I have made a little bit of progress on a few research tasks this week. See below for more details.

Long Hindcast

One of my primary research tasks is to debug the long hindcast. I undefined OMEGA_IMPLICIT in the ROMS header file, recompiled, and re-ran the model starting on 2014.06.13, which is five days prior to the dreaded blow-up date.

Despite re-launching several days in advance, the model still blows up on hour 19 of 2014.06.18. I do not have figures to show this week, but next week I’m planning to plot some vertical profiles that will help us understand the source of the blow-up issue.

Noise

As a follow-up to my last blog post on noise evolution, I re-made the noise videos given Alex and Parker’s feedback.

In the new iteration of noise videos, I have plotted:

  • The difference between the loading and no-loading run
  • of the vertical integrals of total nitrogen (TN)
    • Note: these are no longer volume integrated– simply vertically integrated
  • Normalized by the same value in every frame of the video
    • This value is the maximum value of the difference in vertically integrated TN observed over the course of a year


Fig. 1 Evolution of normalized vertically-integrated TN differences at a weekly interval for one year. Colorbar limits of the pcolormesh range from -10% to 10%.


Near the boundaries of the model domain, I still observe noise on the order of 10% of the max signal (Fig 1). Again, this might be problematic for future OAE experiments.

Over the course of a year in the Salish Sea, I do not observe any noise on the order of 10% of the max signal (Fig 2).


Fig. 2 Evolution of normalized vertically-integrated TN differences at a weekly interval for one year. Colorbar limits of the pcolormesh range from -10% to 10%.


During the first two weeks after initializing the model run, I observe some noise on the order of 0.1% in Puget Sound. By the end of the two-week period, the signal (red) appears to dominate over the noise.


Fig. 3 Evolution of normalized vertically-integrated TN differences at a daily interval for two weeks. Colorbar limits of the pcolormesh range from -0.1% to 0.1%.


It’s difficult to evaluate the size of the noise in Puget Sound since the signal dominates over the noise after two weeks of run time. The remaining questions are: How much is this signal “off” due to noise? If the noise remains on the order of 0.1% of the max signal, is that acceptable?

Side note: I could not detect noise on the order of 1% of the max signal in Puget Sound (video not shown). Thus, I think that noise is closer to 0.1% of our max signal.

OAE module test

As a bonus figure, I’ve included a recent video from my OAE module testing in LiveOcean. LiveOcean runs without throwing an error when I activate the new OAE module, but very quickly the carbonate chemistry fields are swallowed by a wave of nans coming from the Northeast corner of the model domain (Fig 4).


Fig. 4 Surface alkalinity and DIC in LiveOcean + OAE module. The video spans one day of run time at an hourly interval. No additional alkalinity was added to the domain (alkalinity concentrations are ambient values).


Despite this problematic behavior, I think we are okay to check off the OAE module integration task for now as we were able to run the OAE module without any blow-ups or other errors.