CABLE4 planning: meeting notes

Land Surface CABLE
1

This topic records meeting notes and action items from weekly CABLE4 planning.

2

January 25th 2024

For discussion (progress on tasks, questions, issues etc)

  • Meeting structure @RachelLaw
  • Dominant tree type per grid-cell: pros and cons @inh599

Notes

  • Initially work on 3 tile version only for ACCESS implementation but need to be careful not to lose non-POP multi-tree, multi-grass pft capability.
  • Choose how to treat shrubs. Probably need to treat as a tree not a grass. Already being treated as woody in POPLUC.
  • POPLUC currently all vegetated or not. Should be able to handle a vegetated fraction to allow for fractional bareground, lake, ice, urban. Note that ACCESS currently does not allow fractional ice; grid-cell has to be all ice - possibly inherited from what JULES does.
  • Will need to consider how LUH2 (or update) data gets used in the POPLUC-ACCESS workflow.
  • Treatment of crop information from LUH2 will need consideration. Current POPLUC implementation may not be incorporating temporal change.

Actions

  • @RachelLaw to look at grassy types and how many exist in a single grid-cell. Bare ground needed for dust. Should be able to handle a bare ground fraction and lake, ice, urban.
  • LUH2 processing code to be copied from x45 to NRI space. @clairecarouge or team
  • @juergen to check crops from current Trendy experiment
  • Separate executable/inline code discussion scheduled for Feb 1. Let @clairecarouge / @inh599 know if any you have relevant input for consideration.
  • @RachelLaw to cancel Feb 8 meeting (AMOS)
3

February 1st 2024

For discussion

  • CABLE-POP at ACCESS resolution - next steps.
  • Has CABLE-POPLUC been run globally for future scenarios or higher than present-day CO2 concentration?

Notes

  • All files needed for CABLE-POPLUC now copied to rp23. Run not yet tested.
  • @Jhan tried running code with POP switched on from standard branch but need to use separate CABLE-POP branch. Use CABLE-POP-TRENDY branch.
  • No POPLUC dynamics depends on whether c3/c4 - only photosynthesis. So potential to composite grass types when pass to POP. Anything to be considered in LUH2 e.g. harvest fraction.
  • POP has been run for future scenarios (without LUC). Disturbance frequency kept constant. Could look at stress mortality outputs and impact on NPP. Australian work is showing mass die-back and recovery in some areas. Could be ‘quirk/bug’. May be fixed in CABLE-POP-TRENDY branch. Alison looking at in Australian case.
  • POP hasn’t been run with phosphorus on. Will need to re-visit this. Also performance in ACCESS and whether would benefit from recalibration with newer datasets.
  • Offline (CABLE-POP) runs also useful for tuning/calibrating nutrient cycles etc.
  • What do we need/want for evaluation on land? Standardised/automated. Think about this for Hackathon. Also making use of ILAMB. Involvement of Land Working Group - and ESM working group.
  • Separate executable has implications for offline runs as well.

Action

  • @clairecarouge to try a test run of CABLE-POP. @juergen available to help with error messages.
  • ~8 variables at daily resolution. Have these all been cmorised? @inh599 to confirm what is available and what to run (from ESM1.5 or AM3). Note grid shift between ESM1.5 and AM3 so will need two offline set-ups to match the two grids.
    Done. Details here: Run CABLE-POP at ACCESS Resolution - #2 by inh599
  • Other input files/ancillaries will need to be reconciled in a staged process including gridinfo. Gridinfo may have been derived from ACCESS. @Jhan to check - also how ancillaries have been updated for AM3.
  • @clairecarouge and team to look at formatting of ACCESS meteorology and changes required from cmorised files for CABLE-POP to read.
4

February 15th 2024

Previous actions status

  • land-mask split with R converted to Python. @clairecarouge continuing with next steps with @Juergen 's support. Different solution will be required in ACCESS implementation.
  • gridinfo: @Jhan has had a look. Need to confirm where inputs/ancillaries are different or the same between offline/online. @jhan to write summary.

