Accuracy of weather models - climate science Q.


One of Us
Ski Pass
Feb 17, 2010
Melbourne Vic
Hey everyone,

I'm wondering about the accuracy of weather models.
I *assume* that the models themselves make use of some sort of predicted vs. actual statistics and results.
Are these results publicly available?
ie. is there a % as to how accurate GFS is at say 72Hrs out? what about just 24hrs out? What about the ACCESS Models? is that more accurate at 168 out compared to 24 hrs - relative to say GFS that is..

Ideally i would like to see a/some graph(s) comparing different Models predicted scenario vs Actual scenario, at different prediction times.
Also it would be interesting too see accuracy vs. time of year, or accuracy vs. location.
Personally i think Perth has very predictable weather and melbs less so, however i would be interested to see if that is reflected in the accuracy of Models.

Can anyone shed any light on how the Models themselves actually gauge their own accuracy, eg. what metrics are used to use one model over another?

I'm half decent with my maths and statistics so don't be afraid to get technical ion your answers.



Hard Yards
Jun 28, 2007
I have a very basic understanding of this from uni, but most models are numerical models which take equations relating to thermodynamics and atmospheric processes

for example the mean zonal (east - west ) wind

is u= (-1/coriolis force * density )*(meridional pressure gradient force)

so it solves these equations for temperature, pressure, wind in 3 directions, vertical, e-w and N-S and one another which i cant remember,

models usually divide the earth into a grid and solves of about 50km X 50km and calculates each of these equations with inputs (observations) at each gridpoint (there are thousands if not millions of these on the earth)

I think the main thing that varies between models is the resolution of the grid size, i think the ECMWF model uses a smaller grid than GFS, GASP, NOGAPS and i have found it has usually been slightly more accurate.

BOM have another model called MLAPS and LAPS these have smaller resolutions again (i think about 15-20km x 15-20km) if the resolution is smaller it requires more calculations, hence why these models are only 2 or 3 day, it would take too long to calculate a run for 7days.

I have only a basic understanding of this, so others please feel free to add
Remove ads with a
Ski Pass

Log in

or Log in using
Remove ads with a
Ski Pass