Met Office AI

johnalison said:
What are your thoughts about it? Knowing nothing about AI myself, I get the impression that it can be very good at dealing with lots of messy data, such as often found with weather and medicine, and no doubt policing as well. There must surely be a statistical limit to how long we can expect accurate forecasts, but greater accuracy within the current week or so would be nice.
Click to expand...



Like most scientists, my initial reaction to anything new is to be sceptical. AI prediction can only be a pattern matching of statistical relationship approach. Highly sophisticated variants no doubt and way beyond my comprehension. It is, to users generally, a black box approach. Scientists do not like black boxes.
Work so far is encouraging but a full AI system has not yet evolved, as far as I know. The ECMWF AIFS starts for a NWP Model analysis. Starting an AI model using raw observational data rather than a NWP analysis is a different kettle of fish. They will probably get there. However, my understanding is that the likely outcome will be some form of composite system using the best features of AI and NWP.
I suspect that the AI lobby will claim that the operational meteorological meteorologists are dragging their heels. However, there is far more to weather prediction than the tip of the iceberg that we see. In addition, knowing something about the many and various data sources either their highly disparate volumes, resolutions, accuracies, forms generally, my guess is that using raw data with its advantages will be far from trivial.
So, in summary AI will no doubt play a major part in the future but the myriad outputs required will be a major problem. Softlee, softlee catchee monkey. Expect a gradual implementation. Those are my views of the rather cagey Met Office page.
 
IMG-3369.jpg
 
You must log in or register to reply here.

Share this page

Top