Jump to content

It's finally happened: NWS forecasting computers to become 10 times more powerful this year


Recommended Posts

Weather Service forecasting computers to become 10 times more powerful in 2015

 

The National Weather Service’s primary computer model trails competitors in Europe in overall forecasting accuracy.  But today it announced upgrades to its supercomputers that hold great promise to improve its predictions.

 

By October this year, the capacity of the two National Weather Service (NWS) supercomputers will increase by nearly a factor of ten it said.

 

“By increasing our overall capacity, we’ll be able to process quadrillions of calculations per second that all feed into our forecasts and predictions,” said Louis Uccellini, director of the National Weather Service in a press release. “This boost in processing power is essential as we work to improve our numerical prediction models for more accurate and consistent forecasts required to build a Weather Ready Nation.”

 

The anticipated ramp-up in computing power results from a $44.5 million investment in high performance computer with IBM, $25 million of which was supplied by Congress following Superstorm Sandy.

 

These computing upgrades were originally scheduled to occur in 2014, but IBM – its contractor -sold its supercomputing services to Lenovo, a Chinese company.  This raised red flags in Congress, according to reports, and put computer acquisitions on hold.

 

“In late January 2014, NWS was literally weeks away from executing a task order under its current contract with IBM that would have significantly increased its computing capacity and allowed the NWS to meet the goals communicated publicly in 2013,” said a NWS white paper published in November last year. “However, IBM’s decision to leave the x86 server business by selling the line to Lenovo caught NOAA, and everyone else, by surprise and prevented us from moving forward with this procurement.”

 

NWS was forced to start a procurement process from scratch early in 2014 and announced today that  Cray Inc., headquartered in Seattle, WA, will serve as the subcontractor for IBM to provide the next round of computer upgrades.

 

“We are excited to provide NOAA’s National Weather Service with advanced supercomputing capabilities for running operational weather forecasts with greater detail and precision,” said Peter Ungaro, president and CEO of Cray in a press release. “This investment to increase their supercomputing capacity will allow the National Weather Service to both augment current capabilities and run more advanced models.”

 

These planned upgrades follow a three-fold boost in computing power in the works for later this month that will allow NWS to run an enhanced version of its Global Forecast System (GFS) model, with greater resolution deeper into the future.

 

This improvement follows a substantial upgrade to the NWS hurricane model (known as the HWRF), and the operational launch of a high resolution model for forecasting thunderstorms (known as the HRRR) in 2014.

 

Despite these investments, whether NWS will be able to catch up to European counterparts in an ongoing supercomputing arms race is an open question.  The European Centre for Medium-range Weather Forecasting (ECMWF) and United Kingdom Met Office recently invested $64 million and $128 million, respectively, in computer upgrades (although a sizable chunk of UKMet Office investment is for climate modeling, rather than weather modeling).

 

“The US computer capacity will not surpass the European infrastructure without additional investment,” the November NWS white paper said.

 

http://www.washingtonpost.com/blogs/capital-weather-gang/wp/2015/01/05/weather-service-forecasting-computers-to-become-10-times-more-powerful-in-2015/

 

I don't think it's just the weather geek in me that makes me think it's pretty obviously a great investment. Accurate weather predictions save thousands of lives and billions of dollars every year. We take it for granted, but every improvement is worth a lot.

Everett Snowfall (510 feet elevation)

Snow since February 2019: 91"

2023-24: 6"

2022-23: 17.5"

2021-22: 17.75"

2020-21: 14.5”

2019-20: 10.5"

2018-19: 24.75"

 

 

 

Link to comment
Share on other sites

  • Replies 4
  • Created
  • Last Reply

Top Posters In This Topic

Popular Days

Top Posters In This Topic

Cliff Mass has been talking about this for years now and he is understandably quite stoked.

 

A Major Advance for Numerical Weather Prediction in the U.S.
 
This is important. 

Today, NOAA's administrator Kathryn Sullivan announced that the National Weather Service will acquire two very powerful CRAY supercomputers to support U.S. numerical weather prediction.   A machine that will FINALLY allow the U.S. to do world-class forecasting.  (The press release is here).


