further updates for ilamb pre-processing waiting on Romain for approving (currently O/S)
Agreed to go ahead with Australian PFT meeting for Friday. Discussed getting more clarity on which parameter values will be important / most sensitive - both in pft_params.nml and pftlookup file.
@RachelLaw shared canopy height slides - heights based on wood pool for trees and LAI for grasses. Grass pfts have seasonality but grid-cell canopy height dominated by tree height. Using different scaling factor to give taller c4 grass/crop than c3 grass/crop. @Jhan has run test with new canopy heights in vegfunc file produced by @lachlanswhyborn. Seems to be running stably. Looked at daily output for first month to see when results deviate. Occurs from first day so suggests canopy height being updated at start of run, not relying on canopy height from restart file (which was left at the original values). Need to confirm, then look for any impact over longer means.
@alexnorton showed progress on zeric/mesic distributions for Australia and how NVIS data has been processed to produce these distributions. Discussion of the resulting mesic fractions in Central Australia. Due to open mixed woodlands. Agreed to split these 50% to xeric and mesic rather than all to mesic. Considered how to merge with LUH3 data for change in land-use but expect to be reasonable to implement as so much of Australia is classified as non-forest by LUH3.
@RachelLaw showed slides with results from Jhanâs run using spatially dependent canopy height (but fixed via ancillary not interactive). Some changes in temperature, precip, wind speed. Wind speed changes most noticeable for grass/shrub/tundra areas where relative change in height can be reasonably large. Discussion of next steps and whether to try and make canopy height interactive - and what the impact would be under future conditions. Noted that if stayed with using ancillary then need to consider what dataset to use to create the ancillary e.g. long-term mean of LAI rather than individual year as currently used. Agreed that worth coding up interactive case since should be relatively simple but need to be careful about where it is updated etc.
Issue with code/restart for testing Australian pfts
Current status from Jhan:
⢠feat-Aust-PFTs-106 failed at dt=50
⢠Turning off LAIfeedback it worked
⢠BUT fields in restart looked sorta OK
⢠Tested without Aust PFTs in feat-test-mainB leads to same failure at dt=50
⢠Focused on feat-test-mainB to remove potential conflation of issues
⢠gLAI on first timestep seems to be issue
⢠multiple, alternative restarts work. Alternative to PI-02-WithLUH3VegetationMap_2025-03-14.astart, PI-02-WithLUH3VegetationMap.astart
⢠Remaking the Aust PFTs from any of these alternatives will (almost) certainly get passed the initial crash
⢠This is not a guarantee that there will be no further issues with Aust PFTs
Obviously we need to get this resolved ASAP.
@lachlanswhyborn has made new restarts but initial look suggests not what was expected (one file has Australian 2/4 replaced with 12/13 when wasnât expected). Also need to confirm that canopy height in restart not used as set to zero for 12, 13. Previous year tile fraction looks like LUH2 while current year looks like LUH3 - though previous year shouldnât be used without l_luc flag switched on. @lachlanswhyborn will create some more restarts and @Jhan will continue with test runs.
If not a restart problem, have we missed something in the code which is dependent on hard-wired vegetation type - presumably in CASA?
For ilamb prep, @inh599 has written script to adjust dates on files so that can compare difference sections of runs.
Parameters: Rachel and Alex are fine-tuning the âwish-listâ of most important parameters to send to WSU team - Laura Williams and Belinda Medlyn. They will provide trait data where available, Laura has an idea of how to map species-level data to our PFT classes. Juergen and Alex agree that we will need to do a global comparison of tree type PFT parameters following the new assignment of Aus PFT parameters, to ensure the ârelativeâ differences are sensible.
Restart file tests: Changed PFT 2 to 12, and PFT 4 to 13 across entire globe. This resulted in no change in screen temperature - considered a successful test. Next test is to limit the PFT changes just to the Australian region. Lachlan and/or Jhan should have results by next week.
Model test with global swap of 2/12 and 4/13 showing some odd behaviour for type 13. What have we missed?