For discussion

Notes

  • Proposal to create ‘cable tools’ to gather splitting/merging tools, pre/post processing.
  • Systematic test of with/without diffuse fraction to check sensitivity
  • Checking datasets for portability across CABLE-POP and ACCESS and which datasets we need to use from each configuration.
  • Conversion of restart files between offline/online. What are the critical fields that need to be passed between configurations.

Action

5

February 22nd 2024

Previous actions status

For discussion

Notes

  • C3/C4 split at spin-up time is limitation. Do we want to let this change over time as climate changes? Is there a need for mix of c3/c4 grasses in a grid-cell. Is the c4 fraction capability in CABLE still functional? Better to have the functionality to carry different grass/crop types in a grid-cell. Will have deal with how to manage transition and what is happening in the soil under each type within ‘grass’.

Action

  • Discussion on design for e.g. multiple types within grass @inh599 @Juergen @bens
  • @RachelLaw to plot some ACCESS/POPLUC comparisons at individual locations over full timeseries.
  • POPLUC output to be checked against input. Crop appears not to be changing.
6

A post was merged into an existing topic: Run CABLE-POP at ACCESS Resolution

7

February 29th

Previous actions status

  • ACCESS/POPLUC landuse comparisons at a location (@RachelLaw )
  • Does ACCESS resolution gridinfo exist? What did YPW pass onto @clairecarouge ?

For discussion

  • high level configuration options - 17 vs 27 tiles (@inh599 )

Notes

  • SW WA does look odd. Most land use change in 1960s seems OK if a decade early. 2000 onwards looks odd. States and transitions should be consistent. Might be better to start with the states and determine transitions from that but need to consider further.
    INH note (after meeting): POPLUC requires the gross transitions - it is not possible to determine these solely from changes in the states.
  • ACCESS resolution gridinfo available on rp23. Also need to check whether it is on ESM1.5 or CM2 grid. Location of the files on N96 grid can be found here: crujra config at N96.
  • @inh599 - UM limitations on number of tiles, mostly due to UKCA coding. Decision is important now for AM3 work.
  • @inh599 started thinking about multiple grass types and will organise meeting.

Action

  • @Juergen to check POPLUC after 100 years to see how compares with LUH2.
  • @inh599 to talk with Peter Briggs about how LUH2 processed for POPLUC.
  • Audit of where tile/pft types are hard-coded across trunk and CABLE_POP branches. LS team.

Further information regarding the top level discussion around number of tiles:

Screenshot 2024-02-29 120958
Screenshot 2024-02-29 1209581596×895 255 KB

@Jhan is attempting to run a 27 tile version of the proto-type AM3 to assess memory considerations.

8

March 21st

Previous actions status

  • ACCESS resolution gridinfo file. Jhan has checked and has slides. Confirmed at ESM1.5 resolution but doesn’t seem to be exact match - grid-shifted (half-grid difference between CM2/ESM1.5, not clear what extrapolation/interpolation was done). Enough differences to worry about that easier to re-do for transparency etc.
  • 27 tile can be used if we want. Not an issue. Begun testing but haven’t completely verified. Memory OK. Inserted extra 10 spare tiles in the middle of our original 17 tiles as need to keep non-vegetated at the end. Technically should be the same. Jhan hoping to prove this later today. Deposition code may break bitwise agreement.
  • Ian setting up meeting with Peter Briggs. @Juergen expects to work on the LUH2 processing - important for Trendy as well.

For discussion

Notes
Apology: Tilo

  • Possibility of feeding back Australian data into LUH system. Need to do the work first here. Also all the other ‘bits’ of LUH would be a problem.
  • Discussion of @inh599 slides around grass redistribution. Thinking about grass/pasture/crop. New pasture pft on ACCESS side. Carry fraction of carbon pool rather than carbon pool in POP?? Conservation of CNP is priority. Possibility of using gross transitions for woody and for grass, compared with using net as current. Is this achievable on timeline? Is YPW’s code useable for doing the gross transitions? Extend POPLUC code but also challenging? Do land use change at start of year as currently in ACCESS.

