Alexey-Ben meeting 2019-01-22

[ashiklom]

Updates from last week

• Hector -- Spent basically all of my time on this
  • Successful meeting with Corinne and BK. Pretty good idea of how to proceed with paper.
  • Add INI read/write to Hector package (pull request)
  • Fix Hector biomes (branch)
  • Several hours reading; added some notes to paper outline
• FoRTE -- No major updates; spent all of my time on Hector
• Completed GCB review (free color panel!)

Goals for this week

• FoRTE: Target = 26 hours
  • Create UMBS-relevant PFTs in PEcAn database (4 hours)
  • Add TRY data for UMBS species to constrain parameters (4 hours)
  • Look into initial conditions, especially soil (4 hours)
  • Try running ED ensembles formally through PEcAn (4 hours)
    • Make sure parameters are passed correctly, outputs are being read, etc.
  • FoRTE paper reading and outline (6 hours)
  • PEcAn debugging overflow (4 hours)
• Hector: Target = 8 hours
  • Fix biome-specific warming tag; other debugging (5 hours)
  • Background reading and paper outline (3 hours)
  • Start building database of literature values for parameters, calibration targets, etc. (2 hours)
    • The more defensible parameter inputs are, the better. Would be nice to base parameters on formal statistical meta-analysis, a la PEcAn
• Misc: Target = 6 hours
  • Ben meeting + prep (2 hours)
  • Rebecca meeting (1 hour)
  • Other overflow (email, etc.) (3 hours)

Discussion topics, etc.

• Time allocation between FoRTE and Hector
• Upcoming FoRTE deadlines?
  • UMBS winter meeting (2/7) -- do you need any slides, results, etc. from me?
  • Richmond meeting (2/27)
    • Paper 1 results:
      • Minimum: ED2 "default" case results
      • Minimum: Detailed outline of paper introduction and methods, expected results, and topics for discussion section
      • Reach: Results for a few alternative cases
    • Reach goal: Runs for 1-2 alternative formulations done as well
    • Programming/data processing tutorial
• Possible (FoRTE-relevant) PEcAn development (at some point in the future)
  • In general, my thinking here was to get the "default" runs done first, without doing any of this. Then, we can take a look at the outputs, consider how reasonable they are, and, if things look good enough, proceed with the experiment without worrying too much about this stuff.
  • Multivariate meta-analysis -- should add additional (arguably, more realistic) parameter constraint
    • "Hacky" approach -- Perform meta-analysis outside PEcAn, and pass resulting parameter distributions to PEcAn. Need to make sure PEcAn is sampling (or at least can sample) parameter distributions jointly.
      • I can probably do the independent multivariate meta-analysis anyway and just compare the results between mine and PEcAn's. If they're similar enough, we can just proceed with the current PEcAn approach and punt on this.
    • Formal integration is more involved. Main problem is that my current approach assumes (log-)normal priors, but many PEcAn parameters are not normally distributed (many are beta, some are gamma).
  • Other met products -- if we wanted to consider driver uncertainty; also, I have some reservations about the CRU-NCEP product, but it's the only one that goes back to 1900. Might be good to consider at least 1-2 alternatives.
    • A good test of whether or not we need to worry about this is to do some ED runs starting in 1980 (or so) using several different met products but that are parameterized identically. Then, we can look at the sensitivity of ED initial condition runs to climate driver uncertainty. If it's reasonably low, we can punt on this.
    • ED requires a lot of met variables, at high temporal resolution (6 hourly or better, for reliable results). Not a lot of met products satisfy this. In PEcAn, I think only CRU-NCEP and NARR.
    • ERA5 (posted on Slack) has a lot of variables, with great temporal resolution. But only goes back to 1979 right now. Eventually, will go back to 1950.
    • NARR also has everything, but only goes back to late 70's.
    • Hacky solution (sort-of based on how CRU-NCEP works) is just to loop/sample met prior to product start date (i.e. assume stable-ish climate before 1979).
    • More sophisticated, but significantly harder, solution: Downscale these met products against lower-resolution ones that go further back (e.g. CMIP5?). PEcAn already has some utilities for downscaling. Colleagues at JGCRI could be useful for helping develop this. Could maybe become its own paper.

[bpbond]

• GSWP3? https://github.com/PecanProject/pecan/issues/1871

[ashiklom]

Meeting notes

• Hector -- also, work on fleshing out the analysis plan/outline over next few weeks.
• UMBS winter meeting
  • Ben will show a few slides, but nothing necessarily new
  • See how much I get done over next few weeks
  • Maybe put together a conceptual slide in 2 weeks (2/4). If I have preliminary results, consider adding a figure
• PEcAn development -- Get a complete pipeline first. Then, we can think about adding stuff
  • Met downscaling -- could definitely leverage JGCRI expertise. Maybe even hire someone (post-bac?)
  • Met -- also consider GSWP3; used by many ESMs; may give better results than other products
• I have some funding (~150 hours) on PREMIS SERC project
  • At some point, maybe do something with the drone
  • Can go down to SERC with Ben and Stephanie sometime, when I feel like getting out of the office
• Open to mentorship/teaching opportunities

[serbinsh]

@bpbond @ashiklom Note my comment on GSWP3 on the pecan issue, will update when I know more. The current issue is there is not a source for the raw data on a public FTP or similar so that needs to be sorted out first.

[bpbond]

Very cool @serbinsh – thanks for pursuing!