Fesom2.6 recom tra diags and CI test #802
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Define new variables to track tracer changes due to advection and diffusion. We want to save for now diffusion and advection contribution to the tracer changes. Horizontal and vertical diffusion includes Redi parametrization (if it is set .true.).
Fill __ciso directive to ensure that carbon isotope code works. Medusa interface is added.
Use __coccos in src/CMakeLists.txt to include 2 more phytoplankton classes FESOM2.6-REcoM setup with 4 phytoplankton, 3 zooplankton and 2 detritus functional types. BREAKING CHANGE: FESOM2.6-REcoM setup with 4p3z2d
Correct temperature function for small phytoplankton and diatoms. Previous update for the fouth phytoplankton class and new temperature functions for small phytoplankton and diatoms resulted in very low Phaeocystis biomass. We recognised that this is related to missing code lines in the temperature function which we correct with this update.
…ostics Use namelist.io to activate diagnostics. Sinking velocity of particle (fast and slow) are calculated but not written into a file before. We fix the missing diagnostics for sinking velocity of detritus classes. Besides we Extracted the K0, solubility which is computed in gasx.F90.
…esom2.6_recom_tra_diags
We add __coccos directive to get the correct temperature function in silicate assimilation. In the coccos and Phaeocystis version, it should be dependent on the new temperature function, Temp_diatoms. Use the original Arrhenius function in simpler phytoplankton version. Besides, light limitation diagnostics is corrected. It does affect only the diatoms related output.
Correct reported salt flux bug in oce_ale_tracer.F90. #721
We commented out the linear model where degradation is directly proportional to light intensity and the chlorophyll degradation rate is capped at a maximum of 0.3. Reason for this is the unrealistic air2sea co2 fluxes in the 2p3z2d setup. The reactivated saturation model is more realistic. It accounts for light saturation effects. At low light, degradation increases roughly linearly with PAR At high light, the degradation rate approaches an asymptotic maximum
…tives with namelist parameters Replace preprocessor directives with namelist parameters: * `__3Zoo2Det` → `enable_3zoo2det` * `__coccos` → `enable_coccos` * Enhanced inline comments with ecological significance and biological context *Complete variable/parameter documentation with units and descriptions * Clear sectioning with visual separators and variable definitions * Equation explanations showing mathematical relationships and biological processes * Logic documentation for decision trees and conditional blocks * Comments now explain "why" not just "what" **Sequence Validation** * Generates expected tracer ID list based on configuration * Shows exact ID order and position-by-position comparison **Configuration-Specific Clash Detection** * Full Model (both flags): Don't use IDs 1023-1024 * Coccos-only: Use 1023-1028 (NOT 1029-1034) * 3Zoo2Det-only: Use 1029-1030 for microzoo (NOT 1035-1036) **Validation Checks** * Exact sequence matching against expected IDs * Duplicate ID detection *Forbidden ID detection per configuration BREAKING CHANGE: replace directives with namelist parameter for model configuration
JanStreffing
changed the title
Collaborator
Author
|
CI test for recom is on the way. Any help is welcome @suvarchal |
Collaborator
Still needed @ogurses? |
| !#if defined (__recom) ! not necessarily should belong to recom case | ||
| ! tracers%work%tra_advhoriz = 0.0 ! O:G - tra_diag | ||
| ! tracers%work%tra_advvert = 0.0 | ||
| ttf_rhs_bak = 0.0 |
Collaborator
There was a problem hiding this comment.
Has no impact on CI test, so probably ok. Just flagging to be checked by @patrickscholz
Collaborator
last-minute working on compute time reporting and proposal, can take a look tomorrow |
- Add namelist.recom.working and namelist.tra.working to test/input/recom/ - Based on branch owner's working versions with paths fixed for CI - bgc_num=22 for 2p1z1d configuration - REcoM_restart=false for initialization from climatology - Avoids any runtime sed issues that might cause parsing errors
- mkrun does NOT create namelist.recom or namelist.tra for REcoM - Copy from config/bin_2p1z1d/ and patch all file paths - Replace individual filenames in namelist.tra with our test files - Fix REcoM_restart=false and REcoMDataPath='./' - This should finally work\!
Line 178 was incorrectly reading 'pavariables' namelist when it should read 'parecomsetup', causing pavariables to be read twice. This resulted in 'Invalid line in namelist' and EOF errors when trying to run REcoM. Fixed by correcting the namelist read on line 178 from pavariables to parecomsetup, so the reading order matches the namelist file structure.
The code doesn't read nam_rsbc (no variables defined), so remove it from the namelist to avoid parsing errors. This also simplifies the workflow - we only need to patch REcoM_restart and tracer files now.
The 2p1z1d (22 tracers) configuration causes segfaults during initialization, likely due to arrays (wFluxPhy, wFluxDia, etc.) not being properly allocated for the simplified configuration. Use the full 2p3z2d (30 tracers) configuration which is better tested and has all arrays properly allocated.
The wFlux* arrays (wFluxDet, wFluxPhy, wFluxDia, wFluxCocco, wFluxPhaeo) and their global counterparts (GlowFlux*) were being initialized to 0.0 but were never allocated, causing a segmentation fault at line 137. Added allocate() statements for all missing arrays before initialization. This bug affected all REcoM configurations (2p1z1d and 2p3z2d).
- Changed initial year from 1958 to 1948 in clockinit - Updated REcoM forcing year range to 1948 - Added copying of 1948 forcing files in CI workflow The test forcing data files are from 1948, so we need to start the simulation in 1948 rather than 1958.
