I can only imagine about some of the flights across and around the Pacific...what may look benign on satellite isn't always benign. The job those guys do -- both flying and forecasting -- isn't easy in the least, and I hold a great deal of respect for everyone there.

Modelling has gotten a lot better in the past 10 years. The dynamical models got better, finallysurpassing all of the statistical models, but a lot of those gains have come solely from better data and better computing facilities. We've had some improvements in the physics, though not enough yet, but the dynamics have remained largely unchanged (which is fine). Lots more data, particularly from all of the new satellite observing systems, and much better computers to run higher resolution models have been the biggest benefits.

Funding is nowhere near what it could be, and certainly not near the potential damage these systems cost. Unfortunately, not a lot of people want to go into that area of work -- it is very demanding and requires interest in both meteorology and computer science, excelling specifically in the latter. Not many people fit that bill, and some of the big centers -- mainly on this side of the pond -- haven't been able to recruit people to work on modelling like they might have hoped. It's not as big of a problem outside of the US, however, but that's more a reflection of people in the nations and not any policy decisions. Anyway, most money these days goes into climate modelling, which is fine in its own right but not any help when it comes to the weather. There's some left over for the general weather models, but personally I think we'll see a 5-6 year period here where the models consolidate their gains from the past 10 years before taking another big leap near the end of the decade. It'll likely again be technology that drives a lot of the gains, as there isn't a lot of money going into specifically improving upon the physics within the models...we have ongoing research to better understand all of the physical aspects that may need to be forecst by a model, but not a lot of work on how best to represent it in the model. That's probably 10 years down the line, but it'll be a great leap once we get there. Just my thinking, however.