I suggest we use this time to discuss which CABLE3 options we want to run over the summer as a ‘base-case’ and variants for ESM1.6. Also, what code still needs to be updated to do this.
Here’s a list to get us started:
Medlyn or Leuning (@RachelLaw’s Medlyn parameters need to go into repo: cable_pft_params.F90)
refl/taul (CM2 or ESM1.5, also hard-wired in cable_pft_params.F90)
rootbeta or froot (working on making this switchable, also allowing froot to take pft-dependent value)
snmin = 0.11 (needs to go into repo, on switch?, util/cable_common.F90)
fwsoil_switch ‘standard’ or ‘Haverd2013’ (if Haverd2013, need code fixes)
hydraulic redistribution (on/off, all pfts or just tropics)
AM3 improvements (energy/water balance etc)
@bens - catch-up on Juergen’s cohort sensitivity. May be tweakable parameter to give similar results for lower number of cohorts. May also be worth exploring number of patches. @lachlanswhyborn query nlayer=1 - just a placeholder for future development. Worth keeping.
@inh599 - hearing of interest in land-use change from Sustainable Pathways, LUTO etc. Lots of opportunities to connect work.
Is it worth using this time to talk about managing code across repositories, where to make edits, how to keep in sync etc?
clairecarouge
(Claire Carouge, ACCESS-NRI Land Modelling Team Lead)
43
January 16th
Modifications to vegetation distribution
No decision made on which vegetation distribution to use. Focus on getting a distribution that is closer to LUH3.
@inh599 No spatially varying canopy height in ESM, it is veg. type based. Is it worth implementing in ESM? Canopy height in CABLE is not always compatible with the LAI. ESM already reads a spatial file but the values are PFT-based, could be modified to be varying, at least given a seasonality. Could we constrain the canopy height to the LAI? @tiloz It might be possible to have spatially variable canopy height for forests but might not solve the problem on grass.
@clairecarouge BIOS project is still going on. We are still trying to find the cause of an issue. Hopefully once this issue is fixed, we will be able to move on faster on BIOS.
@inh599 not have a gridinfo file consistent with ESM yet. Problem with ice. @Jhan ssat was zero on ice points when running offline. The soil parameters written in the ESM restart file are not set to the values used in CABLE. @lachlanswhyborn overwrite the values on ice from ESM restart with correct values.
Vegetation distribution. If you want to look at the distribution I’ve prepared, there are plots in slides 25-35 of this file: landuse1850.pdf (1.7 MB)
Left figure on each page is the new vegetation distribution, right plot is the difference from ESM1.5. My aim was to make sure we had sufficient trees to match all of the LUH3.0 forest (=primf+secdf) while also allowing existing ESM1.5 tree fractions to continue in LUH3.0 nonforest (=primn+secdn+range+pastr+urban). I also replaced our existing crop pft (9) with LUH3.0 c3 crop (=c3ann+c3nfx+c3per) and added c4 crop (10) to match LUH3.0 c4 crop (=c4ann + c4per).
Getting the 1850 vegetation distribution matched to LUH3.0 makes mapping changes from 1850 to present relatively straightfoward. The only limitation is that there is no easy way to switch tree types over time e.g. if evergreens were being replaced by deciduous etc.
Questions
In some cases, I had to arbitrarily choose which tree or grass type to add - I just did this by latitude but perhaps could do something more like POP does in its vegetation type determination. Thoughts?
Anyone spot any major issues with the new vegetation distribution? There is significantly more forest for high northern latitudes, but switch from Medlyn to Leuning etc has reduced GPP in northern high latitudes so we probably have scope to increase. Also our atmospheric CO2 seasonality at northern high latitudes was rather low for ESM1.5, so this change might improve that.
Do we want further modify Australia? Do we want to have Australia-specific pft types? Pft 12 and 13 are currently available.
Discussion on what work is needed on the vegetation distribution for LUH3. Mention of C4 crops - Is it possible to use it in ESM? Need to revisit the parametrisation for C4. Need to check the coding for C4 crops and grasses as it hasn’t been used in some time.
Priority is to be able to get a test run with the new distribution so we can assess the impact. Need a restart file.
Should we initialise tiles for all types in all grid cells in the initial restart file and not run through the tiles that are not active. The question is how the landuse change code creates new tiles.
Should we use Australian-specific PFTs? Belinda has 22 PFTs but we only have space for 2 additional PFTs in CABLE. Could we reclassify any of the 22 PFTs together to have a type that covers a large enough space to be worth the work. @Juergen to ask Belinda.
POP runs by @Juergen : we need to review what the next steps are.
Vegetation height: it’s possibly causing instabilities. We may want to change the inputs for the veg. height. To discuss at a future meeting. UKMO does not seem to do better than us, maybe CLM would be useful.
offline CABLE in ACCESS-ESM configuration:@lachlanswhyborn has a completed run using 3 hourly ACCESS-ESM meteorology as output. This simulation needs a more in depth look, notably in CNP cycles, to see if functioning as expected.
updated ESM restart creation: We’re looking to use @RachelLaw’s 3 step process as a means to fill in plausible values for all necessary variables, on all tiles. We need to check that we’ve identified all the variables that need to be filled (a first cut will be for any variable that is tiled) - @tammasloughran posted a list of these on the hive somewhere/land use cahnge code will also assist.
