Cauchy-Poisson wave patterns in velocity increment W4DVAR

Discussion about tangent linear and adjoint models, variational data assimilation, and other related issues.

Moderators: arango, robertson

Post Reply
Message
Author
User avatar
pmuskee
Posts: 6
Joined: Mon May 02, 2016 12:54 pm
Location: Ann Arbor, MI

Cauchy-Poisson wave patterns in velocity increment W4DVAR

#1 Unread post by pmuskee »

ROMS Post

I am running the W4DVAR with strong constraint (NADJ=NTIMES) on an 800m resolution grid (40 vertical levels) with initial and boundary conditions taken from 3.7km regional NCOM (USEAST). The surface forcings are taken from a WRF model that is NOT coupled to the ROMS model. I am assimilating ship board ADCP profiles and CTD casts from a research vessel. My header, ocean.in and s4dvar.in files are in the attached zip file. I am getting these “rock in the pond” (Cauchy-Poisson problem) patterns in my velocity increment fields. They seem to be generated at the bottom model level and then propagate all the way to the surface. What I have discovered so far is that this issue does not show up if I use the incremental strong constraint (IS4DVAR). I have also tried smoothing the hell out of my bathymetry and changing my drag law specifications : no improvement. Has anyone else seen this problem before with the W4DVAR? Am I missing something obvious or is there something inherent to the dual form indirect representer based method that could cause this issue. Also I am running the 4DVAR with 1 outer loop and 40 inner loops and the cost function plot shows that it seems to have converged. Lastly, I am including a plot of the background/prior fields to show that these features are only present in the analysis.

Thank you for your help in advance and let me know if you have any questions about what I am doing.
Attachments
Increment velocities at model level 40 (bottom) for time 1 hr, 12 hr, and 24 hr.
Increment velocities at model level 40 (bottom) for time 1 hr, 12 hr, and 24 hr.
Increment velocities at model level 35 for time 1 hr, 12 hr, 24 hr, and 48 hr.
Increment velocities at model level 35 for time 1 hr, 12 hr, 24 hr, and 48 hr.
increment_lvl35.png (633.59 KiB) Viewed 8543 times
Background velocities at model level 40 for time 1 hr, 24 hr and 48 hr.
Background velocities at model level 40 for time 1 hr, 24 hr and 48 hr.
R4DVAR cost-function for one outer loop and 40 inner loops.
R4DVAR cost-function for one outer loop and 40 inner loops.
parameter_files.zip
header file, ocean.in, and s4dvar.in
(36.3 KiB) Downloaded 254 times

User avatar
pmuskee
Posts: 6
Joined: Mon May 02, 2016 12:54 pm
Location: Ann Arbor, MI

Re: Cauchy-Poisson wave patterns in velocity increment W4DVA

#2 Unread post by pmuskee »

So I think my problem was that I had the number of iterations in the biconjugate gradient algorithm used to solve the elliptic equation for the balance operator set too low for my domain. Initially I was using Nico = 200 but now I have it set as Nico = 3000. This causes my problem to disappear. I have a sense that the model grid resolution is the major reason that I need more iterations.

Post Reply