Only trigger REcoM CI test on pull requests to main branch or manual workflow_dispatch. This reduces CI load during development.
The CORE forcing data uses noleap calendar (365 days/year). Set include_fleapyear=False to match the forcing data calendar.
The monthly_event subroutine requires two arguments (do_output, N), but was called with only one at line 1872 in gen_surface_forcing.F90. This caused undefined behavior when accessing the uninitialized N parameter, leading to a segmentation fault during forcing updates. Fixed by adding N=1 (check every month) as the second argument.
The code has a hardcoded default nm_fe_data_file='DustClimMonthlyAlbani.nc' but our test file is named 'DustClimMonthlyAlbani_pimesh.nc'. Since we remove the nam_rsbc section (it's not recognized by the Fortran namelist parser), the code uses the default filename. Create a symlink so the file is accessible under both names.
The dust forcing file is read from ClimateDataPath (test/input/recom/), not from the working directory. Create the symlink in the correct location.
The RiverFe array was declared in recom_modules.F90 but never allocated, causing a segmentation fault at line 2080 in gen_surface_forcing.F90 when trying to set RiverFe = 0.0d0. Added allocation and initialization for RiverFe alongside the other river nutrient arrays.
Added mslp (mean sea level pressure) forcing configuration to enable CO2 flux calculations in REcoM. Without atmospheric pressure, the CO2 flux calculation fails with NaN values. Changes: - Added namelist.forcing section with l_mslp=True - Configured to read from 'slp.' files with 'slp' variable - Copy PHC2_salx.nc and runoff.nc to work directory in CI
Add preprocessing step to process REcoM input NetCDF files using 'cdo copy' (with ncks as fallback) to add proper _FillValue attributes. Files processed: - oxy5deg.nc - si5deg.nc - talk5deg.nc - din5deg.nc - GLODAPv2.2016b.PI_TCO2_fesom2_mmol_fix_z_Fillvalue.nc This ensures proper land masking and interpolation instead of using dummy fill values (10000000000.0 and 0.0). Suggested by colleague to fix interpolation issues.
Processed REcoM input files with 'cdo copy' to add proper fill values: - oxy5deg.nc - si5deg.nc - talk5deg.nc - din5deg.nc - GLODAPv2.2016b.PI_TCO2_fesom2_mmol_fix_z_Fillvalue.nc This is a one-time fix to the repository files, removing the need for runtime preprocessing in the CI workflow.
The default namelist.forcing.CORE2 already has mslp (atmospheric pressure) configured but disabled (l_mslp=.false.). Patch it to l_mslp=.true. in the workflow to enable reading of slp.*.nc files. This is needed for REcoM CO2 flux calculations which require atmospheric pressure (Patm). Removed unnecessary namelist.forcing section from setup.yml since we just patch the default config file directly.
The sed command wasn't matching because dots are regex wildcards. Escaped the dots in .false. to match the literal string. Now l_mslp=.false. will correctly be replaced with l_mslp=.true.
Show l_mslp status in the test run output to diagnose why atmospheric pressure is still not being read (Patm=0.0).
The namelist.forcing file has: l_mslp = .false. (with spaces) Previous pattern l_mslp=\.false\. didn't match. New pattern l_mslp *= *\.false\. matches optional spaces. This should finally enable atmospheric pressure forcing\!
Namelists are working correctly now, so removed verbose dumps. Kept minimal verification that patches were applied: - REcoM_restart setting - l_mslp atmospheric pressure flag This will make the CI logs much cleaner and easier to read.
Problem: Fortran NetCDF library and CDO/NCO tools use different fill value conventions, causing NaN values in tracer initialization. Solution: Make fill value detection more robust: 1. Try _FillValue attribute (Fortran NetCDF standard) 2. Fall back to missing_value attribute (used by CDO/NCO) 3. Use NetCDF default fill value (9.969e+36) if neither exists 4. Broadcast fill value to all MPI ranks This fixes the 'REcoM_DIC: NaN' error during initialization by properly handling land/missing values in input NetCDF files created with different tools. Fixes issue reported by colleague about incompatible fill value conventions between different NetCDF tools.
Added diagnostic output to track where DIC becomes NaN: - Check state(idic) and sms(idic) before DIC calculation - Check DIC after calculation - Print node, depth, lon/lat location when NaN detected - Stop execution immediately to get clean traceback This will help identify whether: 1. The state already contains NaN (initialization issue) 2. The SMS contains NaN (biogeochemistry issue) 3. The addition creates NaN (numerical issue) The error occurs at lon=-43.88, lat=14.55 consistently, so this should pinpoint the exact cause.
Added diagnostic checks at the point where DIC is read from state array and converted for CO2 flux calculation. Since the SMS debug didn't trigger, the NaN must appear between SMS calculation and forcing. This will catch: 1. If state(one,idic) is already NaN when forcing starts 2. If the conversion to REcoM_DIC creates NaN Will print node number, coordinates, and related state values to identify the exact source of NaN.
JanStreffing
force-pushed
the
fesom2.6_recom_tra_diags
branch
from
619ee7c to
c6c21e0
Compare
Collaborator
|
A couple of interesting changes so far. Some of the recom code was missing allocations, that triggered GNU errors. I am left with NANs in one init field that I failed to resolve. @ogurses, @suvarchal: Can you check this out? |
When ballasting is used, vertical sinking is adjusted if the detritus carbon biomass exceeds a treshold value (0.001). This results in large air-sea co2flux uptake. Weremove this constraint.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.