Sed and coastal/clean code #89

charliestock · 2024-08-04T03:43:23Z

This pull request cleans up and adds commenting to sections of the sediment code. It also modifies the formulation of a previously used "coastal" iron source (jfe_coast). This source of iron was used in TOPAZ and early versions of COBALT to simulate coastal iron fluxes from the vertical land face. This was done because coarse resolution often prevented significant sediment iron fluxes from being introduced near the ocean surface in shallow waters: the grid was so coarse, even the first grid cell was far deeper than the euphotic zone in most places. Releasing iron from the land edges provided a means of generating "island wakes" even when the model resolution prevented resolution of shallow nearshore areas.

As the resolution of our ocean model configurations has increased, we have moved away from using this "coastal" source, instead relying on the benthic fluxes that are explicitly resolved by the model. This has worked for nearly all of our CEFI runs. Recent Pacific Island simulations done by @gabyneg, however, suggest that there still may be a role for this parameterized flux in areas with particularly narrow shelves and steep bathymetry. However, when I looked back at the old way the flux was formulated I found the underlying rationale lacking. I thus coded an alternative that made the coastal flux proportional to the benthic flux that would have arisen from the organic carbon flux and O2 in the adjacent water. Conceptually, one can think of this as a net flux that would have arisen from unresolved shallow regions within the first grid cell. The parameter "fe_coast" becomes a constant of proportionality which we need to explore but I suspect should be ~0.1 for cases where we feel it is necessary to use this parameterization.

The default value of fe_coast is 0 and I feel it should stay this way. With fe_coast = 0, this pull request should not change answers. In addition to the code review and commenting, it would be great if @gabyneg could run a few test cases in her Pacific Islands domain to see if we are satisfied with the new approach.

I still have a bit more work to do with the commenting of the sediment section, but I will complete this in a subsequent pull request.

… flux with dzt

charliestock · 2024-08-05T18:15:37Z

Hello @yichengt900, I made the small adjustments we discussed this morning. As expected, it did not seem to fix the cryptic restart issue blocking the pull request. Looking at the code changes, I can't imagine what might be tripping this off. When you get a chance, could you have a look at the restarts and see if they provide any clues?

yichengt900 · 2024-08-05T18:38:30Z

Hi @charliestock,

I reviewed the differences in the restart files, and the variables showing discrepancies after the model integrated the 3600s time step (the model ran with restart for an hour integration) are listed below:

nccmp -dfqS RESTART/MOM.res.nc RESTART_25hrs/MOM.res.nc
Variable Group Count          Sum      AbsSum         Min          Max       Range         Mean      StdDev
htotal   /        40 -4.40209e-19 4.40209e-19 -1.1162e-20 -1.08261e-20 3.35835e-22 -1.10052e-20 9.42072e-23
co3_ion  /        40  4.17849e-15 4.17849e-15 1.02782e-16  1.05723e-16  2.9409e-18  1.04462e-16 7.98836e-19
cased    /         8  1.36514e-14 1.36514e-14 1.70643e-15  1.70643e-15           0  1.70643e-15           0
dic      /        40  8.87918e-15 8.87918e-15  2.2031e-16  2.23346e-16 3.03577e-18   2.2198e-16 8.47963e-19
alk      /        40   1.7754e-14  1.7754e-14 4.42354e-16  4.45824e-16 3.46945e-18  4.43851e-16 7.54041e-19

I will investigate further.

yichengt900

Hi @charliestock,

I think I found the issue that is breaking the reproducibility across restarts.

Please see my comment below.

generic_tracers/generic_COBALT.F90

charliestock · 2024-08-05T21:59:40Z

That seemed to do the trick YC, thanks!

yichengt900

@charliestock , this PR looks good after the fix, I just have a few small comments.

yichengt900 · 2024-08-05T23:17:13Z

generic_tracers/generic_COBALT.F90

    enddo; enddo; enddo  !} i,j,k

+    ! Calculate the bottom conditions and the fluxes to the bottom for diagnostics and benthic flux calculations.
+    ! MOM4/5 used the bottom grid cell, but MOM6 often has a number of vanishingly thin layers overlying the bottom.
+    ! Grid scale noise in these layers can occur, particularly for quantitities with large bottom fluxes.  COBALT thus


Should be quantities

yichengt900 · 2024-08-05T23:20:53Z

generic_tracers/generic_COBALT.F90

@@ -4313,9 +4321,40 @@ subroutine generic_COBALT_update_from_source(tracer_list,Temp,Salt,rho_dzt,dzt,h

