as a grad student in atmospheric science, I am a little skeptical of the applicability and robustness of these approaches. There are people in our group who apply machine learning ideas for prediction, and I use them for data analysis, so I am not against data-based approaches. however, the partial differential equations which govern the atmosphere are known, can be approximated in a computer, and have been done so for decades. It seems silly to pretend otherwise, and the best/current weather predictions use data and physics.
This report make some pretty intense statements about the "limited success" of "analytical techniques", but in the end, they only claim to make an improvement of 1-2%. The kicker is that they make this comparison to the NOAA model, which is one of the crappiest models out there. The model the europeans use is about 5-10% more skillful, and it is generally recognized that the USA has fallen behind in this regard: http://cliffmass.blogspot.com/2012/03/us-fallen-behind-in-nu....
This report make some pretty intense statements about the "limited success" of "analytical techniques", but in the end, they only claim to make an improvement of 1-2%. The kicker is that they make this comparison to the NOAA model, which is one of the crappiest models out there. The model the europeans use is about 5-10% more skillful, and it is generally recognized that the USA has fallen behind in this regard: http://cliffmass.blogspot.com/2012/03/us-fallen-behind-in-nu....