Alexey-Ben meeting 2018-12-20

[ashiklom]

AGU recap

• Mike said he got a bunch of questions about trait covariances in PEcAn's meta-analysis. This will be my next priority for PEcAn development, after we make some significant progress on FoRTE.
• Similarly, this means it would be good to get my multivariate meta-analysis paper submitted (and hopefully published!) soon.
• ED RTM project
  • Mike and Shawn were pleased with the results I presented -- they were an improvement over the results from my dissertation. Mike felt this work was approaching publication readiness.
  • Tossed around a few ideas about ED RTM with Tristan Quaife (U. Reading; works on JULES model; expert on radiative transfer, especially in ecosystem models). He has a new two-stream model implementation that could potentially be an alternative to ED's current one. He also has a snow albedo model, a linearized BRDF model (possibly necessary to convert ED RTM's exitance/hemispherical reflectance output to directional reflectance, which is what remote sensing actually observes), and an alternative implementation of a two-stream model with an analytical solution that could be interesting to compare against ED's.
  • Mike and Shawn also really want to demonstrate at least a single time-step of state data assimilation using the ED RTM approach (which was the final deliverable of their NASA TE grant that paid my grad school salary before I got my NASA ESSF). My main responsibility here is making sure the ED soil-leaf-canopy RTM works properly (i.e. is not systematically biased) -- some combination of Istem Fer and/or Ann Raiho (and/or others) would take the lead on developing and performing the assimilation step, though I might provide guideance or other help.
  • Mike, Shawn, and I (and probably Istem, Ann, and Mike and Shawn's current postdocs) will probably have a call about this sometime in January/February to discuss this.
• Had a long conversation with Shawn Serbin and Jin Wu (his current postdoc; about to start a faculty job somewhere in China) about possible future collaborations. Most promising idea was to use our Bayesian canopy RTM inversion methods (with some tweaks) to generate global map of leaf/canopy traits and their uncertainties. This would definitely require writing a proposal, probably justified by a small-scale proof-of-concept analysis. My thinking was that the proof-of-concept (map traits for a particular AVIRIS flight, a particular Landsat path-row, etc.) could be a good LDRD proposal at some point.
• Another ED2 community meeting will probably happen sometime soon, probably in Boston (hosted by Paul Moorcroft's lab at Harvard). I think this would be a useful meeting for me to attend. I asked his PhD student Miriam Johnston (doing something very similar to me with ED RTM, but for thermal) to keep me posted.

Goals for this week

• FoRTE
  • Start a defensible ED2 100-year ensemble control run at UMBS
  • Set up PFTs for UMBS
  • Create a data-driven distribution for PFT traits/parameters (based on TRY data, my RTM-based calibration, multivariate meta-analysis)
    • May require a bit of hacking around PEcAn's existing code so I can draw parameters jointly (i.e. account for their covariance)
  • FoRTE first paper outline
  • Questions and objectives
  • Methods
• Hector
  • Background reading
  • ~Create and try running one permafrost emissions scenario (as proof-of-concept, and just to make sure I understand how to do it)~ Update: Done with one set of emissions scenarios. Code for this is at https://github.com/ashiklom/hector_permafrost_emit.
• Other
  • ~Peer review for Geophysical Research Letters (requested a few weeks ago, but I asked for an extension through this Friday because of AGU)~ Update: Completed on Monday, 12/17
  • ~(Time-permitting) Submit multivariate trait analysis.~ Update: Basically done on Wednesday, 12/19. Can hit submit as soon as internal review is complete.