Action

  • @RachelLaw to start topic on LUH processing. @inh599 and others starting to look at LUH processing and bringing in Australian information.
  • Copy LUH2 processing scripts to CABLE-tools for evaluation.
9

2 posts were merged into an existing topic: Managing CABLE-POP and ACCESS differences in vegetation distribution

10

March 28th 2024

Previous actions status

  • 27 tile case has had further work/thought. Deposition part still to happen.

For discussion

Notes

  • Comparision of tree demography models for site. @Juergen found many flags in code. Looking at best configuration and closes mass balance. Plans to do runs in mid-late April.
  • LUH2 data for SW WA - discussion of states/transitions (116E, 30S)
  • Got CABLE-POP running in rp23. Will make accessible to others. Trying 100 parallel case - worked from /scratch but not from /g/data. Filesystem issue?
  • Start with running Trendy at ACCESS resolution? But need to deal with LUH2 processing. Peter’s scripts should work. Same chain of workflow. Consideration of re-gridding. Nearest neighbour.
  • When using ACCESS meteorology - take daily data and put into format as for Trendy case. Or @clairecarouge suggests writing something for CMIP style output but trying to keep it fairly generic.
  • Bigger question is how to make more generic input routines. Make better use of namelist for selecting input dataset rather than embedded in code. Weather generator is another consideration.
  • Useful to run CABLE at ACCESS resolution without POP.
  • Tammas working on thinning code in old LUC scheme. Fixed initialisation problem. Running historical case. Checking for change in NBP. Thinning derived from LUH2, aggregation of two terms (possibly out of 4 available harvest). Aim is to remove fraction of wood pool (leaf and root goes to litter). Starts in 1850. Previous tests have done runs with non-zero initial carbon pool. Hoping that some thinning and some initial harvest pool will be best. Is carbon conserved when thinning? Need to double-check.

Action

  • @inh599 to copy Peter’s LUH2 processing scripts to rp23.
  • @clairecarouge will put list on Forum of what needs to be re-gridded and suggested method.
11

April 4th 2024

Previous actions status

  • @inh599 put LUH2 processing code on rp23. Copy of files as made with those scripts as used in Trendy 2023 runs. README provided by @Juergen

For discussion

Notes

  • LUC3 code. Uses gross transition not net. Some capability for harvest. Probably only from primary forest not secondary. Not sure how harvesting is done. Goes via other parts of land use change scheme. Single set of calls for biophysics and biogeochemistry. Different to current ACCESS code so probably helpful - or could be more awkward if all information not available. Assumes a 17 tile configuration and ordering. Different from anything else. Not sure how easy to change. Some hard-wiring. Written for offline so undesirable in some aspects. Recreating gridinfo and number of active tiles as go along. Would need to unpick if want to use. Not sure how it would work with UM style parallelisation due to global read of transitions. Not immediately adoptable but closer than CABLE4 - mostly technical changes but fairly inflexible. @tammasloughran query need to parallelise if just once a year - but @inh599 notes land points split. Could do processor by processor read. Transitions is more information to input that states. Could do as ESM1.5 to update once per year when model restarts. Would need pre-processing code for LUH2 data into the form for this scheme.
  • POP can harvest from a particular cohort, young/old.
  • SE and E Australia plus Indonesian grid-cell - CABLE-POP/ACCESS/LUH2 comparison shows big differences. Need to be aware of land cover versus land use differences. Potential vegetation accounting for sparse trees.

Action

  • @Juergen to follow up on LUH2 processing since also useful to understand for this year’s Trendy.
12

April 11th

Previous actions status

  • @Jhan working through 27 tile AM3 version - runs but gives different results to 17 tile (after couple of timesteps) when extra tiles are all zero. Possibility that hard-wiring for JULES use of 27 tiles is being picked up somewhere with unintended consequences.

