Panasonic claims to have made worlds best weather model

[Ben Holcomb]

Pretty interesting story that @Jeff Duda posted on his facebook earlier. A great read about Panasonic using a 11,000 core supercomputer to create their own global model. I had wondered why no private company had done this - Seems HUGE for the meteorology community at some point.

http://arstechnica.com/science/2016...-has-developed-the-worlds-best-weather-model/

 

[Zac Flamig]

I find their claims about the extra data being the key to be very interesting. It is rare that you will see extra data provide *that* much improvement. Look at the work done post-Sandy on the extra soundings done for that event, from what I have seen the extra data wasn't a significant contribution there.

That said, I'm not sure that the right combination of software engineers and meteorologists has yet been found to truly advance our (public & private) modeling capabilities.

 

[Paul Knightley]

I think that the rising tide of using Big Data will continue to develop this field - it's a trade off because the proponents of Big Data shout about the way it works, within needing to know the reasoning behind it (just crunch the numbers) but it flies in the face of the traditional meteorologists' view of desiring to be able to explain everything!

 

[rdale]

I find their claims about the extra data being the key to be very interesting.

I see what you're saying about global modeling - but then again init is everything On a smaller scale, Dr Mann (DTX) was part of a lake effect snow study that showed just one sounding site on the north side of Lake Superior resulted in massive improvements in LES prediction.

 

[Jeff Duda]

At least the article provided some reasoning for the possibility of an improvement. Since the biggest gains lately seem to be from assimilating more and more data, and if Panasonic provided some observation platforms but also restricted some of the data that they fed to NOAA to use in the GDAS, and then used that data themselves (along with, presumably, all of the other data that NOAA also ingests for data assimilation), then I can see how they could take the GFS source code and improve upon it...

...marginally:

the PWS model has scored a .926, compared to a .923 for the ECMWF and .908 for the GFS

Wow...a whole 0.003 and 0.018 higher for 500-mb AC compared to the ECMWF and GFS, respectively...excuse me while I make an obscene gesture.

How about low-level temperature? Moisture? Precip?

 

[MClarkson]

Interesting... very interesting. There are other companies that run their own weather models, (often local scale WRFs for renewable energy forecasting), but I don't think there are many(any...RPM?) that even try straight up to compete with the full globals.

I would also like to see surface wind, precip, temperature, dewpoint verification stats, the things that matter most to most of us.

The slide about the snowstorm is a little dishonest... notice how they left off the GFS, which was similar, in at least a few of its short term runs, in correctly predicting an offshore miss for the NYC area and points south.

 

[rdale]

Remember - this isn't a cheap endeavor. It would probably be FAR less expensive to just grab ECMWF if this wasn't notably better for their purposes... Granted I'm not saying take their word blindly - but unless their CEO is a weather geek and doesn't mind blowing money for no reason (doubtful) I'd give this some merit.

 

[MClarkson]

Oh I agree, Panasonic is a large company with plenty of resources and a good reputation, so they are capable of pulling this off. The amdar data links are awesome.

But I am going to need actual verification scores on boundary layer parameters before trusting this model... or any model.

 

[Stan Rose]

I'm highly skeptical. TAMDAR is nice but has limitations. If they want to be taken seriously they need to release more of the data for public scrutiny. Otherwise, they're no different than any other company making outlandish claims (and rdale, that's really all they have to do--make the claim--because they're are plenty of suckers willing to buy!) I don't always agree with Cliff, but he's right here.

 

[rdale]

I see what you're saying - but they aren't selling access to it as far as I know. There is no reason for them to plug it unless they really can back it up.

 

[rdale]

WxGeeks is apparently going to discuss this model next week.