The trk variable for the .pss files is set at 1, I think, but maybe it should be different for different patches/ecosystems? I don't really know how it works

https://github.com/EDmodel/ED2/wiki/Initial-conditions#files-types-and-formats-for-nlied_init_mode6

Increase runs to more sites besides N/S transect to get data to train a gridded model like in cct-datascience/model-vignettes#89.

And possibly run tests with old version to compare.

The history files from Mike were created with ED2.1.0 at some point (not sure exactly when) and are meant to work with IED_INIT_MODE=3, which is now deprecated in favor of IED_INIT_MODE=6. The .pss files have a different format with the new mode. Also, the .css files contain some PFTs that have been replaced (12, 13, 14), so these need to be replaced with some modern PFTs. Other changes might be necessary also.

Accidentally committed a large file on #19, now can't push to GitHub. Untracked large file, but still can't push. Can I squash commits or something??

Some jobs apparently don't start when using the run.sh script. The jobid is NULL and the workflow. Not sure if there is a pattern to this or if it's stochastic.

It seems like we are either getting close to the compute limit, or there's some other limit about the number of simultaneous jobs/nodes/cores.

(puma) [ericrscott@wentletrap ~]$ squeue -u ericrscott
             JOBID PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)
           6753242  standard PEcAn-SA ericrsco PD       0:00      2 (AssocGrpCPUMinutesLimit)
           6753209  standard PEcAn-SA ericrsco  R   17:28:25      2 r1u11n2,r4u29n2
           6753173  standard PEcAn-SA ericrsco  R   17:40:42      2 r1u26n1,r2u32n1
           6753207  standard PEcAn-SA ericrsco  R   17:34:33      2 r4u08n2,r4u13n2

Currently PEcAn overwrites ED2 defaults with its own defaults resulting in errored runs.

The simplest fix is to just turn off this behavior from PEcAn and let ED2 use it's own defaults, but this will never be merged into the develop branch.

1. Keep a fork/branch that we install PEcAn.ED2 from
2. Try to fix PEcAn.ED2 in a less invasive way:
   • re-name ED2IN.r2.2.0.github to ED2IN.rgit so history.rgit is used for defaults
   • possibly edit ED2IN to use ED 2.2.0 defaults for everything (i.e. IMETRAD, IALLOM and ISTRUCT_GROWTH_SCHEME) (compare with what other ED2IN files in inst/ do)
   • update suspected bad values in history.rgit

The sensitivity.output...Rdata file loads dataframes with all NAs in most recent run

NPP seems way too high
AGB seems way too low
transpiration seems way too low

• Inspect raw data coming from .h5 files
• Triple-check calculations and conversions
• Ask in ED2 discussions if I'm getting the units and conversions right

Revise work plan and get approval from David/Kristina before starting more runs

• Select 20 sites from SEUS that we haven't run yet
• Generate run files
• Change number of cores (programatically in setup script?) to encumber fewer cpu hours
• Start all runs

Now that PecanProject/pecan#3155 is merged, should I update the container on Welsch? I'd have to wait until a set of runs finish to do that. Maybe not worth it until these 9 sites are finished.

Currently planting Setaria in an already existing forest is only working for a site with a single PFT. Need to figure out if it works in other sites and if not, why not.

Create table with columns: pft number, number of sites in PNW, number of sites in SEUS

• mort2
• growth_resp_factor
• leaf_width
• root_respiration_rate
• Vm_low_temp

Marked as not relevant to current study. Remove these from SetariaWT pft to use ED2 defaults.

The initial idea of having three patches representing three distinct ecosystems didn't appear to work. After just a few timepoints in the simulation it appears that all PFTs were in all patches. Rather than figuring out how to make patches not interact and not split, it might be better to just do these as separate runs. It'll take ~3x longer, but we already know how to do it.

turn on SA

Train gridded model with data from #35. Use that model to predict AGB of setaria across the southeast US. Create a heatmap.

Need to also modify runlist, and this should be fine according to Mike

Work on understanding Kristina's code and getting it to run with new results.

https://github.com/cct-datascience/model-vignettes/blob/master/ED2/ensembles_modeling/predict_growth.Rmd

Files from Mike have ED2 pft numbers, but we need to create a lookup table to decide which PEcAn PFTs these match to, potentially taking into account whether the sites are in PNW (pacific northwest) or SEUS (south east US). Also keeping in mind that 3 PFTs are "upgraded":

12 -> 9
13 -> 10
14 -> 3

Although I'm suspicious about that last one since in modern PEcAn ED PFT 3 is broadleaf_evergreen_tropical_tree and it doesn't make sense that it occurs in the PNW sites

Can start work on this after brainstorming figures (https://github.com/danforthcenter/sentinel-detection/issues/421)

• Extract and clean data from .nc files
• re-create SA plots so they are customizable
• create timeseries plots of ensemble analysis

R CMD BATCH doesn't load .Rprofile, so doesn't use renv and uses different package versions.

