We receive quite a few submissions applying some flavor of ML/AI to weather forecasting. Most we decline, because the general point has been made that the technique works, and at least for @nature there usually isn't a case for another demonstration.
This one was different. First because it addressed a long-standing challenge in NWP. Second - and really intriguing for me - is that a key evaluation came from human weather forecasters, who judged the deep learning forecasts to be more useful/realistic than other approaches.
The paper follows on from a couple others I've handled on the overall topic of ML/DL/AI in geosciences: rdcu.be/cyE7M and rdcu.be/cyE7N
Definitely interested in the topic going forward but increasingly we're looking for major advances, rather than demonstrations that the approach works.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