For discussion

  • @Juergen: comparison of CABLE-POP LUC output with LUH2 data

  • Who/what/when (including CABLE4 vs AM3/CM3)

Notes

  • For LUH2, need to be aware of difference between primary potential forest and primary potential non-forest and how these are interpreted in CABLE-POP. Australia largely primary potential non-forest and CABLE-POP maps to a fraction of tree/grass depending on the biome classification derived from climate statistics e.g. may be 0.4 tree and 0.6 grass. Comparison of CABLE-POP and LUH2 aggregated to 1 degree shows timing of transitions looks correct but magnitude not always easy to understand.
  • For ACCESS will need to decide how to determine the potential vegetation/1850 pft distribution.
  • ‘Flowchart’ of work streams for further discussion next week.
    CABLE4-plans.pdf (54.7 KB)

Action

  • @Juergen to continue LUH2 comparison focussed on grass types and transitions.
  • Identify who is working on which parts of flowchart now (edit powerpoint, shared separately or let @RachelLaw know suggested edits)
  • Consider what tasks are not resourced but should be a priority for near-term work.
13

April 18th

Apologies: Ian, Tilo

Previous actions status

For discussion

Notes

  • No meeting April 25th (Anzac day)
  • @Juergen reported on email exchange with Peter B around LUH2 processing and @RachelLaw reported on email from @inh599 with relevant notes here.
  • Discussion of shrubs - continuum in woody vegetation, short to tall. Arbitrary to discriminate shrubs and tree. Mallee/short savannah could be sparse woodland. ESM currently using shrub but perhaps better not to use. POP treats shrub as woody and runs for shrub. Check how shrubs differ - just in parameters. vcmax. Vegetation height prescribed in ESM but would come out of POP. ESM has different treatment of shrubs when cleared. Also relevant - treatment of bare ground - can it have LAI or not if precip occurs etc. Likely different assumptions in POP and ESM1.5.
  • Trendy_v12 in repository. Preparing POP branch for ESM1.5 met forcing. Transparent inputs rather than hard-wired. User provide template for dataset. Inspect netcdf attributes. Pass reader day/year. Applying to met inputs in POP branch. Iterate on template design. Question about dealing with cases that have different variables available e.g. diffusive fraction. One input method but not too restrictive, just CF compliant. Deal with calendar variable. Performance of I/O. Current default will be day by day read - safe option. User instructions for chunking data? Not yet. Get working first. LPG-Guess experience is chunk size, compression, ordering of dimensions makes big difference. Documentation and guidelines for best practice will also be prepared.
  • Who/what/where - NESP context. ESM1.6 for Fast Track but not with CABLE4 - importance of work on understanding how to use LUH data.
  • Refactorisation discussion. Worth doing. Restructure of directories and code modularised in parts. Only logical way forward to do CABLE-POP refactorisation. @jhan started work on this. ‘Plumbing’ issues rather than science code changes. Radiation/albedo can take mostly from CABLE3. Canopy/soilsnow more work. NRI could help as time-consuming - and NRI happy to help. @clairecarouge to think about who.
  • CABLE4 needs 27-tiles. Running OK, almost no performance penalty. Need to check output.
  • Jhan to start on CASA in AM3. Rachel to help. CABLE3 in ESM1.5 rplant bug fix still needs to be addressed. Documented in trac tickets.
  • IAV of carbon fluxes with CABLE3 still needs understanding. Suggest AMIP runs with AM3, ESM1.5 and ESM1.5 with CABLE3 to check GPP.

