CABLE-POP runs - global parallel setup (without MPI) and sensitivity tests

The aim is to set up CABLE-POPLUC so that it can be run by a wider group of users, for example by ensuring necessary forcing and input files are in a shared space (under rp23). This set-up can then be used for sensitivity testing etc. The focus will be on the global set-up as used for Trendy which runs multiple parallel jobs (rather than MPI).

  1. @Juergen provide demo of how to run - Done. A recording is available.
  2. @clairecarouge to define directory structure for forcing/input files and copy files.
  3. @inh599 to check dependencies
  4. Write a quick start guide / cheat sheet
  5. Test runs with POP turned on/off
  6. Test different number of cohorts (default 20) and patches (default 60). Plot the results (e.g. global NBP) and see at what point they converge to our satisfaction.
  7. Try and run CABLE-POP with prescribed carbon stocks. This involves getting familiar with the POP restart files and would extend (6) to check whether the size of the POP restart files is practical.

@RachelLaw I’ve edited to add check boxes in the list to improve on readability. To do it simply add [ ] after the list marker:

- [ ]
1. [ ]  

Then one just needs to click on a box to mark it done. Not sure who has the right to do so on a normal post so I’ve also changed your post to a wiki post to make it editable.

Here is the directory structure chosen:

├── data/
   └── no_provenance/
       ├── casa/
       ├── gridinfo/
       ├── landmask/
       ├── luc/ 
       ├── met_forcings/
       └── parameters/

The no_provenance/ directory contains data where the provenance is unknown or not yet fully described.

Each directory and sub-directory contains a README that explains what the directory contains and a description of the data within.

Now I will copy the files for the POP experiment. I will update here when the copy is finished and the demo script runs from this data.

The TRENDY files I have access to are now copied under /g/data/rp23/no_provenance.

I have sent an email to Jürgen to check, add information and copy the files I don’t have access to where I can see them.

Since it is missing some files, I haven’t been able to test a modified version of the scripts that uses these files. I will do so in the second half of January.

The recording of Juergen’s demo is available, but where? Can we please put a link to it in the first post here?

Dependencies from CABLE-POP

These notes from @inh599 email.

A quick look through the svn repo (CABLE-POP_TRENDY, in particular anything to do with v12) indicates that we may (if different from files in AUX) need to manage

  • An R packaged library (on x45 – this appears to be using the R raster and ncdf4 libraries)
  • A python package (on x45)

The R package appears to be used exclusively in the first ‘splitting the land up into nruns subregions’ step. The python package is for the merge of the subregions into 1 output file. The simplest way to avoid having to use these packages may be for @inh599 /@clairecarouge /@SeanBryan51 to run the creation bit (we have access to this) and then place the resultant landmasks somewhere on rp23 and then don’t recreate each time.

Ancillaries and other data inputs

  • CRUJRA daily meteorology (on /x45/CRUJRA2023/daily_1deg/ - this includes annual and monthly CO2 information)
  • LUH2 preprocessed land-use transition files (on x45 - x45/LUH2/GCB_2023/1deg/EXTRACT/)
  • A somewhat random gridinfo file associated with the runs – this is sitting at /g/data/pr09/TRENDY_v12/AUX/
  • the modis_phenology text file associated with the runs - also sitting at /g/data/pr09/TRENDY_v12/AUX/
  • the glob_ipsl landmask sitting on the branch (under scripts/aux/landmasks/)
  • (possibly) the trendy.grid.1deg sitting on the branch (under /scripts/aux/)
  • the def_soil and def_veg parameter text files and the pftlookup.csv sitting in on the branch (under /params/v12/)

There are other files indicated in the run scripts that could conceivably be needed – especially if we attempt to activate BLAZE. However my reading is that these are for science options that we are unlikely to/will not use (e.g. isotopes) or for particular sites – so can be dealt with later if needed.

Note the need to access stuff on pr09!! I can’t see this directory so I suspect this is @Juergen’s UWS project space and there are copies of these files in the /scripts/aux/ directory on the branch – @Juergen , please confirm.

In order to run anything ourselves we will have

  • Modify the main script to point to the relevant places
  • Modify the script – I think it’s just the PBS stuff at the top but this is REALLY important.
  • Ensure that the serial cable executable has been created in the relevant place (as indicated by the run_TRENDY script)

For initial testing, @inh599 suggests running with experiment=”S2”, nruns=100, run_model=1, merge_results=0 and modify the script to only run the first of the generated land_masks (section 2.2 in the run_TRENDY script).

The svn repo of CABLE-POP_TRENDY was last updated 3 months ago so it is likely that it (and the git copy) are up to date.

Some update about running the TRENDY configuration.

I have now what seems to be a python version of the R script to create the land mask (split_landmask.R). I have to add the capability to handle regional runs but it seems to work for global runs. And add the capability to handle input land masks with 0 for ocean (instead of missing values) so that it has the same range of capabilities as the R script.

I still need to run the R version to check we get the same results at the end of both versions.

LUH2_CABLE-POP.pptx (1.0 MB)

The presentation from 15 Feb 2024: LUH2 and CABLE-POP and how the agricultural pool is calculated.