Solution:
at the top of workflow.R source(".Rprofile"). This doesn't work though, unless you provide an absolute path or setwd(), both of which aren't great solutions. Maybe can do somethuing like setwd(../..)?

Make a slide for the April MSR

Split workflow.R into setup script, submit runs script, and process results script. This will hopefully reduce fragility and allow for some manual checking while maintaining some automation.

Also create checklists (could be programmatic, could be literal list) for each stage.

Longer runs, more ensembles, all the sites, SA with ± 1, 2, 3, SD, etc.

Use bash script to edit met headers instead of complicated R code.

1. copy to HPC
2. Edit files in place with sed -i
   cct-datascience/organization#1049 (comment)

1. Sensitivity / elasticity plots
   • For Setaria AGB
   • Can be default output by PEcAn
2. Timeseries plot (median ± uncertainty from ensemble run)
3. Pointrange, slab, raincloud, or something like that showing median ± uncertainty AGB after 10 yrs (https://mjskay.github.io/ggdist/articles/freq-uncertainty-vis.html)

For some reason model2netcdf.ED2() fails for the prairie run at site 1123 for ensemble member 1 at 2003.

> model2netcdf.ED2(
       # .x,
       outdirs[[1]],
       29.365195,
       -82.810137,
       '2002-06-01',
       '2012-06-30',
       c(
         SetariaWT = 1L,
         sentinel_ebifarm.c3grass = 5L,
         sentinel_ebifarm.c4grass = 16L,
         ebifarm.forb = 12L
       ),
       process_partial = TRUE
     )

2023-03-20 14:57:05 INFO   [model2netcdf.ED2] : 
   ----- Processing year: 2002 
2023-03-20 14:57:05 INFO   [read_E_files] : 
   *** Reading -E- file *** 
2023-03-20 14:57:15 INFO   [model2netcdf.ED2] : 
   *** Writing netCDF file *** 
2023-03-20 14:57:15 INFO   [model2netcdf.ED2] : 
   ----- Processing year: 2003 
2023-03-20 14:57:15 INFO   [read_E_files] : 
   *** Reading -E- file *** 
2023-03-20 14:57:25 INFO   [model2netcdf.ED2] : 
   *** Writing netCDF file *** 
Error in ncdf4::ncvar_put(nc = nc, varid = varid, vals = vals, start = start,  : 
  ncvar_put: error: you asked to write 48 values, but the passed data array only has 44 entries!
In addition: Warning message:
Returning more (or less) than 1 row per `summarise()` group was deprecated in dplyr 1.1.0.
ℹ Please use `reframe()` instead.
ℹ When switching from `summarise()` to `reframe()`, remember that `reframe()` always returns an ungrouped
  data frame and adjust accordingly.
ℹ The deprecated feature was likely used in the PEcAn.ED2 package.
  Please report the issue to the authors.
This warning is displayed once every 8 hours.
Call `lifecycle::last_lifecycle_warnings()` to see where this warning was generated. 

Wait for PecanProject/pecan#3140 to be merged and docker image to be pushed

Currently I've added Setaria by editing .css files to include Setaria in every patch with dbh = 0.6 (cm) and n = 1 (plants/m^2). dbh =0.6 came from what @KristinaRiemer used in previous runs with .css files and n=1 came from @dlebauer's suggestion. If these numbers are fine, this can be closed. Otherwise let's decide on a method.

So that model2netcdf.ED2() runs on the HPC

We said we'd try for ~ 50 sites. Someone needs to come up with criteria and/or a sampling procedure for narrowing down the sites.

https://docs.github.com/en/get-started/using-git/splitting-a-subfolder-out-into-a-new-repository

Remove most priors from PFTs. Keep a few parameters so SA and ensemble analysis generate variation. Replace uniform priors when possible.

Params to keep:

• SLA
• Vcmax
• stomatal slope
• cuticular conductance
• quantim efficiency
• fineroot2leaf

Some more "custom" pfts like the ebifarm.forb might need to have more priors to make it more distinct from the default ED2 PFT it's based on.

Remove .h5 files if .nc files were successfully created at the end of a run so Welsch doesn't fill up.