Action

  • @jhan to package up refactorisation work and organise handover to ACCESS-NRI with @clairecarouge .
  • @Jhan and @RachelLaw to meet April 24th (Aspendale) to look at AM3/CASA and set up of AMIP tests for GPP. [and site-based output].
14

May 2nd

Previous actions status

  • Site-based output from ACCESS - some progress in looking at potential locations for output (following 42 site benchcab sites)
  • CASA implementation, AM3 and for ESM1.6-CABLE3

For discussion

  • LUH2 interpretation for CABLE-POP @Juergen @inh599
  • CASA clean-up, rplant bug fix etc.
  • ESM1.6 for Fast Track

Notes

  • refactorisation work for CABLE-POP. Claire meeting with M. Cuntz to check impact on users.
  • see slides - rangeland quirk. Discussion of what this means and how much should be done as pre-processing. Minimum requirement is consistency and therefore need to include all transitions i.e. don’t neglect primn2past, primn2range,secdn2past,secd2range. Bug found in BIOS implementation. Impact on RECCAP2 budget but very minor.
  • Crops - see slides. No new crops from 1700 but can remove as calculated as a residual. ptoc read in but not used. Commented out about 4 years ago when fixing a difference between serial and parallel versions. Also need to consider BIOME1 factors.
  • crop harvest (and pasture harvest) need to be looked at. Fraction being removed needs to be looked at. If currently 0.9 then too high. Removal once a year, then decays over next year.
  • Reminder of importance of good comments in the code.
  • watch out for spincasa routines etc kept in sync with any changes
  • c13 code. Support of not? Lots of ifdefs.
  • Also how to manage SLI in context of CABLE4 planning.

Action

  • @inh599 and @Juergen to coordinate code fixes across versions ready for next Trendy. Post slides in LUH2 topic.
  • CASA clean-up. Keep logging issues. @RachelLaw to work with NRI team to test and implement.
15

May 16th

Previous actions status

For discussion

  • CASA clean-up and testing
  • New Met input routines for POP-TRENDY

Notes
No meeting May 23rd?
@Juergen started re-coding to deal with land-use bug. Also clean-up to make clearer re hard-wired numbers.
Pause refactorisation of CABLE-POP branch while bug fixes happen in LUH relevant code.
MPI work to start in July.
Offline at ACCESS resolution currently in CABLE-POP branch. Can happen in parallel to MIP work.
@inh599 still has revision of slide pack on LUH quirks on his to-do list, for putting onto Hive.
Query flag primonly - what is it doing? Grid-cell that grassy as potential veg, can’t grow forest on this. Possible ‘wrinkle’ that needs to be checked - plays into crop fraction etc.

@lachlanswhyborn Aim is to move away from file specific I/O to something template based. Configuration options opaque. Pulled these out to be namelist options in cru namelist - how forcing is recycled. @inh599 Configuration options also have impacts elsewhere so need to watch out for those. @Juergen notes should already be accounted for in run script. @lachlanswhyborn Will post in ACCESS resolution topic and where to find branch. Looking for feedback. Aim to also improve run scripts. @inh599 has simplified run script which may be helpful.

ACCESS grid-cell timeseries output as forcing for offline CABLE. Input radiation onto same timestep as other output. Then match to format of single-site file. Need to watch out for ancillary information that is embedded in file. Also initial conditions. Gridinfo is default, then overwritten from restart or forcing file.

For CASA clean-up, document test configurations on Forum. Possibility to save an ACCESS test as a payu configuration.

Action
See notes.

16

June 6th

Previous actions status

For discussion
Juergen can report on progress on the constant crop issue (short presentation) and a few other updates.
Site-based output from ESM1.5 used to run CABLE offline (Rachel)

Notes
Problem with static crop fraction (at pre-industrial levels). 100 site test shows fixed (some transitions were commented out). N American test - crop tile shows mostly higher GPP. Harvest now variable. Checking magnitude for Europe location. Harvest index reduced from 0.9 to 0.5. Seems to give plausible magnitude. Test in full S3 run and compare with last years TRENDY. Next step to look at prim_only attribute in more detail - calculate in run rather than as an input? Various implications of how prim_only currently being used (e.g. doesn’t allow secondary forest to grow on initially non-forest land). Also rewriting of parts of LUC code and pre-processing scripts.

