Skip to content

CTSM FATES integration

Jump to bottom
will wieder edited this page Sep 23, 2021 · 14 revisions

Goals Notes and milestones for CTSM-FATES integration

Sept 2021 conversation on LULCC

See the FATES project board for additional details / updates

Notes:

  • Peter shared complexity of land use transition matrix with LUH2 dataset in CTSM
  • Charlie emphasized need to use area weighted harvest (not mass weighted as currently used in CTSM5)
  • Peter and Dave asked what versions of FATES we want this working with
    • Charlie wanted to keep it generalizable so that patches keep land use labels.
  • Dave wanted to keep this discussion focused on LULCC, but not the crop model in the HLM.
  • Dave confirmed that we need to integrate into current logging scheme implemented in FATES.
  • Jennifer echoed desire to coordinate & work together with ELM and CTSM efforts.
  • Continued previous conversation re. use of stream, vs. modified surface datasets or land use timeseries
    • Erik noted flexibility provided with streams for interpolation and maintainability are advantageous see #1077
    • This may not be best for ELM, since they aren't using nuopc, which would mean using something similar to the land use timeseries at different resolutions.
    • Jennifer said she'd look into this for ELM

Peter's initial focus will be on passing this information from LUH2 into FATES

  • Subsequent efforts on the FATES side will have to be on deciding on what to do with this information in FATES
  • Ryan suggested parallel efforts would be helpful on the FATES side addressing FATES #450
    • Determine how to handle primary and secondary forests as separate columns or patches.
    • How to handle pastures?
    • Considering how to grouping / combining patches onto columns.

June 2021 conversation

What does the CTSM-FATES integration look like?

  • With Adrianna coming on board NCAR can play a bigger role.
  • How do we bring in the different threads (especially development that have been largely external to NCAR) into CTSM?
  • Gordon suggested that improved communication and coordination seems important (without being overwhelmed by meetings).
  • Charlie echoed concerns related to perceived development (outside of NCAR) and governance.
  • Rosie suggested developers that could use a hierarchical global calibration system could be the 'glue' that provides a common structure for us to work under.
  • Peter suggested we have different use cases for different users & applications (e.g. paleo, C4MIP, etc).
  • Dave noted that there's a split between what the land model needs to support (ecology / land only vs. climate / CESM applications).

How do we bridge this intermediate stage for global simulations with a FATES model that's under development?

  • Rosie suggested that FATES-Hydro in SP mode would be a good first use for global simulations
  • Need to build capabilities for:
    • global parameterization for pfts (starting with FATES-SP)
    • transient land use
    • memory optimization & efficiency
    • standardization of history file output

CESM3

  • Bringing in new features for CESM3 will take resources

    • do we bring this into FATES too?

  • Long term goal will be abandon the big-leaf model so we're just supporting FATES

    • This would require a simplified (2 stream radiative transfer model) for FATES (similar to big leaf model).

  • FATES-SP version will be the first step for doing this, so future model development can build off this.

  • Gordon suggested FATES for boreal (or polar) could be a good use case for NCAR to contribute to.

    • Rosie noted that we're ultimately trying to build a global model...
    • Jackie noted that there are communities that would use regional applications of the model (e.g. tropical, boreal, CONUS with biogeography) would can help build the user base of the model.
    • Adrianna emphasized that the regional applications ultimately will serve the global development (with closer goal posts).
    • Charlie asked, why not release a global version (even if it's not very good), to serve as motivation for future work.

  • Can we have a version release of the model with a frozen set of physics of CTSM-FATES that can be vetted & supported, in advance of CESM3.

    • maybe start with I2000 case (since LULCC won't be possible)
    • a paper could focus on CO2 fertilization or biophysical effects of land cover change- questions that SP simulations are good at making.
    • use case with physiology that works to explore differences in CTSM5.1 vs. FATES
    • Then Adrianna can build on this by activating competition modes of the model.

Next steps:

  • Make strategic goals / 3 year plan for FATES development and integration with CTSM / CESM.
  • What does success for Adrianna / NCAR in 3 years?
  • We'll meet monthly, alternate PPE time slots + CLM meetings as needed.
  • down the road we'll communicate this in the newsletter.

v0 integration: Goal Sept 1, 2020

  • Reduced complexity FATES [Rosie Fisher]

    • Create a reduced complexity versions of FATES for testing, development and application including FATES-SP and fixed biogeography (with and without competition).
    • Marked as completed, but maybe more work needs to be done here?

  • LULCC implementation in FATES [Charlie Koven]

    • Beta version of LULCC applied in historical, transient simulations

      • Initially uses area weighted harvest (CLM5 currently uses C mass harvested)
      • Complicated to investigate how biases may influence primary and secondary forest in FATES
      • FATES want to know just the fraction of vegetated land unit for each column that's being harvested (which is directly provided from LUH2).
      • FATES disturbance / harvest happens daily, but harvest data from CTSM provided annually.
        • Currently FATES harvest just happens 1x/year.
      • FATES marks trees that are harvested, surviving trees that are moved to degraded (secondary) land. Additional collateral damage happens from treefall and roadbuilding.
        • LUH2 assumes 30% slash, collateral damage from harvest

    • More work needed to investigate potential issues with forest age classes being tracked in FATES

    • Also questions about using area harvest vs. biomass harvested.

    • Current implementation uses area weights and seems to produce sensible changes in secondary forest area

    • Charlie has questions for Bill and Erik if the implementation is correct.