Note: this may be done in two steps - something quick for our current ESM1.6 needs and then something more rubust (with provenance) later. There is interest from the paleo community in a more flexible tool (e.g. for simulations with changed land-sea masks).
POP runs: Discussion identified some further topics to assess in the current runs
Analysis on other fluxes (NPP, respiration) in order to understand where the differences in biomass originate.
Global values of the variation of biomass, GPP etc. with number of cohorts.
(in time permits) how sensitive are the results to a modest changing in number of patches (e.g. 40 from 60).
To follow up: how are the age ranges for the cohorts set up when number of cohorts changes? what’s the role of NPATCH and NPATCH2D (i.e. total and partial disturbance)? Given that only total disturbance is used currently, can the code be simplified. (same question also applies to NLAYER)
Australian PFT@Juergen has had initial discussion with Belinda Medlyn - the suggestion is to review the aggregations devised by de Kauwe et al. (2020) which work on aridity gradients. There will need to be further conversation/decisions around i) which 1 or 2 PFT groupings to add, ii) how to determine all parameter values for the new PFTs, iii) how to determine and cover fractions (including through time) and iv) the location within CABLE PFT table (there are technical advantages to adding these within the trees section but this creates additional work and likely represents a barrier to backwards compatibility).
POP work and review of land use code has not progressed.
Vegetation height ancillary@inh599 (largely driven by AM3 needs) has started looking at expanding/revising the vegetation height ancillary used in coupled models. For AM3 and ESM1.6 this is a required field for all PFTs - currently provided by a monthly climatology (similar to the LAI climatology). For CABLE4-ESM3 we could conceptually replace this field with the POP generation variable - however @Juergen notes that this likely requires additional work to ensure performance/fit for purpose as the current allometry does not seem to produce reasonable values.
The current suggestion is that vegetation height information for the grass, crop, tundra and wetlands be linked to LAI, whereas that for the trees is developed from one of the emerging global canopy height products, e.g. Potatov et al. or Land et al.
ESM restart has been run through both ESM1.5 and ESM1.6 - with no apparent issue. Thank you @lachlanwhyborn. Feedback from @RachelLaw is that the phenology phase information in the restart is likely redundant (i.e. a inheritance from UM/MOSES) whereas CABLE uses the information from the modis phenology file. Still it would be worthwhile checking this.
ESM restart An assessment of the long run climate associated with the new restart needs to be conducted - and include an assessment of seasonality and sea-ice (since that showed up as being sensitive in ESM1.5 land-use and CM2-AMIP runs via changes to clouds)
canopy height Discussions around updating the canopy height. @alexnorton highlighted the existence of several lidar-based global canopy height products. @MartinDix has informed that GC5 comes with an updated canopy heights ancillary (which has a more realistic spatial variability than the GC3 version and the file currently used in ESM1.6). Discussion tended towards i) consolidating codebases now and leave updating canopy height to later, ii) regridding and splitting the GC5 canopy height, rather tan using the lidar-base products (and use these as part of CABLE4 development/calibration).
wood thinning We were reminded that the wood thinning code was implemented into ESM1.5 - it is not clear whether this has been ported to ESM1.6. Certainly some ESM1.6 tests with this capability should be conducted.
Energy balance:ESM1.6 Analysis on current runs has shown that none of the various updates applied has resolved the (relatively) poor energy balance as diagnosed from the outputs. @MartinDix has demonstrated that some of this originates from the 3-hourly nature of the UM radiation scheme, with the remainder likely linked to sublimation. There is a small amount of work necessary to fix the sublimation but we may need to transition our routine assessment of energy balance onto other fields.
CABLE4 timeline We had a short discussion around timeline for development of CABLE4 - this has not notably progressed recently (although some completed tasks will be useful). Discussion tended to a position that we should continue to hold off on work and use the next 4-6 months on ESM1.6 work.
ACCESS->iLAMBNot discussed but should have been. Ahead of formal efforts around tuning etc. we need to establish the work flow necessary to automatically convert ACCESS-ESM1.6 output into a format suitable for input to iLAMB.
To do
check with @tammasloughran around the implementation of the initial harvest pool sizes. Do these need to go into @lachlanswhyborn’s revised restart file? or do they get input via a different route?
More detail on the phenology question - at least in the ESM -
CASA does not pass any phenology variables (so phen%, casabiome%) back to the UM-MOSES via implicit_unpack (so that info cannot get into the restart nor output).
CASA reads in and initialises these variables/parameters via the read of ‘casafile%cnpbiome’ and ‘casafile%phen’ on the first call to CABLE each run (the info in the restart is not getting into CABLE)
so I think the phenology phase is being passed to and from the restart file OK.
clairecarouge
(Claire Carouge, ACCESS-NRI Land Modelling Team Lead)
51
FEbruary 27th
updates/discussion
CABLE4:
Timeline for CABLE4: consensus from leadership meeting the timeline for CABLE4 can slip because we are aiming the MIPs and not the FastTRACK.