Progress on parameter values
Vegetation distribution without Australian pfts. @alexnorton has prepared a new version, waiting on @RachelLaw to review.
Configuration âto doâ table. Review of land items.
Notes
PFT swaps:
Rachel: The GPP is collapsing (to 0) for PFT 13 after several years of simulation.
Lachlan: run with zeroing out the PFTs 2 and 4 is creating differences in the run while these have a tile fraction of 0.
It would be useful if the collapse is reproducible if starting the run on that year to make a shorter testcase to work with.
Alex: Is there any chance there is a PFT-dependence left in the coupling? Could we use Rachelâs single point simulations to track the issue? Or the offline CABLE code although it isnât in sync with ESM1.6 (Claire)
Parameter values:
Rachel working on the pftlookup file. Trying to identify the parameters we are using and we need values for.
Rachel: In the CABLE documentation, it would be good for the parameters to identified if they are used only with certain options.
Comparison of Walker with before for seasonal cycle of GPP, soil respiration, NPP and NEP. Big suppression of NEP boreal maximum.
NEP suppression in evergreen needleleaf. Suppression not seen in plant respiration. Massive increase in leaf respiration: Leaf respiration ratio to GPP is pushed to 0.5 which is way too high.
vcmax increasing by 2.5 times but leaf respiration increases by 8-9 times. Trying to figure out what is happening.
Other results seem to be ok with the Walker vcmax.
Ian: there could be some connection with the climate variables or other parts of the code that need to be brought in with the Walker vcmax calculations.
Juergen: worth checking the respiration and vcmax temperature response.
Alex: worth looking at tweaking the sla parameter to see the effect on the leaf respiration while still looking to explain what is happening.
Ian: from Jhan, it will not be easy to get vcmax out of ESM1.6. It wonât be easier in AM3.
Australian PFTs
2/12 and 4/13 swap is not collapsing GPP anymore, caused by using an old parameter file.
Jhan chasing down reproducibility. There may be MOSES parameters being used and making progress when parameters are made to agree
Alex sent list of parameters for input.
Rachel to review the veg. distribution with Aust PFTs and without to see if we want to set a min. fraction that is not worth simulating, as we have done in the past.
Before we used a 10% fraction threshold: a grid cell never reached more than 10% of that type with land use change. Alex: what was done with these 10% then? Rachel: it was distributed proportionally to other PFTs in the grid cell.
Rachel: issue with 10% on the coast lines though.
Alex: should we reduce the threshold instead of no threshold at all? Something like 1-3%.
Ian: Are the new PFTs evergreen:
Rachel: yes, the code assumes everything is evergreen unless there is something in the phenology file. Itâs the simplest for now to keep them evergreen.
Ian looked into the physical parameters but most important will be the bgc ones. (@inh599 edit: and biophysical trait parameters such as stomatal function)
@RachelLaw, lots of parameters are also completely unused, it would be best to get rid of them completely.
Swapping PFTs
2 runs,JuneSpinupStandard and JuneSpinupJuggle, should use the same except for the restart files. They still diverge very quickly. Nothing is looking crazy but it is different.
Question on what happens on the first UM radiation time step.
LUH3.1.1 data
Loss of temporal resolution in recent years. Decision from LUH data providers.
EGBL appearing in northern Russia: probably a bug to check
Tile fraction not quite summing to 1 at each time step. Problems in some Canadian tiles.
Changes between 3.1.1 and 3.0 for 1850:
Switch of EGNL to DNL in high latitudes. Also change noted in Tibetan plateau.
Reduced bareground, increase C4 grass in Australia
Decrease in trees and increase in grass in Europe.
Change between present day and 1850:
largest differences in DBL, more change from grass to crop
Chair: Alex
In Attendance: Alex, Ian, Claire, Lachlan, Jhan, Tilo, Juergen, Ben
For discussion
Notes
Historical Land-Use Change:
Alex benchmarked the LUH3.1.1 derived ACCESS PFT maps against the HYBMAP land cover map (2001-2020). Showed significant overestimation of forest/tree PFT area in high latitude regions. Many tundra PFT regions were replaced by tree PFTs (EGNL, DNL), while HYBMAP shows these areas as shrub/grass. Fix: Include tundra in the âforestâ LUH3 category so its 1850 distribution is largely retained.
