when forecasting for KBTV (Burlington) which is good. It means I get to keep my job I guess.
As I mentioned were coming off my best months. August 2012 was my best month this year, with the least forecast error. Since I forecast the seven-day period, I have thirteen highs and lows to calculate. Not 14 because I forecast day 1 high, then day 2 low and high, day 3 low and high, etc.
My forecast error for all 13 highs and lows together, for August 2012, was 33.5° F. I averaged 33 and a half degrees off for the entire 7-day period. I'm happy with that. However in September so far I've had an average error of 43.1° F.
I thought of writing this blog for a couple of reasons. One I want to show you some of the behind-the-scenes work I do here at Fox & Abc and how I'm constantly trying to get better. The other is to show you an example of weather forecasting. I think forecasting in fall and spring can be much tougher than summer and winter. Fall and spring are transitional months. The jet stream can take sudden dips and sudden rises. Pockets of cold air plunge south, while warm air surges north. It can be an active period for weather. It can be a windy one, and certainly busy which lowers forecast confidence.
Here's another example. Yesterday (Wednesday) I was forecasting a mostly sunny day for Friday and a high of 66 @ BTV. Now that has changed due to recent model updates and more information, to 58 @ BTV Friday with thickening clouds and rain moving in. The forecast time was only -48 to -24 hours and these sudden changes can happen.
So as you watch the forecast know that the first four days are generally close. My data shows I am within +/- 4 degrees with days 1-4. However days 5-7 my error increases to around +/-5 degrees or slightly more. This error calculation is part of a larger project I'm working on to assess how I'm doing in each month and how I can improve. Next step: Precipitation forecasts and accuracy. I've already done some, and I'll release that probably later this year.