  • Rebase FATES to CTSM MAIN [Gregory Lemieux]

    • FATES needs to migrate to MAIN, not the released version of CTSM.
    • Need conversations about maintaining code & integration between two large and active development communities

  • Coupled model smoke test [Keith Oleson]

    • Need appropriate compsets [Erik]
    • Initially just run CAM-CTSM case, as chemistry likely won't work.
    • Worth running chemistry, so we can see what else we'll need (e.g. dry deposition).

  • Performance, FATES costs

    • Dave thinks we should have cost estimates for FATES. Rosie notes that we don’t currently do a formal analysis.
    • Greg notes that we note the memcomp and tutcomp warnings in the regression tests only right now. Ryan has done spot checking of cost (notably with PARTEH refactor).
    • Erik notes that the cost estimate is going to be somewhat difficult due to dynamic nature of FATES. Patches and cohorts spawn and merge initially (mostly an issue during spin up from bare ground).
    • Smart load balancing as a way to address performance issues, but Bill notes dynamic load balancing would be challenging.

v1 integration: Goal Jan 1, 2021

  • Code refactoring to allow FATES to run on it's own land unit [Erik]

    • Erik will scope out what this will look like, but the idea would be to have FATES run on seperate land units from big leaf CTSM & Crops to make FATES-Crop simulations possible

  • Running FATES with the crop model

    • requires code refactoring (above)
    • If crops are running with nitrogen, and natural veg without, need to think about how to initialize nitrogen content in newly-initializing crop columns with dynamic land use. Charlie: FATES does have a nitrogen model that could be turned on.
    • Need to consider fire related to LULCC, clearning, agricultural fires, etc (see next task)?

  • Fire & human-fire interactions [Jackie]

    • Lots of additional complexity WRT fire, LULCC & crops.
    • There is a framework to add in population & GDP; Jackie is planning to incorporate them in the next year.
    • May need to add some capabilities for global fire: allow spatially-varying parameters, or have some of the existing parameters depend on other variables
    • Significant discussion on ignition sources related to landuse change & deforestation forest
      • Peter noted roughly 1 of 3 Pg C y-1 fire budget that’s currently handled ‘implicitly’.
    • Jackie is cautiously optimistic based on the range of regions over which this is being tested.

  • Parameterization & Individual PFT performance investigations.

    • Calibrate dynamic vegetation model: Ideally with a relatively simple PFT set. If individual PFTs are performing sensibly, this should be easier.
    • Drought deciduous phenology: This remains challenging, as per CLM5. Some clear structural errors need addressing as well as parametric issues.
    • C4 grass: This is historically over productive in CLM. Anthony Walker made a start at looking at this in MAAT.
    • Shrub introduction: Allow shrubs to tolerate harsher growing conditions v.s allometric limits to height etc. Maybe collaboration with EMERALD/NGEE-arctic/?
    • N.b. This is a place where domain experts can potentially take charge of, e.g. of the testing of a new PFT type. Need to think of a way of making those studies publishable. Sensitivity tests? Regional simulations?
    • Also FATES makes this easier than big-leaf mode in that you can run with one-PFT at a time.
    • Jackie highlighted the importance of coexistence for this target. Noted Polly is working on this.

  • Seed carbon conservation?

  • Shared code base for parameterizations, functions and routines between FATES and CTSM. This should likely live in the FATES repo, but would need to be accessed by CTSM for big-leaf simulations (e.g., phenology & photosynthesis code).

    • Avoid redundancies in code and allow for smoother development in the future
    • Afford opportunities for multi-hypothesis testing in both FATES and big-leaf
    • Requires SE attention on both CTSM and FATES side

  • Test and decide whether HYDRO would be on in default

  • Isotopes

    • Carbon: the functionality is basically in place for this
    • Water: if plant hydraulics are off, FATES doesn’t handle any water pools / fluxes, but just hands conductance back to the host land model. If plant hydraulics are on, then this will need some work
      • Whether to turn on plant hydraulics is still an open question

  • Need for dry deposition to work with FATES. And maybe fire emissions passed to the atmosphere.

  • Benchmarking, additional FATES specific benchmarks: NOTE, there are additional discussion on benchmarks in the workshop meeting notes. This short list is intended highlights priorities that are of greatest utility for global, transient simulations.

    • PFT Distributions
      • Sandy Harrison: kappa statistic, simulated vs. observed statistic (Monserud & Leemans 1992)
      • Charlie: Regarding PFT distributions, part of what they’ve been pushing back on is the idea of canonical PFTs. So separating into trees & grasses is fine, but want to be a little cautious about trying to match canonical PFTs.
    • Abby: idea of comparing distribution of traits rather than PFT distributions per se (e.g., isohydricity / anisohydricity, etc.)
    • Tree height data, tree density data
      • Peter: in CLM5-SP, we specify canopy height from ICESat, could be used for benchmarking instead
      • Lidar-based global canopy structure and height distribution (GEDI) https://gedi.umd.edu/ GEDI science team also has a large database of forest metrics for modeling biomass - they are usually willing to share
      • Emulators of direct satellite observations (NASA MAP project?). This would also allow comparison of forest change after disturbance from LandTrender or other satellite obs.
      • S. S. Saatchi, N. L. Harris, S. Brown, M. Lefsky, E. T. A. Mitchard, W. Salas, B. R. Zutta, W. Buermann, S. L. Lewis, S. Hagen, S. Petrova, L. White, M. Silman, and A. Morel. Benchmark map of forest carbon stocks in tropical regions across three continents. Proceedings of the National Academy of Sciences of the United States of America, 108(24):9899–9904, June 2011.

  • Hillslope interaction? Probably not an initial target. The main challenge here is the interaction between hillslopes and LULCC.

Clone this wiki locally