          ! Iron from sediment (Dale, 2015).  The maximum release from the sediment is set by ffe_sed_max.  The
          ! hyperbolic tangent requires the flux of carbon to the sediments (as mmoles m-2 day-1) in the numerator
-          ! and the bottom water oxygen concentration (in microMolar units) in the denominator:
+          ! and the bottom water oxygen concentration (in microMolar units) in the denominator. Note that ffe_sed_max
+          ! was converted to moles Fe m-2 sec-1 during parameter input, so ffe_sed has is in moles Fe m-2 sec-1


has is should be is

yichengt900 · 2024-08-05T23:22:02Z

generic_tracers/generic_COBALT.F90

+          ! the vertical face of the land mass has thus been included.  The flux is posed as a fraction (fe_coast) of
+          ! the sediment Fe flux (moles Fe m-2 sec-1) that would have result from the sinking organic matter flux and
+          ! O2 level of the adjacent waters.  This then spread across the layer mass (rho_dzt(i,j,k)) to give an input
+          ! in moles Fe kg-1 sec-1. Conceptually, this can be though of as a net iron flux resulting from the fraction


yichengt900 · 2024-08-05T23:42:50Z

generic_tracers/generic_COBALT.F90

@@ -4249,6 +4256,7 @@ subroutine generic_COBALT_update_from_source(tracer_list,Temp,Salt,rho_dzt,dzt,h
             ! Since burial is highly uncertain and often used in global earth system simulations to balance inputs and
             ! outputs, a dimensionless scaling factor (cobalt%scale_burial) has also been included.
             fpoc_btm = cobalt%fntot_btm(i,j)*cobalt%c_2_n*sperd*1000.0
+             ! Should we use ztop?


Hmm. This depends on how we define our benthic layer. If we consider the model bottom layer as the benthic layer, then using ztop would be appropriate. However, if we define the benthic layer as one layer below our model's bottom grid, then cobalt%zt would be suitable.

I know I have looked at this directly before, and maybe I'm just thinking myself in a circle, but am I right in thinking that grid_kmt is equal to the number of vertical tracer grid cells (e.g., 75 for most of our regional applications) in MOM6. So "zt(75)" is the bottom of the 75th grid cell (i.e., the sediment)? I just need a quick sanity check on this.

gabyneg · 2024-08-06T14:49:45Z

generic_tracers/generic_COBALT.F90

+          ! Coarse resolution models and/or intermediate resolution models in areas with exceptionally steep bathymetry
+          ! can under-represent coastal iron because they don't resolve shallow regions. An option to add iron through
+          ! the vertical face of the land mass has thus been included.  The flux is posed as a fraction (fe_coast) of
+          ! the sediment Fe flux (moles Fe m-2 sec-1) that would have result from the sinking organic matter flux and


change "result" to "resulted" or remove "have" in this sentence "the sediment Fe flux (moles Fe m-2 sec-1) that would have result from the sinking organic matter flux and O2 level of the adjacent waters."

Got it. Thanks.

gabyneg · 2024-08-06T14:50:11Z

generic_tracers/generic_COBALT.F90

+                   "nitrate release per kg of ice melt",units="mol N kg-1 ice melt", default= 2.0e-6)
+    call get_param(param_file, "generic_COBALT", "jpo4_iceberg_ratio", cobalt%jpo4_iceberg_ratio, &
+                   "phosphate release per kg of ice melt",units="mol P kg-1 ice melt", default= 1.1e-7)
+    ! fe_coast is effectively the fraction of the equivalent benthic iron flux that you would have been generated by


Remove "you" after "... flux that" in "fe_coast is effectively the fraction of the equivalent benthic iron flux that you would have been generated by", and change "were it to encounter sediments" to "when encountering sediments".

gabyneg · 2024-08-06T14:52:25Z

generic_tracers/generic_COBALT.F90

+          ! the sediment Fe flux (moles Fe m-2 sec-1) that would have result from the sinking organic matter flux and
+          ! O2 level of the adjacent waters.  This then spread across the layer mass (rho_dzt(i,j,k)) to give an input
+          ! in moles Fe kg-1 sec-1. Conceptually, this can be though of as a net iron flux resulting from the fraction
+          ! of the sinking flux that would have been intercepted at shallower depths were the model resolution finer.


add "is" after model resolution

I think this one might be OK as is, but I did need a "The" after finer.

gabyneg

I just made a few edits on the comments, and will start working on creating a few model tests to see how this changes the Main Hawaiian island domain.

gabyneg

I set up 3 experiments with the updated fe_coast formulation, where fe_coast =0.001, 0.01, and 0.1. Below I show the differences in surface dissolved iron compared to a base run without any changes to fe_coast (fe_coast =0).

