Alexey-Ben meeting 2018-10-30

[ashiklom]

Last week's goals:

• Outputs from SIPNET 100 year runs -- Downloaded and calculated annual means. Posted on Slack.
  • Proof of concept: We can easily do 100 year runs at UMBS, including getting at least one climate data product. And we can do it all with PEcAn.
  • Now, need to look more carefully at how we run the models, especially:
  • PFT definition -- what species go into it? What trait data are available? For traits, I can probably constrain values more by adding TRY data.
  • Initial condition -- I currently have no idea where SIPNET gets this. Will need to figure that out.
  • Parameter uncertainty -- With PEcAn, should be straightforward to run ensembles across parameter ranges. I should actually do this!
  • Driver uncertainty -- Good idea to run SIPNET using several different met products (should be at least a few reanalysis products available across this range). The met products themselves should also be compared to local tower observations.
• Run ED at UMBS -- Can successfully start runs with PEcAn, but model crashes early in the first spring because of numerical integration errors.
  • Mike had a few tips, but said ED is just numerically unstable at low LAI and was pessimistic about bare-ground runs.
  • ...but, I emailed a contact in the Moorcroft lab (Xiangtao Xu, postdoc), and he said he's done lots of bare-ground runs with ED in temperate forests in the past. He had some suggestions, and offered to look at my input files to help debug.
  • Next step -- look closely at PEcAn-generated parameter files. Might be a parameterization/configuration error. If everything seems to be in order, send some input files to Xiangtao to see if he sees anything.
• Run SiPNET and ED at SERC -- Didn't get to, but shouldn't be too hard.
  • Unless it's needed for something else, I'd like to take SERC off my to-do list for the near future. Any time I would have invested in SERC specifically, I would rather invest in PEcAn more generally (both code hardening and adding models). Once it's a bit more solid, I can help set Ben and Stephanie (and others?) up with it so they can do the runs they need themselves.
• Background reading -- Spent a lot of time this week on general background reading, particularly catching up on Google Scholar alerts and journal RSS feed updates from last few months.
  • No updates to FoRTE first paper outline. Need to start regularly adding to this moving forward.
• ED RTM -- Completed translation of ED two-stream RTM into R (link), but it's giving some crazy output.
  • Next step -- Run ED RTM natively with a few extra "print" statements to have some reference variables for comparison. Look in particular at RTM inputs. I may be giving bad inputs for incident solar radiation (ED does a lot of extra normalization of various fluxes; odds are good I haven't correctly normalized something).
• Hector -- Found a few papers with permafrost C emission trajectories, but haven't done anything with them yet, and need to give them a closer look. See also Hector meeting notes.
• General PEcAn development
  • Added THREDDS data server functionality. Makes it easier to access PEcAn run outputs remotely, including enabling remote subsetting. See PEcAn PR #2164.
  • 90% of the way there to a PEcAn R API, where runs are submitted to a remote Dockerized instance directly from a local R session. Once fully functional, this will make it much easier to run integration tests and generally to run models (by avoiding need to manually click through web interface each time).

Goals for this week:

• FoRTE
  • Try to get ED bare ground runs working.
  • Try to get more reasonable SiPNET runs.
  • Background reading on SiPNET
  • Flesh out the FoRTE paper, focusing on introduction and methods
  • (Time-permitting) Dockerize and try to run a few other PEcAn models at UMBS
• Hector
  • Review of permafrost carbon literature
  • Reach out to Kevin Schaeffer and co-authors about output from their 2011 paper.
  • (Time-permitting) Digitize the plots from that paper myself, and try to set up Hector runs where those emissions are included.
• ED RTM work (for AGU)
  • Debug R implementation.
  • If this proves too difficult or time-consuming, switch back to original approach. (Time-permitting) Do some basic profiling of executable to see where I can potentially skip time-consuming steps.
  • (Time-permitting) Parameter sensitivity analysis, and test Bayesian calibration.
• (Time-permitting) PEcAn
  • Clean up, document, and test PEcAn API code for a pull request

[ashiklom]

Alexey's notes from meeting:

• If adding more models to FoRTE analysis, consider LPJ-Guess
• Alexey will eventually set up two PEcAn cloud nodes -- one more stable, one more experimental
• Consider initializing ED from observed non-bare ground state one or two years later, if bare ground runs don't pan out
• Alexey's AGU work should be top priority
• Ben will work on adding Biome-BGC to PEcAn in January. Alexey thinks it should be pretty quick.
• Fine to take SERC off Alexey's to-do list. Teaching others how to do runs at SERC is a better idea. Postdocs interviewed this week may also be interested in using PEcAn.
  • Also, Lisa Haber (VCU, FoRTE project) may be interested in doing a modeling chapter for her PhD. We should keep in mind a list of low-hanging fruit PEcAn-enabled modeling analyses that would be good candidates for this and similar efforts.
• Hector can be lower priority on Alexey's to-do list until AGU
• FoRTE modeling -- look into observational data for validation
  • Direct comparison of absolute values
  • More advanced comparisons of trends, distributions, correlations, etc.
  • Maybe discuss on Thursday...but main focus is Jeff's kill map
• Consider the University of Delaware temperature and precipitation climate data as a model driver?
  • Only has air temperature and precipitation -- many models need more than that. But may be a useful point of comparison against other met drivers, or for other work.

[bpbond]

• Lisa's first committee meeting is November 12, and I'll know more about their plans after that.