Opened 7 years ago

Closed 5 years ago

#1101 closed Defect (fixed)

Strange behaviour at very small timestep sizes

Reported by: hadcv Owned by: nemo
Priority: low Milestone:
Component: OCE Version: release-3.4
Severity: Keywords:
Cc:

Description

I am currently working on the C1D model configuration as part of the UKMO contribution to this year's work plan.

As part of this work, I have been using mooring data with a sampling frequency of ~2-10 min to force the 1D model. In doing so however, I have observed strange behaviour in the model results, where the inertia of the top model level seems to decrease as one approaches very small timesteps (that is, the acceleration of the top model level in response to the surface wind stress seems to be stronger for small timesteps).

The attached PDF illustrates this for 1D and 3D model results. The timestep and nn_fsbc parameters have been altered proportionally so that the SBC sampling frequency is always 45 minutes. The 1D model results are for the OS PAPA site, using OS PAPA mooring fluxes. Additionally, all 1D model runs (except those on page 3) have been run without the Coriolis contribution to the dynamics for simplicity.

Page 1 shows depth-time hovmuller diagrams of the zonal momentum trend in the 1D model. These illustrate that the impact on the model when reducing the timestep is mostly limited to the top model level.
Page 2 shows the impact of using a very small timestep in the 1D model. In this configuration, the surface zonal momentum trend becomes 'noisy' and acts to increase the vertical mixing via an increase in the mean shear.
Page 3 shows the impact of several timestep values (with nn_fsbc varied accordingly) on different model variables, illustrating the impact of the increase in vertical mixing with reduction in timestep. These runs include the Coriolis contribution to the momentum trend.
Page 4 shows results for the same two parameter combinations presented on page 2, but here for the full 3D model (using CORE2 fluxes). There is a small impact on the surface current which appears to act to restratify the summer mixed layer in the Southern Ocean, whilst having additional impacts elsewhere globally. Generally the response seems to be a reduction in vertical mixing, in contrast to the 1D model results.

This seems to suggest a problem with the tridiagonal solver in dynzdf.F90 at small time steps, but this goes beyond my mathematical knowledge.

Commit History (0)

(No commits)

Attachments (2)

Illustration_of_the_timestep_issue.pdf (1.3 MB) - added by hadcv 7 years ago.
timestep_nodyncor_TINT_sokaraml.png (109.8 KB) - added by hadcv 7 years ago.

Download all attachments as: .zip

Change History (4)

Changed 7 years ago by hadcv

comment:1 in reply to: ↑ description Changed 7 years ago by dupontf

I am not clear that dynzdf would be the culprit. I would rather speculate that the time-interpolation of your forcing is the issue (and maybe the turbulence).

I would try different sensitivity experiments with C1D:

-first run with no turbulence scheme (key_zdfcst) and a fixed value of avm0/avt0 (say 1e-2)
-change your time-interpolation option. Right now there appears to be none, which means that at each new data coming in (every 45min) and then held constant for rdt=60s. A shock is then created in the model. At rdt=2700s, each new data comes at the right time and no interpolation is needed. So try instead to smooth your input by choosing to time-interpolate your forcing for the case rdt=60s.

Changed 7 years ago by hadcv

comment:2 Changed 5 years ago by clevy

  • Resolution set to fixed
  • Status changed from new to closed
Note: See TracTickets for help on using tickets.