The Bureau of Meteorology installed a multi-gazillion dollar supercomputer a year or two back, with the aim of improving the accuracy of weather forecasts.
How accurate are they? Well over about a week, I compared forecasts (including up to a week ahead of the day) against the actual weather reported the next day. I was seeking to answer these key questions:
- Do they alter forecasts markedly as the day gets closer?
- Are the all-important forecasts the day before actually accurate, so you know if you can leave the house without taking a brolly?
- Is this kind of detail so deadly boring and sleep-inducing as I’m suddenly suspecting it might be?
|Day||Forecast days in advance||Actual weather|
|Mon 9th May||Mainly fine||Mainly fine||Mainly fine||Fog then fine, 18||Fog then fine, 18||Fog then fine, 18||Fog then fine, 18||Fine, 17.3|
|Tue 10th May||Mainly fine||Mainly fine||Mainly fine||Fine, 18||Fine, 18||Fine, 18||Fine, 18||Fine, 18.5|
|Wed 11th May||Mainly fine||Mainly fine||?||Shower or two, 17||Early shower, 17||Early shower, 17||Shower or two, 17||Fine, 19.1|
|Thu 12th May||Mainly fine||?||Mostly fine||Fine, 18||Fine, 18||Fine, 18||Fine, 18||Fine, 16.7|
|Fri 13th May||?||Mostly fine||Mostly fine||Mostly fine||Mostly fine||Mostly fine||Fine, 18||Fine, 18.3|
|Sat 14th May||Mostly fine||Mostly fine||Mostly fine||Mostly fine||Mostly fine||Fog then fine, 19||Fog then fine, 19||Fine, 21.8|
|Sun 15th May||Mostly fine||Mostly fine||Fine, cool||Mostly fine||Mostly fine, 17||Mostly fine, 17||Fine, 17||Fine, 19.3|
Notes: ? — oddly, on Saturday 7th of May, they weren’t brave enough to forecast more than 4 days ahead. And don’t ask me what the difference between “Mostly fine” and “Mainly fine” is.
If you’ve got nothing better to do: See all the figures (XLS, 19.5Kb)
In conclusion, it seems they’re pretty accurate, with temperatures usually within a couple of degrees of that forecast the previous day. Even the advance forecasts don’t change a great deal before the day. Mind you, we did have a run of fine autumn weather. But still, I’m suitably impressed.
7 replies on “How accurate are weather forecasts?”
Sounds like my Year 12 (1992) Maths project
Crunched (copy typed) 5 years worth of actual and forecast data into Quatro Pro running on an XT machine
Compared accuracy to a number of simple forecast moving averages etc….
Remember writing formula, coppied it down the several thousand cells and when to have dinner while the XT whirred away to finish the calculations..
Stuck a wizard of ID cartoon on the front that had the town cryer anouncing “There is an 60% chance of rain tonight, 40% chance of rain tomorrow and an 80% chance we are wrong on both counts”
Do you feel better now? *LOL*
I found the other week that Edwin Maher is working in China now as a weather forecaster. It was probably the only good thing in the melbournemagazine.
I’d like to see the results during a more changeable Melbourne weather period. Mid – late autumn is hardly a challenging sample.
I agree, that looked like a pretty stable and consistent week anyway so it probably isn’t a complete picture.
I’ve noticed (here in Adelaide) that there can be pretty stable weeks, so the temperate provided 7 days in advance can be a good indicator, but other times the temperture for a given day will change significantly throughout the 7 day forcast and ultimately end up nothing like originally predicted.
Earlier this year I started keeping track of Melbourne and Glasgow forecasts and actuals … for about a week, until I forgot to check the weather sites one day, and never got started again. I was intending to look at the same question as you, but also to compare Melbourne’s variability to Glasgow’s.
Even without the data to back me up, I’d say that the local forecasts are inaccurate more often than Melbourne’s, and even when they’re right, it’ll generally be for a shorter period in the day.
bout the weather,