This blog post addresses an ongoing issue in which we are observe bit-reproducibility issues in the biogeochemistry fields of my loading experiments.

Summary of prior knowledge: I have two runs that I am comparing: Natural and Anthropogenic. They are identical except the Natural run has no NH4 or NO3 discharging from WWTPs, and the Anthropogenic run has realistic nutrient loads. The salinity, temperature, and hydrodynamic fields are identical between these two runs. However, I am observing noise in the biogeochemistry fields that develops along the northern half of the western boundary of the model domain.

New work: This week, per Jilian and Parker’s advice, I tried to quantify the size of the noise in the DO field relative to the DO signal in Puget Sound. Note that these results are only for the year 2013, which is still a spin-up year. However, 2013 is when it is easiest to distinguish between the signal and noise (because they haven’t propagated towards each other yet).

In summary, the background noise is generally several orders of magnitude smaller than the signal in Puget Sound. However, it is nonzero.

After two weeks

After two weeks of run time, we already begin to see bit-reproducibility issues along the northwest corner of the model domain. Figure 1a shows a map of surface differences between the Natural and Anthropogenic test cases. Figures 1b and 1c show boxplots of the differences over the whole domain, over Puget Sound, over the Salish Sea, and over the Western Boundary. Note that outliers are not shown so that we can see the boxplots more clearly. The boxplot associated with the full domain appears to have a smaller range of data compared to Puget Sound and the Salish Sea solely because the outliers are not being drawn. To clearly list the entire range of data, I have written all meaningful values in Table 1.

Note that I have omitted zeros from these analyses so they do not not bias the analyses towards zero. There are a lot of placces where Natural = Anthropogenic, and we are only interested in places where they are different.


Fig 1. Natural minus Anthropogenic DO after two weeks. (a) shows a map of the surface differences. (b) shows boxplots of surface differences. (c) shows boxplots of bottom differences. The navy dots are mean values.



Table 1. Values of Natural minus Anthropogenic DO [mg/L] in the different domains after two weeks.


The mean and median values of the background noise are much much smaller than those of in the Salish Sea and in Puget Sound.

After two months

Figure 2 and Table 2 show the Natural minus Anthropogenic differences after two months. It is around this time where we begin to see the real signal propagate towards the noise, and it become harder to distinguish the two from each other. We also see boundary noise appear further south along the western boundary.


Fig 2. Natural minus Anthropogenic DO after two months. (a) shows a map of the surface differences. (b) shows boxplots of surface differences. (c) shows boxplots of bottom differences. The navy dots are mean values.



Table 2. Values of Natural minus Anthropogenic DO [mg/L] in the different domains after two months.


Again, it still seems like the DO difference in Puget Sound and the Salish Sea are much larger than those along the western boundary.

On September 1

Finally, I took a look at the Natural minus Anthropogenic differences on September 1, when we are most interested in DO differences. Results are presented in Figure 3 and Table 3. Note that there also appears to be some noise along the southern boundary as well.

Once again, the differences in the western boundary region are generally much smaller than in Puget Sound and the Salish Sea.


Fig 3. Natural minus Anthropogenic DO after eight months. (a) shows a map of the surface differences. (b) shows boxplots of surface differences. (c) shows boxplots of bottom differences. The navy dots are mean values.



Table 3. Values of Natural minus Anthropogenic DO [mg/L] in the different domains after eight months.