80s next Wednesday!? Is it possible? sure. Is it likely? No.
I had many people reach out with screen shots of their iPhone forecasts Monday morning. “Is it really going to be in the 80s next week?!” people were asking.
It’s pretty obvious the iPhone forecast either used exclusively, or was heavily influenced by NOAA’s American (GFS) computer forecast model. That model has been notoriously awful lately, especially on temperatures beyond 5 days out. It’s a good reminder that those ‘app forecasts’ are often not actual forecasts but computer model data or maybe an average of models. Your local human forecasts are still the best bet.
The American model’s Sunday evening run had a high of 87 next Wednesday, April 12th. The very same model, just 6 hours earlier had a high of 48 for that exact same date in the future.
6 hours later than the evening run, the overnight Sunday night had a high of 75. So, the same model had a high temperature for Wednesday, April 12th ranging from 48 to 87 all within the same day.
Computer models typically go back and forth this time of year especially. March and April bring some of our biggest temperatures swings of any point the year and we know just how volatile it can get. The last few days are a great example; from a late season raging blizzard Friday night to our first 50 and 60 for southern Minnesota Sunday afternoon.
One way we can measure this volatility in statistics is standard deviation. For April 12th, that number is 11 degrees (in case you’re curious the standard deviation in mid August is just 6 degrees). That means that 68% of the time, the high temperature on April 12th will fall within +/- 11 degrees (one standard deviation) of 53 (which is the median number for the date using 1873-2022 data). 95% of the time it falls +/- 22 degrees (2 standard deviations) of that date. Basically, temperatures can land anywhere within a 44 degree range in mid April. The use of an ‘average high’ as we all know, is kind of pointless in Minnesota much of the time, but the range in mid August, the least volatile time of year, is about 26 degrees.
You’ll notice, a high of 87 falls outside that biggest range. Statistically speaking, a high of 87 has a less than 1% chance of happening, and in fact, the record high is 83 for the date. So, the American model is basically forecasting a record high temperature next week. That’s not very believable considering we just finished a March that was more than 4 degrees colder than normal. The model itself doesn’t even believe it and has been bouncing all over.
If we look at two other longer range models, the European and Canadian models, we get numbers of 55 and 62 for high temperatures next Wednesday respectively.
A responsible forecaster would give you a high temperature that’s the most probabilistic at this point. When we look at the whole suite of models, including ensemble models (which are the models run many different times with slightly tweaked conditions to get different outcomes), we get a median high temperature of about 63 degrees next Wednesday. The odds of hitting 70 next Wednesday come out to be about 12% and a high of 80 is likely, at most, a third of that.
This is why we meteorologists always caution you against looking at one computer model run’s most crazy, extreme run posted by someone that goes viral. We see this TOO often in the winter, but it also rings true for things as simple as temperatures. A responsible forecast is one that looks at many model runs over time. We look for trends and we know each model’s bias, strength’s and weaknesses. That’s why, at least for now, the job of a human meteorologist is safe. AI may be able to write a high school paper, but it cannot forecast weather… yet.
Comments