Action

17

June 13th

Previous actions status

For discussion
@RachelLaw update on single-site testing

Notes
@inh599 - looked further at LUH2 use in BIOS2. Lost land? Perhaps more from set-up or analysis rather than how LUH2 used in code. Still checking some transitions. Tweak in pre-processing. Will need to be done separately for ACCESS. Less fundamental problem than initially thought.
@Juergen - full TRENDY run done with crops corrected. To be compared with original.
@lachlanswhyborn - had branch of Trendy branch with more generic met forcing routines. Need to consider what to do with data at other resolutions. Discussion of restart files and how to get data from ACCESS restart into format for CABLE restart.
Is ANTs useful? Check with Davide/Martin. Will also need to be thinking about what can be interpolated and what should be nearest neighbour. Reconfig doesn’t work for tiled variables. Current issue for @Jhan in getting C, N, P pools from ESM1.5 to AM3.
@tammasloughran update on thinning. Wood removed from thinning not going to wood flux. Have added it into code to then check for carbon conservation. Working on the CABLE3 code. Test runs may have plant respiration bug.
@RachelLaw qh/qle now better matched with ACCESS (using t,q, from lowest model level instead of screen level). Seeing impact of snow initialisation. Still to fix. Odd (zero difference) between tair and tscrn for DNL grid-cell for large part of year in offline case only. Needs following up. Reference heights and displacement height may be impacting this.

Action

18

June 20th

Previous actions status

For discussion
@Juergen: short updates on crop fraction, next steps.

Notes
@Juergen - crop fraction update. 2022 crops now better but lower than luh2 - due to assumption about initial biome. Example: potential is assumed 80% grass, 20% forest. Conversion to crop only coming from forest not from grass. Terminology - recommend primary veg instead of primary forest - otherwise misleading. Crop fraction important for harvest but doesn’t change dynamics of model. Test change in small domain.
@Juergen - plans for this years Trendy and then additional simulations
Discussion of patches/cohorts and sensitivity to number. Possibility to test for selection of locations rather than full global. Should we have a pop namelist to pull out number of cohorts/patches etc.? Targeting annual total carbon vs cohort number.
@RachelLaw - not much progress on single site analysis. tscrn does have a bug. @inh599 following up.

Discussion of design of code for CABLE4 in ACCESS e.g. what new code is needed such as mapping between veg types in/out of POP. Probably needs a larger chunk of time that this meeting.

Action

19

June 27th

Previous actions status

For discussion
@RachelLaw - latest on site testing

Notes
GPP/NEE at sites looking better when fix cfrd3 and nslope to ESM1.5 values. Need a method to keep track of parameters. Continuing to document what has been done on Hive-Forum. Will move onto ESM1.5 with CABLE3 as a follow-up step.
CABLE4 code design - when, who, how. Do we all need to be in the same room? Pre-meeting prep on what we want to target. Need to be as specific as possible. Understanding of development workflow. What tests to run. Need to review material Ian has already prepared. Need science and software together so longer conversations required to tease out intent. Some unresolved science but is it a roadblock?
Start with an offline implementation but with the right tiles for ACCESS - to include mapping, variables required.
Need to run a 27 tile case in offline - get rid of hard-wired pft types in CASA. Worth trying test cases with simplified vegetation distributions. Also single-site with multiple tiles. @Juergen has done this with Trendy setup.
Need to start preparing a work plan.

Action
Start a topic ‘issues to address in a workplan’. @RachelLaw
Flowchart slide(s) to describe what aiming at. @inh599
@clairecarouge to run Jul 4 meeting - discuss issues for work plan. Everyone to try and contribute to the list before then.
@clairecarouge to investigate NRI funding if need a face-to-face meeting - or use time after NRI workshop.

20

July 3rd

Previous actions status

For discussion
CABLE4 work plan

Notes
The CABLE4 work plan has been edited during the meeting. @inh599 @Juergen and others to edit/comment/review as required.
The implementation for several pieces of work is still vague and needs more scoping, probably including the offline and coupled codes.
We will need a discussion on test cases required for each step in the work plan to scope the work required for creating the test cases.
There are still questions about resourcing for this activity. This exercise around the work plan has identified some items with fewer science requirements that might be easier to resource.