Config for CABLE4: we’ll want to use POP + POPLUC. Extend POP capability to more than 3 tiles.
Questions around crop handling and around what else could be in scope.
ESM1.6:
phenology in UM restart done correctly.
LUH3 land cover fractions and wood thinning: the land cover fractions need to come first. Rachel has a first draft of these for the historical period. Tilo: need to get a base map first. Ian: do we need more work in the restart handling work? We need a script that will give reasonable values to the harvest pools in the restart file based on what the map for the start of historical is. Tammas to send script to Lachlan for addition to the restart handling script.
GPP from LUH3 run: Same order of magnitude as released ESM1.5. Need to look more at the variability and variation across PFTs
ESM WG meeting at 1pm where Pearse will present results from ESM1.6 long simulation.
clairecarouge
(Claire Carouge, ACCESS-NRI Land Modelling Team Lead)
52
March 6th
Apologies: Juergen
Direct work on CABLE4 still paused because of lack of resources.
Work on BLAZE needs to be done in the next couple of weeks.
POP namelist work on going.
ESM1.6:
Ian: an ESM-AMIP run has highlighted issues with the veg. distribution map. And we have holes in initial values for some grid cells in the restart.
Rachel investigated some problems in the veg. distribution, has identified a problem in her workflow. Working on a solution.
We should be initialising the values for all the PFTs at all grid cells even if the tile fraction is too small to account for anything. Lachlan says that all PFTs that existed in 1850 should have had initial values at all grid cells. Unclear if the problem in diagnostics
Let’s fix the veg. distribution (i.e. negative fractions) first and then see if there is still an issue.
We should still be going with an update of the version of ESM1.6 to bring in updates for ocean, cice and land together. There is a question on which version of each component to use, possibility of running several options in parallel to check the effect.
Analysis of ESM output in ILAMB:
Progress done. Rhaegar successfully produced 18 variables, still has to deal with tasmax and tasmin. Some issue on the DRS but nothing difficult on that side. Need to check what Rhaeger has done. Alex to help on this.
Wood thinning:
Harvest pool initialising, looking at LUH3 data. Rachel came up with a 22 PgC carbon pool. In Tammas’s script, he used a 19 PgC of total carbon pool. Same order of magnitude. The distribution of the initialised carbon pool should not matter for the results of the simulation.
Good to do a historical run with CMIP6 forcing relatively soon to help with this type of decision.
Initialisation of gridcells:
We should try initialising all tiles in all gridcells (likely done) and run all tiles as well (probably not done). See the impact on the model performance and then decide on what we should do. Except ice, lake and wetlands are invariant so no need to run them everywhere (and we can’t sustain ice in the tropics). Try to run with everything except the obvious exclusions, this can be done by setting the fraction to 0.
Vegetation types for Australia:
We need to get back to the question of introducing Australian veg types.
Actions:
Rachel working on veg distribution
Claire, Alex and Rhaegar to continue work on ILAMB
Keep mixed CABLE4/ESM weekly meetings.
clairecarouge
(Claire Carouge, ACCESS-NRI Land Modelling Team Lead)
53
March 13th
Apologies: Ian
Land use change with the new vegetation distribution
ESM1.6 is currently crashing with land-use change but working with old ESM1.5 land-use files. Problem with the files based on LUH3.
Tilo: can we use the files for the spin-up or should we wait to incorporate it? The new files work in pre-industrial but there is a risk that changes for the historical period might impact the 1850 distribution. Rachel: Pretty confident the 1850 distribution is good enough to incorporate in spin-up now.
In CMIP6, we simulate all the tiles that cover 0.1/land fraction at any point in historical and scenarios. But in the new files, we now have tiles that come and go.
Question about the labile carbon and how it works in CABLE. It does not do anything functionally so it is left alone in the LUC (transfer of labile carbon from a tile losing area to the labile carbon of a tile gaining area).
The labile pool can be used as an indicator of the nutrients limitation. The nutrients are transferred to the new tile using the same pathways as the carbon so it makes sense for the labile pool to be transferred as well.
Discussion on wood harvest pool. Name isn’t intuitive, talking of wood product pools might be better. Tammas’s work shows that the initialisation of these pools is important for the total NBP of the simulation.
For LUH3, do we want to use the same cut-off of 0.1/land fraction and simulate all these tiles? The other way around would be to add cases in the code for newly appearing tiles. The second case is attractive because it would cover the case of climate-dependent tiles where new tiles appear dependent of the climate which means we don’t know what tiles might appear.
Conclusion: worth having a look at the code to see if we could have tiles appearing and disappearing.
Alex to look at the LUH3 processing code from Rachel. Possibly rewrite to Python to make it more user friendly.
Australian PFTs
We have 2 spare PFTs we could use to introduce Australian-specific PFTs. Juergen to bring this back to discuss soon.