I got this month's edition of 'Weather' today and found an obituary of Lorenz who died in April. He was Professor of Meteorology at Massasuchsetts Institute of Technology in the early sixties and was the one who coined the 'butterfly effect' phrase in 1972, where a tiny event in one place can have a profound effect on the weather in many others. And he was using a computer to simulate weather predictions and found, by accident, that if he only put in the very same data, but minus some decimal points - like 0.304218 shortened to 0.304,- the end result forecast was completely different.

He used his experience to suggest that long range forecasting more than a few days could never happen. Too many variables.

In this I think we see a parallel with our own model watching. Clearly GFS and all the others are constantly updating their data and even in six hours this will change a lot more than a few decimal points. Hence the reason why the 00z can be very different out at 300 hrs from the succeeding 6z run. Todays computer models are vastly more sophisticated and faster than in Lorenz's days but there are still too many different routes to take every few seconds.

I think the best we can do, as many members have seen, is to look for trends, particularly where the forecasting computer keeps on picking up on something basic and/or recurring, eg, a preponderance of mobility/frequency or positioning of anticyclones etc. I love looking to the furthest extremes of GFS (maybe UKMO is very wise to restrict theirs to a week) but know full well that a snow-lover's dream model can be snatched away with the next set of charts. And I've learned, through bitter experience, never to see anything as set in stone. "Life's a cruel teacher but you learn, by God you learn" as Antony Hopkins said in another context in Shadowlands. It applies to us.