(the columns represent base and differences between experiments (exp - base)). I have also looked at how surface chlorophyll

and the deep chlorophyll maximum at station ALOHA were modified by this change.

(DCM doesn't change between experiments). The changes between these three experiments do not seem to look linear, so I am having trouble understanding why the smallest changes in fe_coast (0.001), seem to show the largest differences in dissolved iron along the islands. I have verified my xml and code to make sure I didn't make a mistake plotting the figures.

andrew-c-ross · 2024-08-12T22:59:50Z

generic_tracers/generic_COBALT.F90

       endif !}
    enddo; enddo  !} i,j
+    ! CAS: Is the necessary?
    do k = 2, nk ; do j = jsc, jec ; do i = isc, iec   !{


I don't thinks so, since these are initialized with 0s here

Agreed. I have deleted. and we will confirm there are no answer changes.

Hi Gaby. Thank you for adding the figures and sorry it has taken me so long to look at them. I too am a bit surprised to not see a clearer progression from the small to the largest flux, but I think this may be partly due to the prominent boundary condition noise. I will be good to continue testing this in combination with Theresa's boundary nudging fixes.

charliestock · 2024-08-20T03:36:35Z

It looks like this has passed checks. Since the default fe_coast = 0 won't effect any answers, perhaps we can merge it in?

yichengt900

Thanks, @charliestock. The changes look good. I tested them with the NWA12 regression, and the results are identical. Approved.

charliestock added 3 commits August 3, 2024 21:27

sediment clean up, modified unused coastal iron source

de914e3

a few additional fixes found during testing

ad251d4

made Fe flux depend on adjacent O2 rather than bottom; properly scale…

07b9647

… flux with dzt

yichengt900 requested changes Aug 5, 2024

View reviewed changes

generic_tracers/generic_COBALT.F90 Show resolved Hide resolved

reset k = grid_kmt(i,j) before sediment caco3 calculation

d57dbe9

yichengt900 reviewed Aug 5, 2024

View reviewed changes

gabyneg reviewed Aug 6, 2024

View reviewed changes

gabyneg reviewed Aug 9, 2024

View reviewed changes

andrew-c-ross reviewed Aug 12, 2024

View reviewed changes

minor fixes in response to code reviews by YC, GNG and ACR

388b1b3

yichengt900 approved these changes Aug 20, 2024

View reviewed changes

yichengt900 merged commit 1e3b73b into dev/cefi Aug 20, 2024
1 check passed

yichengt900 deleted the sed_and_coastal/clean_code branch September 20, 2024 20:56

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Sed and coastal/clean code #89

Sed and coastal/clean code #89

charliestock commented Aug 4, 2024

charliestock commented Aug 5, 2024

yichengt900 commented Aug 5, 2024

yichengt900 left a comment

charliestock commented Aug 5, 2024

yichengt900 left a comment

yichengt900 Aug 5, 2024

charliestock Aug 20, 2024

yichengt900 Aug 5, 2024

charliestock Aug 20, 2024

yichengt900 Aug 5, 2024

charliestock Aug 20, 2024

yichengt900 Aug 5, 2024

charliestock Aug 20, 2024

gabyneg Aug 6, 2024

charliestock Aug 20, 2024

gabyneg Aug 6, 2024

charliestock Aug 20, 2024

gabyneg Aug 6, 2024

charliestock Aug 20, 2024

gabyneg left a comment

gabyneg left a comment

andrew-c-ross Aug 12, 2024

charliestock Aug 20, 2024

charliestock Aug 20, 2024

charliestock commented Aug 20, 2024

yichengt900 left a comment

Sed and coastal/clean code #89

Sed and coastal/clean code #89

Conversation

charliestock commented Aug 4, 2024

charliestock commented Aug 5, 2024

yichengt900 commented Aug 5, 2024

yichengt900 left a comment

Choose a reason for hiding this comment

charliestock commented Aug 5, 2024

yichengt900 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gabyneg left a comment

Choose a reason for hiding this comment

gabyneg left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

charliestock commented Aug 20, 2024

yichengt900 left a comment

Choose a reason for hiding this comment