In yesterdays Look Ahead video I explained the difficulties models are having in predicting conditions over the coming weeks.

It's always hard to point to exactly why the problem is occurring, and almost as impossible to put the issue into numbers.

However, one good graph to look at is the verification of 500mb heights between 20N and 80N across models.

Below is the verification of the 5-day forecasts from several models (colour coded). A perfect anomaly score would be 1 and that would give an accurate forecast. The verifications are daily back to 21st August.

Notice it's the ECMWF (red triangles) which has proved most accurate when forecasting 5-days in advance, the GFS is close and the Canadian isn't to far away either. Can you see the general drop in forecast accuracy, especially the ECMWF and GFS?

Of course these numbers cover the whole of the northern hemisphere and so it doesn't tell us how accurate individual forecasts have been but it is certainly a good guide.

Now here's the verification for the 10-day forecast charts (240hours). Of course, you'd expect these correlations to be low, but just look how low they have become. Below 0.5 is really getting into next-to-useless territory and many of the models have been within that.

Again, a downward trend since August, with the GFS and Canadian models being notably poor.

This is why we have been focussing on the ECMWF as it does have a better track record, especially in recent weeks. Of course, that is not to say things won't change!