FoRTE team meeting 2018-11-12

[ashiklom]

First modeling paper -- establishing the baseline

• Research questions
  • Boring: Can models reproduce observed forest growth following stand-replacing disturbance?
    • Pro: Much easier to do; can just stick to models that are easy to run.
    • Con: Harder to draw meaningful ecological conclusions from the results.
  • More interesting: Why do some models get some processes right or wrong, and what does that tell us about the importance of the underlying processes?
    • Con: Harder to do. Requires careful thought about model selection so we can isolate effects of specific different process representations.
    • Pro: Results could be much more ecologically relevant.
    • Follow the general framework of Fisher et al. 2017, but actually run the models! All of their conclusions are largely speculative, or pulled from isolated examples in the literature (different sites, compositions, biomes, etc.). Our study would be novel by running all models using (as much as possible) the same inputs and running conditions.
• Methods
  • "Observed growth" is broadly defined. Typical modeling studies set a low bar -- reproducing absolute carbon fluxes (flux tower), maybe radial growth/aboveground bionmass. We can do better!
    • Evaluate against multiple observations at the same time -- radial growth and NPP and LAI and composition and surface reflectance (Landsat) (others? Demographic rates would be really cool)
    • Beyond absolute values -- phenology and other "dynamic" variables (e.g. growing season length, max annual NPP), interannual variability (e.g. sensitivity modes of climate variability), spectral analyses (e.g. Fourier decomposion, wavelets), demographic trajectories
  • Model selection
    • Based on processes we expect to be important in driving succession (or, put another way, based on alternative hypotheses about the processes underlying succession).
      • Competition for light
        • Alternative representations of radiative transfer (Fisher et al. 2017)
        • Alternative representations of photosynthesis (Rogers et al. 2017)
        • Trait plasticity in sun vs. shade (Keenan & Niinemets 2016)
      • Competition for nitrogen -- dynamic nitrogen cycle
    • So...which models?
      • ED2 -- Cohort-based model (groups individuals of same PFT and age/height); several different radiative transfer schemes (can run in big-leaf mode, two-stream, or experimental multiple-scatter; also, optional finite canopy radius model); can toggle N cycle; can toggle trait plasticity (experimental). Subdaily timestep. Fully integrated with PEcAn. Code is open access.
      • JULES -- Same general RTM approach as ED (separate two-stream solution for each canopy layer), so a useful point of comparison. Code is not open access.
      • LPJ-GUESS -- Cohort-based, like ED. Similar RTM, but with more competition among cohorts with similar heights? My understanding of its RTM is fuzzy. Has dynamic N cycle as of version 3.0. Daily timestep.
      • Important downside: Code is not open access. You have to email a guy to get it. I'm tempted to omit the model from the analysis and explicitly call them out on this in the discussion.
    • CLM/FATES -- Several different forms to work with, which makes it useful for testing alternative hypotheses. Code for CLM (CTSM) and FATES is on GitHub. Experimental support in PEcAn.
    • Maybe individually-based models like SEIB-DGVM, SIBBORK, and LINKAGES. Only LINKAGES is in PEcAn.
  • Simulation experiments
    • Use PEcAn to run models using the same data drivers/parameters (same climate drivers, same data source for parameters, same PFT-definitions, to the extent we can) from bare ground at UMBS for ~100 years (1900 to present-ish)
    • Compare against all the site data we can.
• Immediate Next steps
  • Get any version of ED2 to run without crashing from bare ground at UMBS (1 week)
  • Production-level runs (properly calibrated ensemble) for ED2 at UMBS (2-3 weeks)
  • Try running ED2 with different "switches" (RTM type, with/without N cycle, finite canopy radius, trait plasticity) (3-4 weeks)
  • Try runs with other models (??? weeks)
• Longer-term goals
  • (Reach) Disturbance treatment runs of ED2 done in time to inform treatment
  • Paper submitted late Spring/early Summer

Second modeling paper -- Response to disturbance

• Direct follow up to first paper: If we feed the models a disturbance treatment, how do they respond?

Other ideas (down the line)

• Interrogating the model development process -- E.g. imagine a plot of Git commit on X-axis vs. model performance benchmark(s)

[bpbond]

I'm tempted to omit the model from the analysis and explicitly call them out on this in the discussion.

Yes. 👍

[bpbond]

Would this paper lay down future predictions as well? E.g. as a sort of preregistration.

[bpbond]

Interrogating the model development process -- E.g. imagine a plot of Git commit on X-axis vs. model performance benchmark(s)

Again, yes. That would be really cool!

[cmgough]

Would be interesting to contextualize/compare with a well-known ecological theories on the mechanisms underpinning long-term C cycling processes, e.g., bormann likens, odum

[cmgough]

Factors that may affect long-term NEP, and could be manipulated in a model (e.g., ED): radiative transfer trait plasticity N cycle plant hydrodynamics *gs