Over the last couple of weeks, I was able to successfully add a dye_01 variable to LiveOcean and conduct a dye release test originating from West Point WWTP. More details below.

Dye release results

Adding a dye_01 variable required interventions to the ocean forcing files, the river forcing files, and the dot in file. ROMS already had a “dye_” variable in the varinfo.yaml file, so all I really needed to do was be consistent with the ROMS naming convention, and the model handled everything else automatically. I imagine that adding additional BGC variables will require more intervention in the code.

Results from a one-day West Point dye release experiment are below.


Fig. 1 Dye release test from West Point WWTP. Every frame corresponds to an hourly snapshot, and the video spans one day. Discharge concentration was 10 kg/m3 of dye.


I used intentionally small colorbar limits to see whether noise would appear in Sinclair Inlet. It appears as though inert passice tracers do not have noise issues.

One last noise test?

In the back of my mind, I am still curious about Alex’s question of how the twin variables would solve our BGC noise issue. I propose one final noise test to probe into whether the twin variables is likely to solve our problem:

Let’s make NH4 zero everywhere in the model domain and let the only source of NH4 be West Point WWTP. This experiment is similar to the dye release test above, except that NH4 is a nonconservative tracer.

I envision two possible outcomes to this experiment:

  1. My hypothesis is that noise is generated by the nonconservative nature of the BGC variables. If this is true, then I expect noise to appear in this final NH4 test. (Note that since NH4 will be zero everywhere but West Point, I do not need to conduct a differencing experiment to see NH4 noise).

  2. If my hypothesis proves incorrect and we do not see noise in this test case, then I suspect that noise only appears if there is other NH4 present in the model, and our differencing experiments somehow perturb what is already present. In other words, BGC noise perturbations can only occur if there is something there to perturb in the first place.

It occurs to me that regardless of the outcome of this experiment, we will still come to the conclusion that there is a noise issue in the BGC fields of LiveOcean. Perhaps the noise will be equivalent in two twin variable test conditions such that the noise subtracts out to zero. However, this would still mean that there is inherent noise in the BGC fields of the model, which could contribute to model uncertainty. Is this something to be concerned about?