Welcome, Guest
Username: Password: Remember me
HELPDESK

Here we can describe more what should be posted here

TOPIC: Noise in MSLP fields

Noise in MSLP fields 4 months 15 hours ago #2251

  • Colm Clancy
  • Colm Clancy's Avatar
  • OFFLINE
  • Fresh Boarder
  • Posts: 6
  • Thank you received: 2
Hello all,

In some of our recent operational output (cycle 40h1.1++), I've noticed some noise 'features' appearing in the MSLP fields. An example below:


IREPS-atl-MSLP-2018120606031-mbr000.png


The noisey patch appeared for a number of hours then disappeared, i.e. no model crash. In all cycles it was appearing around the same time, same place, and has since happened in other situations

A cross-section of vertical velocity through this shows oscillations at the top of the domain:

w_sectionWE_201812060631.png


The zonal wind:

u_sectionWE_201812060631.png



Conveniently, this behaviour has proved to be reproducible in experiments. It can be removed by reducing the time-step from 75s to 60s, or by using a quadratic or cubic grid.
Experiments with things that sound like they might affect upper behaviour (LUNBC, LCOMADV, LNSPONGE) haven't been successful in removing the noise.

I was curious if others see these kind of patterns?

Noise in MSLP fields 3 months 4 weeks ago #2252

Hi,

This is not something new. In our operational configurations, we have decided not to risk and stick to ts<=60s quite a while ago, after having suffered similar effects at 2.5 km resolution (more than a few times, but those ussualy do occur during important situations), including the (usually) following operational model crashes.
On the other hand, that might be quite painful in terms of the targeted wall time (in case you had any) in an EPS configuration, and if that is the case, perhaps you might want to explore the grid_type options if those help at ts=75s further.

Noise in MSLP fields 3 months 4 weeks ago #2253

  • Colm Clancy
  • Colm Clancy's Avatar
  • OFFLINE
  • Fresh Boarder
  • Posts: 6
  • Thank you received: 2
Thanks for that.

Interestingly, our only operational crash in recent years was with cycle 37 at tstep=60s, during a major storm. And I have been unable to reproduce the crash with cycle 40, even at 75s.

Noise in MSLP fields 3 months 4 days ago #2259

  • Sander Tijm
  • Sander Tijm's Avatar
  • OFFLINE
  • Administrator
  • Posts: 24
  • Thank you received: 2
Dear Colm and others,

one thing that I just remembered was the predictor - corrector option. If this does not happen with the PC-scheme, then the PC-on-demand, the work of Jozef Vivoda, may be an option to pursue? It should be available in Cy43 or Cy45?

Noise in MSLP fields 2 months 1 week ago #2277

  • Colm Clancy
  • Colm Clancy's Avatar
  • OFFLINE
  • Fresh Boarder
  • Posts: 6
  • Thank you received: 2
An update on this.

Brute-force approaches to removing this noise include reducing the timestep, switching to quadratic grid, increased spectral diffusion, and off-centring of the nonlinear terms. Also, a predictor-corrector with SLHD instead of COMAD seems to work.

However, it can also be improved with a simpler namelist change: setting (LGWADV, LRDBBC) = (TRUE, FALSE). These are the values generally used with the predictor-corrector, but by default have the opposite values in our HARMONIE set-up.
Attachments:

Noise in MSLP fields 2 months 1 week ago #2278

  • Colm Clancy
  • Colm Clancy's Avatar
  • OFFLINE
  • Fresh Boarder
  • Posts: 6
  • Thank you received: 2
I'm curious: does anyone else have experience with LGWADV and LRDBBC in a SETTLS set-up (i.e. not predictor-corrector) ?
Time to create page: 0.294 seconds