In a number of my previous blogs, I complained about the inferior computing resources available for numerical weather prediction in the National Weather Service (NWS).  I noted that groups such as the European Center for Medium Range Weather Forecasting (ECMWF) and the UKMET office have had substantially more skillful global predictions and that their superior computer resources gave them a substantial edge.  I have also described how the U.S. has lacked high-resolution forecast ensembles (where the model is run many times), which undermined the NWS ability to predict severe thunderstorms and other small-scale features.

Everything can change now, IF the National Weather Service uses these powerful new computers wisely.

So here is why you should be excited.

Numerical weather prediction is the underlying technology of weather prediction and it dependent on computer power.

More computer power allows more resolution:  the ability to simulate smaller-scale weather features (like thunderstorms or mountain precipitation)

More computer power allows better physics, which includes the description of how clouds form or the turbulent processes in the lower atmosphere (among others).

More computer power allows better data assimilation, the use of observations to create a physically consistent three-dimensional description of the atmosphere.

More computer power facilitates ensemble forecasting, in which running the model many times allows forecasting the probabilities of weather events.


The Details

Currently, the National Weather Service has two computers (one is for backup and research), each with a throughput of roughly .21 petaflops (quadrillion operations per second).   The European Center for Medium Range Weather Forecasting has two computer that run at roughly 2.5 petaflops.   In other words, they have ten times more computer power to do FAR LESS than the U.S. NWS (the European Center does not do high resolution local prediction, the NWS does).   And it shows.

This month the National Weather Service will upgrade to two machines of .75 petaflops.

The new machines announced today will be 2.5 petaflops each, roughly TEN TIMES what we have today.

A major part of the announcement is that the new computers will be provided by CRAY, the leading supercomputer maker for numerical weather prediction (they are supplying the machines to the European Center and the UKMET office, among others).   IBM, who has held the computing contract with NOAA, could not supply the machines, because they sold their server division to a Chinese firm, Lenovo.  Thus, they subcontracted the acquisition to Seattle-based CRAY.

I have learned that the new machines will be CRAY XC-40 systems.    I suspect they will each have 50,000-100,000 processors.

The Potential

Properly used, this new computer power can revolutionize and greatly improve the skill of U.S. numerical weather prediction, with huge positive impacts for the country.  Problems in U.S. numerical weather prediction have not been limited to lack of computer power and these problems need to be addressed, such as an inability to entrain the huge knowledge based in the huge U.S. academic research community or inefficiencies/duplication of effort in U.S. government research and development.  These are issues that must be taken on now.   But the excuse of lack of computer power is gone and a renaissance in U.S. NWP is possible. 

Finally, it is important to acknowledge the leadership of NOAA's Kathryn Sullivan and NWS Director Louis Uccellini in making this happen.  The environment in NOAA seems to changing in a positive way and they deserve some credit for it.

 

http://cliffmass.blogspot.com/2015/01/a-major-advance-for-numerical-weather.html

Everett Snowfall (510 feet elevation)

Snow since February 2019: 91"

2023-24: 6"

2022-23: 17.5"

2021-22: 17.75"

2020-21: 14.5”

2019-20: 10.5"

2018-19: 24.75"

 

 

 

Link to comment
Share on other sites

So now the question is, will they go forward with a proportionally better GFS model? The upgraded GFS was for the tripling of computer power coming this month. So if we end up multiplying it by 10, will we see an even better model at the years end? Hope so!

Winter 23-24: Total Snow (3.2")    Total Ice (0.2")     Coldest Low: 1F     Coldest High: 5F

Snow Events: 0.1" Jan 5th, 0.2" Jan 9th, 1.6" Jan 14, 0.2" (ice) Jan 22, 1.3" Feb 12

Link to comment
Share on other sites

It should allow for higher resolution. I'm not sure they need to run the model every 6 hours though. The consensus on here seems to be that the 18z is a POS.

There's no statistically significant variance in model performance by cycle. The only issue is that $$$ used to run the model 4x a day could be put to better use under a 2-cycle system.

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Unfortunately, your content contains terms that we do not allow. Please edit your content to remove the highlighted words below.
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Loading...

×
×
  • Create New...