Still need to finalise this historical land-use change map: Assign minimum fraction threshold and redistribute PFT fractions below this limit, add future potential PFTs with 1e-6 fraction.
Northern land seasonal amplitude underestimation bias:
Seasonal amplitude issue: Ben flagged that northern ecosystems need temperature sensitive phenology, which is still missing in CABLE. This could go a long way to resolving seasonal amplitude issue and incorporate a trend in GPP. Ian noted that Vanessa had developed a climate-sensitive phenology module which we should try to integrate in CABLE4.
ILAMB:
Ian is looking at ILAMB again. Comparisons with ILAMB benchmarks also show underestimation of global seasonal amplitude (GPP and NEE). Modeled albedo also looks like it is overestimated in mid-year at global average. Still working on getting the settings functional to look at specified regions.
Benchmarking ACCESS-ESM1.6 with CABLE:
Lachlan is testing an ACCESS-ESM1.6 build with CABLE compiled into a static library from the CABLE repository directly. This should give the same results as the release version of ACCESS-ESM1.6, which has the CABLE code included as source in the UM7 code (i.e. the files just copied into UM directory). This is a benchmarking test with the access-esm1.6-configs. It is not giving bit-wise equivalent results yet. He is running more tests, double checking flags and making more settings explicit.
Interactive phenology:
CABLE does have a simple interactive phenology module. This is independent of more advance capability such as POP, SLI etc. and sits in MAIN - see cable_phenology.F90
To activate this capability would require some addtional work - notably
4 additional variables added to the UM-restart (for %chilldays, %gdd5, %dtemp and %dmoist) and threaded through to CABLE (implicit) for use.
cable_phenology to be refactored so as to not depend on CABLE-offline structures such as the patch% TYPE. cable_phenology has been modified recently as part of the attempt to insert Australian PFTs (i.e. hard-wired indices should be no issue).
creation of a much-simplified version of cable_climate that calculates only the relevant information
calls to cable_climate and cable_phenology inserted in the relevant places in the time stepping (cbm from cable_implicit) and/or the namelist phenology setting changed.
Vegetation distribution from LUH3-1-1. Readiness for testing? Need a restart.
Impact of attempts to make EGNL more productive
CABLE as a library update
Notes
Vegetation distribution from LUH3-1-1
need to adjust the restart to the changed tiles.
Rachel looked at the min. threshold we want for tile fraction. 0.01 looks like a decent threshold. Need to bump up tile fraction to 1e-6 (or 1e-4?) for tiles that will become used.
Alex to try Lachlanâs tool to create the restart starting from a restart from the June spinup.
Need to try the AMIP run with land-use change with this new veg. distribution.
EGNL productivity
Rachel sharing results. Sudden drop in productivity, LAI and NPP after about 10 years of simulation. First guess is to look at what is happening in the nitrogen cycle. Supported by the increase in labile carbon.
Rachel to look at all the pools information.
A step change in the drivers (at the start of the run) can also cause this.
Is the spike in leaf respiration realistic since productivity in going down? It could be happening because there is a step to calculate vcmax that doesnât know about the nutrient limitation that would lead to a higher vcmax and higher respiration.
CABLE library
Identical results with a library or not with a compilation without optimisation. Optimisation
Need to confirm results with optimisation are acceptable: GPP, NBP in AMIP and PI-control
New testing system coming together as well
ILAMB results
Not a lot of results given by ILAMB for overall score. More info available for individual scores. Donât have the spatial maps either.
GPP: phasing of the annual cycle is ok but not the amplitude.
Albedo: not getting a dip in albedo in the southern hemisphere. Annual cycle in the boreal regions is not quite correct.
