Actually its the humans that came up with the laws of thermodynamics that totally own the model output. To the OP, the equations do not change unless Euler or Newton or Einstein have recently been undone by some other genius. What does change and can only get better is the initial start data and the boundary conditions that are used for each run. And, of course, these could well use historical data and blend that into the programming, but at this point its getting very complex. I would suggest looking up Navier-Stokes for some basic fundamentals on fluid flow analysis. Also check out multiphase flow. Honestly, we cant get this stuff right in a straight pipe never mind in the whole atmosphere, but its getting better all the time and the GFS Parallel has been kind of consistant. edit - of course you're right and the computer programme can get more accurate and take on more data ad infinitum (on what its working with) and the grid can be smaller - like John says vertical and horizontal - but this does all come down to two words that hasn't been cracked yet. Chaos Theory.