I think there are two basic issues with the model output: firstly, the data that is plugged into the models as the starting point is not 100% accurate and therefore it doesn't matter how accurate the mathematical equations might be, those opening errors, however tiny, will multiply and eventually render the predictions wrong.  That is, of course, why they run ensemble suites, to test how tiny changes in the opening position will alter the evolution; as we have all seen, sometimes the effects are really quite significant within just a few days - whilst on other occasions there is reasonable stability for perhaps 5-6 days out (as we have seen recently). There is also the fact that measuring 'stability' by looking at the effect at a single geographical point isn't especially reliable: if that point is at the margins of an air mass boundary it will look different to somewhere that is more stable.
The second issue is that the mathematical equations are, and can only be, an approximation to how the atmosphere behaves; that approximation has clearly got better over time, and the more grid points within the model then the better the results should be - as evidenced by the very high resolution short-term ones. Even then, as we saw last weekend with the development of that small LP that brought snow to the south-east and Home Counties, some of even the high res ones simply didn't pick it up. If they had the same opening data then that suggests errors in the model.
Yes, this is the correct thought train on modelling. đ
Â
Originally Posted by: Gandalf The White