Australian pft parameters - Alexâs summary slides
CABLE as a library
EGNL productivity / nutrient limitation
Notes
CABLE as a library: Removing the optimisations didnât work in fact. But checked that the CABLE code from the library branch gives the same answers as before if compiled inline. It definitely has to do with the compilation.
Australian PFT parameters:
Mallee more suited to Xeric. Should change before testing.
Should we use a precipitation-based threshold to define the separation between Mesic and Xeric? Should tall Eucalypts be EGBL or Mesic?
Right now, just need to implement what we already have and test that. Alternatives could be explored in a research project later on.
Leaf N:C and N:P given by Laura, based on 10th-90th percentiles, show a very large range compared to what is used in CABLE. Ask Laura for the ranges for 25th-75th percentiles.
EGNL productivity / nutrient limitation:
Still seeing issues with 1pctCO2 run.
Canât reproduce the same behaviour in productivity with offline.
Need to check nitrogen deposition. Units?
This paper shows CABLE-POP (v11 or 12) show very high veg. C:N ratio in northern high latitudes compared to other models. But CABLE-POP dynamic could be different from CABLE3 because of some parameters and other code changes that arenât reconciled.
Data request:
What time averaging to choose?
Pools are useful as instantaneous variables to check for carbon conservation, but we should only have monthly mean values for production runs. We still have the carbon conservation script.
Historical land cover updated to remove tiles that always stay under 1%. 1e-6 is the min. start fraction for tiles that pop up later in the historical period.
Question about the future scenarios and what happens for new tiles? Not aiming for consistency, we would adapt the restart files, initialising the new tiles.
Jhan to give Alex the PFT parameters file in new format.
Going in 2 steps:
change the distribution without the Aus. PFT params
change the parameters for the Aus PFTs.
Tilo to send to Alex info on how the land-use change was done in ESM1.5 to prepare for ESM1.6
Need to review parameters for C4 crops. Some of the parameters in pftlookup file donât look good.
The root distribution/depth parameter needs some sensitivity tests. Running with the same root distribution for all PFTs and only changing the max. root depth (only in Haverd2013).
g1 parameter: Laura gave information to Alex during the week.
Documentation on the provenance of the parameters for the Aus. PFTs: Alex keeping a document updated. There will be a model description paper but the Australia PFTs might need something separate.
Parameter file versioning: vk83 provides versioning. We can keep information in the CABLE documentation on the differences between these various versions. This will not work for non-Gadi users of CABLE, we will look into this when we provide supported configurations.
PFT 12/13: seems there is something wrong somewhere. Jhan followed this to an uninitialised factor in the convection, looking in UM code history to find a fix.
EGNL productivity / nutrient limitation
Test with nitrogen deposition / 365 in CABLE offline: broke the relation gpp - npp = autoresp. Probably coming from hitting the case where NPP < 0.
Is ACCESS more sensitive to low nitrogen than offline?
Data request
Progress made by Rachel and Ian. Havenât found any requests that would require changes in the code.
Some changes might be required in MOPPeR. Some outputs might be better picked up from the tiles variables rather than the UM variables.
AM3
Jhan canât find any CABLE fields in the output. Might be missing the STASH request.
Grid-cell outputs should be getting out. Other variables (like tiled soil) might need extra work.
Need to check the outputs from AM3 for the release. It seems to be outputting more than we need.
Do we want to harmonise the standard diagnostics for config releases? Spencer has done some work to have light, medium and heavy diagnostics requests. But being revisited with CMIP7 data request. We may want to use this work when itâs ready.
AMIP
Need a new test with the current spinup before moving to the CICE5 version.
Australian pfts - testing âsprintâ, experiment design
Other âfinalâ CABLE changes
Notes
Australian PFTs
Rachel found another hard-wired number in CASA that assumes trees are the first PFTs. Looks easy to fix. It is in the thinning code. Other impact in distribution of wood to litter, reset to 0 for the trees but wonât find the 12/13 PFTs as trees. We could use another variable here to fix this. Issue is open, might be easier for Jhan to test.
Test with no impact: best to use the ESM1.5 distribution so we can turn on land-use change.
