Add FATES cross-grid seed dispersal mechanism #2077
Conversation
|
@ekluzek just a heads up that from the FATES software meeting today, I'm going to move forward with this as the next FATES api update. |
Rename hist fields to track them down more easily Renaming history fields to make easier to find in lists, e.g. when using ncview. For example, litter fields like MET_LIT and STR_LIT will be LIT_MET and LIT_STR.
If fates is not using dispersal mode, which is set via the parameter file, do not run the nearest neighbor algorithm to avoid performace impact.
Change order of history fields to improve performance on derecho Instead of just ordering history fields alphabetically, order them first by the name of their level dimension (with fields without a level dimension appearing first), then alphabetically within a given level dimension. This changed ordering gives a significant performance improvement especially noticeable on lustre file systems such as on derecho.
This corresponds to a fates-side refactor that moves the seed_out and seed_in variables from the bc_in/out types to site. Corrections to the column type names are also made
|
Cross posting that running fates regression tests on cheyenne, I'm seeing all As discussed in the ctsm software meeting this morning, this seems likely to be due the MPI_AllReduce not being able to maintain b4b when layouts change. |
|
Also per discussion during the software meeting, I'm going to add a namelist option to allow user's to turn off seed dispersal even if the seed dispersal parameters are not unset in the parameter file.
|
@billsacks was spot on with this being due to the non-associativity of floating point sums with MPI. I refactored the code to utilize a non-reducing MPI collective and it appears that the ERP tests are passing now. |
Since fates seed dispersal can occur at different user defined cadences, allow the user to define this via the namelist.
|
Per discussion today around #2151, I'm going to update the testmod to adopt a shell script from @adrifoster to call
|
This update implments a shell script to call ncgen and the fates modify parameter file tooling to generate and update the default fates parameters on the fly with values appropriate for seed dispersal mode.
Zeroing of wood product fluxes on fates columns products are otherwise zero'd in a p2c() routine that is incompatible with fates. When wood product fluxes become available via fates, these routines will be updated. These fluxes were previously left as uninitialized, which was causing math issues on some compilers.
|
@ekluzek and @adrifoster I think I've resolved all the requested changes. Would you take a look when you get a chance to see if they look good? |
SNICAR snow albedo scheme updates A few substantial changes in the SNICAR module for the following updated snow albedo calculation features: - Updated ice optical properties from Flanner et al. (2021), with multiple types for ice refractive indices. - Updated aerosol optical properties from Flanner et al. (2021) with multiple dust types & new BC and OC optics. - Updated downward solar spectra from Flanner et al. (2021) for multiple condition types. - More accurate radiative transfer solver (adding-doubling) from Dang et al. (2019). - Nonspherical snow grain scheme from He et al. (2017). - BC-snow internal mixing scheme from He et al. (2017). - Dust-snow internal mixing scheme from He et al. (2019). - Hyperspectral (480-band, 10-nm spectral res) capability with all the above features. - New namelist controls for aerosol in snow and additional snow albedo diagnostic output variables. Specific notes - Code contributors: Cenlin He (NCAR/RAL) with advice from Dave Lawrence (NCAR/CGD) and Mark Flanner (UMich). - The manuscript to report this update is Cenlin He, Mark Flanner, David M Lawrence, Yu Gu: New features and enhancements in Community Land Model (CLM5) snow albedo modeling: description, sensitivity, and evaluation. Authorea. June 08, 2023. DOI:10.22541/essoar.168626390.01530324/v1 - These updates will change the snow and surface albedo results along with other surface fluxes changes. - There are a few new namelist options related to SNICAR scheme added to the namelist control. New namelist variables: snicar_numrad_snw = 5 or 480 wavelength bands, default 5 snicar_solarspec, default "mid_latitude_winter" among six available options snicar_dust_optics, default "sahara" among three avail. options snicar_snw_shape, default "hexagonal_plate" among fourn avail. options snicar_use_aerosol, default .true. snicar_snobc_intmix and snicar_snodst_intmix, default .false. means do not activate bc-snow and dust-snow internal mixing do_sno_oc, default .false., already appeared in previous code but in caps use_snicar_frc, default .false., existed before fsnowoptics now points to an updated 5-band file and gives the option for a 480-band file
|
Regression testing with Cheyenne location: Testing |
|
Regression testing on izumi completed over the weekend. All expected tests are b4b/ Location: |
Description of changes
This PR enables FATES to disperse seeds across neighboring grid cells using MPI. The API update includes calls to new fates dispersal procedures and as such should be coordinated with NGEET/fates#1005.
The design documentation can be found in the FATES Developer's Guide.
Specific notes
Contributors other than yourself, if any: @YanlanLiu
CTSM Issues Fixed (include github issue #):
Are answers expected to change (and if so in what way)? Answer should not change for non seeddispersal mode
Any User Interface Changes (namelist or namelist defaults changes)?
New fates_seeddisp_cadence item and requires changes to the FATES parameter file
Testing performed, if any: FATES testing, will do aux_clm