Let's Talk About Forecast Error (again)
If you are an avid reader of my weather blog (I know everyone has this website bookmarked of course) then you probably read this article I posted back in June:
Well, it is time to expand on this post now that we have finished the year 2020. You may or may not know, but I have a huge spreadsheet in which I enter all of my forecasts for the year. I track error, climatology, and trends. It is really awesome (if you are into statistics of course). Here is a sneak peek of one month's sheet and what things look like:
During 2020 I forecasted 166 different times (Quite a lot right?) I wanted to share some of the takeaways from the data I have compiled.
First, let's take a look at the raw data. This every single forecast for the year averaged out and graphed by day. There are a few things worth pointing out here.
1. My 7-day forecast has an average error of under 4 degrees for each low and high (Good Job Cory!) of course there are days with busted forecasts that have much higher errors but we try to forget that right?
2. The further out you get in time the higher the forecast error will be. (As mentioned in the June article as well).
3. Overall, my high-temperature forecasts are better than my lows. Except though for the big decrease in error for the day 1 lows.
Now let's took at some weighted data! I call this Cory Error Index, it takes into account how many days out of the month are actually forecast to adjust the raw monthly averages. These numbers are arbitrary but allow us to see trends.
So there are two takeaways here but I will put them together. During the winter months, the error tends to be much higher. This is expected since during this time of the year the troughing and ridging cause temperature swings every few days. This makes predicting the future much harder. You can also see this as the spread in temperature error is much thinner in the summer months than in the winter. This indicates a theoretically "easier" forecast.