Started By
Message

re: IBM introducing the World's Highest-Resolution Global Weather Forecasting Model

Posted on 1/8/19 at 10:30 pm to
Posted by soccerfüt
Location: A Series of Tubes
Member since May 2013
65655 posts
Posted on 1/8/19 at 10:30 pm to
quote:

IBM
Me too.

Forecast is 100% chance of shower afterwards.
This post was edited on 1/8/19 at 10:31 pm
Posted by slackster
Houston
Member since Mar 2009
84856 posts
Posted on 1/8/19 at 10:40 pm to
So this is basically a global version of the HRRR? The HRRR has a 3km resolution. So does the NAM.
Posted by Tyga Woods
South Central Jupiter Island, FL
Member since Sep 2016
30054 posts
Posted on 1/8/19 at 10:42 pm to
quote:

NAM.



The only easy day....was yesterday
Posted by slackster
Houston
Member since Mar 2009
84856 posts
Posted on 1/8/19 at 10:43 pm to
Posted by TigerstuckinMS
Member since Nov 2005
33687 posts
Posted on 1/8/19 at 11:58 pm to
quote:

For hurricanes it probably just moves the cone earlier and more often than current models to gain accuracy.


I bet it's optimized for triangle math, too.
Posted by The Cool No 9
70816
Member since Jan 2014
9958 posts
Posted on 1/9/19 at 10:41 am to
Could be a really helpful tool for us on the gulf south. Good post - Sounds interesting.
Posted by Clames
Member since Oct 2010
16566 posts
Posted on 1/9/19 at 11:16 am to
quote:

The kicker is that you don't need huge computers to do the modeling. Once the training is complete the forecast run in 1/10 of the time.


On what data? That's the problem you seemed to have glossed over and why physics-based modeling is never going away. It's not the physics and our ability to accurately model the phenomena, it's obtaining and processing the vast amounts of data. It's a granularity problem, like modeling solid objects in CAE software, the size of the mesh determines the precision of the output, but using a finer mesh requires an exponential increase in the number of equations that need to be solved. Machine learning isn't going to solve what is fundamentally a hardware problem.
Posted by toddts
Fort Worth, TX
Member since Feb 2005
62 posts
Posted on 1/9/19 at 11:33 am to
quote:

On what data? That's the problem you seemed to have glossed over and why physics-based modeling is never going away. It's not the physics and our ability to accurately model the phenomena, it's obtaining and processing the vast amounts of data. It's a granularity problem, like modeling solid objects in CAE software, the size of the mesh determines the precision of the output, but using a finer mesh requires an exponential increase in the number of equations that need to be solved. Machine learning isn't going to solve what is fundamentally a hardware problem.


Yes and no. Software right now is what is helping to move Infrastructue (server and storage) technology capabilities forward despite the diminishing returns/ending of Moore’s law. Go look at IBM technology...hell, look at what Intel/AMD/NVidia is having to manipulate for their gains. It’s software and GPU’s that are driving AI/ML/DL and and it’s why infrastructure technology matters.

However, you’re right in that current classical based High Performance Computing (HPC and GPU) + AI/ML and Current SW Dev have their limitations (which we’re still not hitting - yet) . That is until quantum computing is unleashed in scale to the market ...which is 3-5 years away from being commercially available to all enterprise companies outside of just a handful that are in private bleeding edge partnerships with the few (less than three) tech companies who actually have Quantum h/w up and going today.
Posted by Duke
Twin Lakes, CO
Member since Jan 2008
35610 posts
Posted on 1/9/19 at 11:41 am to
quote:

So this is basically a global version of the HRRR?


Yeah, basically. They're going to pull data from weather stations (like the private ones populating wunderground). Which is good but it's not like it'll be pulling soundings from those locations.
Posted by CivilTiger83
Member since Dec 2017
2525 posts
Posted on 1/9/19 at 11:50 am to
So is this like the IBM smarter planet initiative or the IBM Watson is going to change the way businesses work initiative?

I can almost guarantee this will be all bark and no bite.

If they are able to come up with an extended accurate forecast that stretches to 21 days instead of the current 7 days that would be a gamechanger. I am skeptical with it being IBM and their propensity to hype the crud out of everything they do.

IBM also purchased Weather Underground and promptly proceeded to run that site and the WunderMap app into the ground while taking away historical gage data/functionality and putting it behind a pay wall.
Posted by aggiegeog
Tyler, TX
Member since Feb 2013
51 posts
Posted on 1/9/19 at 11:53 am to
A lot of the issues upper air data is generally only collect twice a day and the grid of that data collection is very coarse. And it is almost non existent over the oceans except from airplanes. Until we have better upper air data over the Pacific forecasting individual storm systems out more than a few days over the US is very problematic. Now large scale changes are much simpler to forecast at longer ranges. Such as the flip to a colder pattern for the E 2/3 of the US that is coming in a couple weeks has been forecasted since before Christmas. Individual storms are another matter as many are just a ripple over Asia a week out from impacting the southern US.
Posted by CivilTiger83
Member since Dec 2017
2525 posts
Posted on 1/9/19 at 12:06 pm to
Some more background on what they did to Weather Underground's WunderMap, which they are using for this analysis…

Once upon a time a couple years ago, Weather Underground set up a system so people who owned private weather stations could connect with to see their own data and others gage data and put it in WunderMap application for free. Great site for looking at historical rainfall data.

Fast forward a couple years ago and IBM buys Weather Underground. They proceed to strip the historical data from the site beyond a year, and start charging people for access to data and they don't even own the devices. Now they are trying to leverage it further for more profit. And they are throwing so many ads on the Weather Underground pages that it makes it glacial to navigate.
This post was edited on 1/9/19 at 12:08 pm
Posted by corym52
Member since Jul 2007
721 posts
Posted on 1/9/19 at 12:26 pm to
I thought the weather channel bought weather underground? Regardless, what they have done to the “storm” app is an embarrassment.
Posted by CivilTiger83
Member since Dec 2017
2525 posts
Posted on 1/12/19 at 9:40 pm to
quote:

I thought the weather channel bought weather underground? Regardless, what they have done to the “storm” app is an embarrassment.



IBM bought Weather Underground and Weather.com in 2015 (but not The Weather Channel).
first pageprev pagePage 2 of 2Next pagelast page
refresh

Back to top
logoFollow TigerDroppings for LSU Football News
Follow us on Twitter, Facebook and Instagram to get the latest updates on LSU Football and Recruiting.

FacebookTwitterInstagram