Len W
Feb 17, 2020, 4:51:20 PM2/17/20
to weathera...@googlegroups.com
Not sure of the benefits of such an expensive number crunching machine.
It seems it's all down to being able to forecast at 1 km resolution.
Keeping up with ECMWF who are already planning to use the American's most powerful in the world computer at Oak Ridge.
'Numerical weather prediction and climate models cover the globe in a
grid and simulate the evolution of the Earth system for each grid box.
The smaller the grid spacing, the more Earth system processes can be
explicitly simulated by the model.'
Yes, bags more number crunching to do. Solving second order, non linear partial differential equations.
But how about the physics? Can we parameterize at 1 km?
Surface energy transfer is horribly complicated at that scale.
Len PhD ( Boundary Layer Meteorology)
Jack Harrison
Feb 18, 2020, 2:57:23 AM2/18/20
to Weather and Climate
I can’t put it in quite the scientific language that Len uses but here is my take
Surface area of Earth 500 million sq.kms
Computer will use a grid of 1.5 x 1.5 kms = approx 2 sq.kms
Requires some 250 million input pieces of input data. How will this be achieved?
This massive number-crunching machine will only be as good as the input (nothing new in that concept)
Jack
Smartie
Feb 18, 2020, 3:47:46 AM2/18/20
to Weather and Climate
On Monday, February 17, 2020 at 9:51:20 PM UTC, Len W wrote:
Keeping up with ECMWF who are already planning to use the American's most powerful in the world computer at Oak Ridge.
ECMWF new HPC hardware is going to located in Italy.
But how about the physics? Can we parameterize at 1 km?Surface energy transfer is horribly complicated at that scale.
Effort already underway and has been for sometime, building on work started after Boscastle
I think this is a coup for Endersby getting the commitment to injvest in exascale HPC- machines and personnel.
Smartie
Feb 18, 2020, 3:57:00 AM2/18/20
to Weather and Climate
On Tuesday, February 18, 2020 at 7:57:23 AM UTC, Jack Harrison wrote:
I can’t put it in quite the scientific language that Len uses but here is my take
Surface area of Earth 500 million sq.kms
I don't think the idea is to have a 1.5 km global grid, probably something a bit coarser and then finer grid spacing (~1km) over regions eg UK and ~100m over local areas eg London. but it will depend how they can get the new models to scale on the hardware-
Computer will use a grid of 1.5 x 1.5 kms = approx 2 sq.kms
Requires some 250 million input pieces of input data. How will this be achieved?
This massive number-crunching machine will only be as good as the input (nothing new in that concept)
Jack
The key is the data assimilation system but the UM and EC-IFS have state of the science 4DVAR systems with the ECMWF beiing particularly successful. But new techniques are always needed. It's not widely appreciated that DA is more computationally expensive than running the actual NWP model.
Len
Feb 18, 2020, 4:57:43 AM2/18/20
to Weather and Climate
Yes Dave, I know about ECMWF's new ATOS computer.
but I was referring to research they were going to do with the Oak Ridge supercomputer in USA.
Of course you are right about the data assimilation process.
Hugely computationally expensive in the operational forecasts.
I get a bit confused about the finances involved in improving weather forecasting.
Presumably UKMO need £1.2 billion just because they operationally forecast.
ECMWF are set up to do joint European research.
Also costing UKMO as partners for a new ATOS computer,
Is this doubly expensive?
Better let Boris and his mate Cummings know about this.
After all Brexit means Brexit.
;-()
Len
Smartie
Feb 18, 2020, 5:27:21 AM2/18/20
to Weather and Climate
On Tuesday, February 18, 2020 at 9:57:43 AM UTC, Len wrote:
Yes Dave, I know about ECMWF's new ATOS computer.but I was referring to research they were going to do with the Oak Ridge supercomputer in USA.
Oh- thanks, I wasn't aware of this. Japanese have done this at something like 2-3km.
Of course you are right about the data assimilation process.Hugely computationally expensive in the operational forecasts.I get a bit confused about the finances involved in improving weather forecasting.Presumably UKMO need £1.2 billion just because they operationally forecast.
I believe the it's not just for the hardware but also infrastructure and running costs for 10 yrs NWP and climate work. Presumably also human resources- I'm told the LFRic/GungHo project is under-resourced at present...
~£1B may not turn out to be too much.
Jack Harrison
Feb 18, 2020, 11:15:37 AM2/18/20
to Weather and Climate
Better let Boris and his mate Cummings know about this. After all Brexit means Brexit.
Had you forgotten Len? We've taken back control. That European lot couldn't organise a proper flood.
Jack
Reply all
Reply to author
Forward
0 new messages