Ian has a problem with the fix to the convection, unsure if this has been changed yet.
Jhan ran 1 year with new veg distribution from Alex. Seems to have run fine.
Tests in AMIP runs until the PI-control has the CICE and ocean grid changes.
Need to ensure the phenology file is modified so PFT 13 changes from deciduous to evergreen.
EGNL low production
Lowering the limit to the leaf C:N ratio has an impact on the NPP, less reduction at the peak of the seasonal cycle in CABLE offline.
I have 5 year AMIP runs completed or underway for testing the Australian PFTs and land-use change.
All output is in /scratch/p66/rml599/access-esm/archive
Runs 1-4 use input files that have been modified to include pft 12 and 13 information with edits to config.yaml and cable.nml:
cable_vegfunc_N96.anc
pftlookup_cable3_1213_v1.csv
pft_params.nml
modis_phenology_csiro_nophase.txt
Control run, starting from restart790 from JuneSpinUp run with date modified for amip run. No land use change.
test-790orig-test-790orig-d0d4b96e
Run with restart file from @alexnorton, /g/data/p66/ajn563/ACCESS-ESM/restart_luh3.astart adapted from restart790, further modified for amip start date. No land-use change.
test-ANveg-test-ANveg-f98bdeea
As 2 but with type 13 changed to evergreen via update to phenology file (modis_phenology_csiro_13evergreen.txt). No land use change.
test-13evergreen-test-13evergreen-140c52e5
As 2 but with land use change switched on. First attempt failed in script for updating vegetation distribution at the end of the year. Suspect differences in the dimension/variable names between @alexnortonâs new file and ESM1.5 file. Fudged file to match ESM1.5. Ran.
test-luc-withbug-test-luc-withbug-219556e8
As 4 but with updated parameters for c4crop (pft 10). Required updated pft_params.nml (mostly g1) and pftlookup_cable3_1213_v2.csv. No change to phenology file.
test-c4crop-test-c4crop-b845de22
As 2 but with froot values varied by pft (but no change to zr). New pft_params.nml copy of /g/data/p66/rml599/amip-test/pft_params.nml.altroot
test-altroot-test-altroot-df6479a6
As 4 but with executable as B i.e. with mlogmax removed.
test-luc-newexe-test-luc-newexe-f655aa74
Runs A and B use original ESM1.5 vegetation distributions
A. Control run with land use change, no changes to dev-amip
orig-amip-orig-amip-9cb87380
B. Test run with new executable to remove âmlogmaxâ and assumption that tree types always at start of pft ordering. Aim is to reproduce case 5. Appears to match exactly.
orig-amip-newexe-orig-amip-newexe-0718de83
Latest ESM1.6 testing. AMIP tests described here with more planned for pft 12, 13 parameter testing.
Notes
Restart issue with Aus PFTs - carbon pools are zero for PFTs 12 and 13:
Need to double check Lachlanâs code to ensure the PFTs 12, 13 can use data from PFTs 2, 4. There is a setting in the config to do this:
Parameterisation tests for Aus PFTs:
Alex has a list of sensitivity tests to run in AMIP configuration with corresponding parameter input files ready.
How to benchmark these tests?
Put AMIP runs through ILAMB. Ian @inh599 to show Alex how to set this up.
Phosphorus limitation on Vcmax for Aus PFTs (12 and 13):
Do we want to implement this? Yes, likely need to apply P limitation given nutrient poor soils in Aus and that in ESM1.5 Australia had mostly EGBL (PFT 2) which already had P limitation.
We should test the P limitation switched on/off for PFTs 12 and 13. Requires code change in casa_feedback.f90.
@lachlanswhyborn - we have been running amip runs based on dev-amip but with different parameter files, both pft_params.nml and pftlookup. What we need help with is the canopy height for the Australian pfts 12 and 13. This goes into the âvegfuncâ file as a constant at all grid-cells, and probably into the restart file as well. @alexnorton had a go at doing the vegfunc file but ran into problems. He has the values for the new canopy heights that we wanted to use.