Now that I have finally passed quals (yay!), I have turned my attention back towards my PhD research. This blog post summarizes my current thoughts on Chapter 2.

There are five main considerations that we need to think through before setting up a loading/no-loading experiment:

  1. What version of LiveOcean to use? The old biogeochemistry module which overestimated DO? A newer version?
  2. Bit-reproducibility issues manifesting as noise in the biogeochemistry fields
  3. Shortening model run time by outputting daily averages rather than hourly data
  4. New loading data from Ecology and USGS for point sources
  5. What are our test conditions?

1. Version of LiveOcean

Depending on how we decide to handle issues 2 - 4 below, we may need to run a new “loading” run for the Chapter 2 experiment (rather than using the hindcast data I used for Chapter 1).

If we need to run a new “loading” run, then we have the flexibility to consider an improved version of the biogeochemistry module.

2. Bit-reproducibility

A few months ago, Jilian was actively investigating an issue where we saw bit-reproducibility manifest in the biogeochemistry fields as noise– particularly along the boundaries of the model domain.

One concern for my Chapter 2 experiment is that this noise can propagate into Puget Sound, creating artificial differences in DO between the loading and no-loading runs. However, when I previously tried to quantify the noise in the DO field, it appeared to be relatively small compared to the DO signals we are interested in.

In any case, it is still worth considering a solution to this bit-reproducibility issue. Jilian suggested an improvement that worked for her: increasing the strength of “nudging to climatology” for the biogeochemistry fields along the domain boundary. This improvement requires that I re-run both the “loading” and “no-loading” conditions.

3. Daily average outputs

Another concern with the Chapter 2 experiment is the duration of run time. Will it take six months before having results? Is there any way to speed things up?

Recently, Jilian suggested that I recompile ROMS with an option to output daily average values, rather than outputting hourly data. Reducing the amount of data that LiveOcean needs to output will speed up the simulation process.

I was originally very excited for this idea. However, I would like to compare the fluxes of nutrients, detritus and DO into the terminal inlets between the loading/no-loading runs. The way I calculated DO fluxes in Chapter 1 was using the TEF code, which requires hourly model output. Therefore, I think we necessarily need to output hourly data rather than daily average data.

I’m happy to chat more about this, though. It would be great to know if we could still make daily averages work for us!

4. New WWTP loading data

New WWTP loading data

As I mentioned previously, Ecology and the USGS have published new monthly loading data for WWTPs from 2005 - 2020 (links below). These WWTP loading data are used as inputs to a SPARROW model– a joint effort between Ecology and the USGS to estimate the seasonal nutrient loads discharging from 19 different basins across Puget Sound. The SPARROW model takes into account an assortment of upstream nutrient sources such as animal feed operations, septic tanks, Red Alder nitrogen fixation, etc.

For our purposes, I am only considering incorporating the new WWTP loading data into LiveOcean. I intend to leave the tiny river climatologies as-is because the outputs of the SPARROW model are resolved at a seasonal time scale rather than a daily time scale. However, I’m happy to chat more about all of our options in more detail.

Why upgrade the LiveOcean WWTP loading?

The old TRAPS climatologies were generated from an older dataset that spanned from 1999 through July of 2017. One reason why I generated climatologies was to expand the dataset beyond July 2017, so we could use it as model input for these later years.

However, Stefano has previously mentioned to me that some WWTPs have been upgraded over the years, and their loads may have changed abruptly. Thus, climatologies may not be the most appropriate input to use for our loading/no-loading experiment. I’m curious to look into the WWTPs with changing loads in more detail and compare our climatologies to the new data for these specific plants. Stefano is out of office currently, so we’ll wait to learn more once he’s back (I don’t remember which WWTPs had changing loads).

As a first look, Figure 1 shows a comparison of the West Point nutrient load climatology compared to the new nutrient load dataset.


Fig 1. West Point WWTP discharge. New Ecology/USGS loading data compared to the climatologies currently being used in LiveOcean.


Implementation thoughts

It’s going to take me at least one week of full-time focus to integrate the new data into the LO system. Some challenges I anticipate include:

  • Restructuring my TRAPS code to handle and process the new data format and metadata
  • Identifying which new WWTPs correspond to which old WWTPs (because the names aren’t the same)
  • Deciding how to deal with loading inputs after 2020
  • Differentiating between WWTPs and industrial facilities and fish hatcheries
    • Do industrial facilities and fish hatcheries also discharge from the sea floor?
    • We only want to modify WWTP loads, not the loads of industrial facilities and fish hatcheries. Thus, I need some way of tracking these different point-source types.

5. Test conditions

Lastly, we need to decide on what test conditions to run for the “loading” and “no-loading” scenarios.

My original plan was to run both simulations with identical hydrodynamics, and simply reduce the WWTP nutrient concentrations to zero. Let’s revisit this idea.

In chatting with Mike a few weeks ago, he expressed interest in participating in the experiment design plan. He mentioned that it would be good idea to include Ben in this discussion as well. Perhaps if we plan together as a group, we can devise experiment plans that are more conducive to future inter-model comparisons.

Another thought I had after looking at the new WWTP loading data (Fig 1): Are we interested in analyzing a loading/no-loading experiment for only one year? There is interannual variability in the loading, so do we simply pick a year? Or perhaps we stick with climatologies and examine a “typical” WWTP load.