Other notes

• First-authored papers in the pipeline
  1. Multivariate TRY analysis (dissertation chapter 1; in prep for Ecology Letters)
     • Already published as pre-print. Close to submission ready (estimate 0.5-1 days work) for Ecology Letters.
     • Already formatted for Ecology Letters. Only changes I need to make are (1) flip the order of subsections in the results (and tweak text to make sure it still makes sense) and (2) flesh out the concluding paragraph. Mike had a few other ideas about things to add to the results/discussion, but this never came up in previous reviews so, in the interest of time, I'm inclined to ignore them for now and revisit if requested by reviewers.
     • Relevant to FoRTE (2) because trait covariance can provide additional parameter constraint. Can cite the preprint for initial submission.
  2. ED2 parameter and structural uncertainty analysis at UMBS (FoRTE year 1 paper)
     • Outline stage
     • Aim to submit by end of May 2019?
  3. Hector permafrost emissions paper
     • Pre-outline stage
     • Aim to submit by end of May 2019?
  4. ED RTM calibration (AGU presentation; dissertation chapter 4)
     • Complete rough draft done (dissertation chapter) but needs substantial revision, and to be updated with latest results. Analyses need to be cleaned up. Estimate 3-4 weeks of work.
     • Relevant to FoRTE (2) because it can help constrain canopy strucutre parameters (which generally lack good trait data) for ED. Can always cite as in prep for initial submission.
     • Aim to submit by end of May 2019?
  5. Global(ish) leaf spectra analysis using PROSPECT (dissertation chapter 3)
     • Decent draft in dissertation. Needs to be reframed (per committee comments), but content is fairly solid. Estimate 2-3 weeks of work.
     • Relevant to FoRTE (2) because it provides an informative prior for leaf optical properties. Can cite as in prep for initial submission.
  6. FoRTE year 1 experimental response paper (and prediction)
     • Idea stage
     • Goal: Generate (pre-register?) an ED2 prediction of the experimental responses before we start observing them
     • Reach goal: Generate ED2 predictions for different killing scenarios before we do the experiment (to inform which trees we kill)
     • Submit by end of August 2019?
     • This paper is "Here are our predictions for the FoRTE project." Then, follow up with papers as updates about the status of our predictions for subsequent years. Publication of model forecasts is the ultimate form of pre-registration.
  7. Impact of multivariate constraint on model uncertainty projections
     • Pre-idea stage
     • Paper to accompany implementation of multivariate meta-analysis in PEcAn.
• FoRTE paper idea -- Maybe re-frame / exend the scope to include drivers of uncertainty in ED2 predictions:
  • Parameteric uncertainty (by running ensembles over trait uncertainty)
  • Consider the role of trait covariance? E.g. Here are predictions with vs. without accounting for covariance. But, this might be better off as a separate paper when I fully integrate the multivariate meta-analysis into PEcAn
  • Structural uncertainty (our factorial ED "scheme" experiment)
  • Besides the processes we are already thinking about, an interesting interaction between structural and parameteric uncertainty is PFT definition. E.g. We could run one ED2 ensemble with a standard PFT specification ((early, mid, late) x (hardwood, conifer)), and another with more finely resolved PFTs, maybe even down to the species level (e.g. aspen, birch, red oak, red maple, pine). This would increase the parameteric uncertainty somewhat, and could expose problems related to limited similarity. The latter opens up an interesting discussion about real, significant differences between species that are not represented by ED (e.g. clonal reproduction of aspen but not birch, different ectomycorrhizal associations)
  • Driver uncertainty (i.e. meteorology data)?
  • I saw several AGU talks (including one from Gordon Bonan) that highlighted how different various meteorological reanalysis products are and how that can affect predictions about various processes (Bonan showed it's an important source of uncertainty for CLM projections)
  • One challenge is finding climate data products that go back to 1900 that have all the variables ED needs (in my experience, downwelling radiation is a variable that is often missing) and have a good-enough temporal resolution (at least 6-hourly). If there's at least one other one that works more-or-less off the shelf, this shouldn't be hard to do. An alternative is to try to use some stats to derive some of the other variables and/or downscale, but that's complex-enough to warrant its own paper.
    • Another reason to punt on this for the first paper is that adding another factor to our factorial experiment represents a significant additional computational burden.

[ashiklom]

Also, just spoke to Mina and it sounds like we should briefly discuss my billing targets for the next year. She mentioned that you may want me to reduce FoRTE and add some LDRD-related work...?

[bpbond]

Useful, and interesting–thanks. A few other notes, esp. wrt papers:

• Re driver data uncertainty, yes, I've seen this a bunch too. As part of E3SM work, I have site- and global-scale runs using two different BGC modules and two different driving datasets (GSWP3 and HadCRUT), i.e. a structural versus driver data uncertainty test. I'm supposed to lead a paper on this in 2019, and your expertise would be useful and welcome.
• "Signal emerging from noise" paper–what you and Jinshi and I talked about a bit, based on my Friday talk. There are a number of different ways this could go, some more directly relevant to your areas of expertise and some less, but regardless it would be great to have your involvement if of interest.
• FoRTE paper on structural/parametric uncertainty–seems promising.
• Glad Mike and Shawn were pleased with development of ED RTM work! 👏 Aiming to submit this within 4-6 months seems good.

See you tomorrow, B