jgcri / an2month Goto Github PK

including information noted in Issues #17 and #16

To whatever standards @crvernon has established.

Forgot to open issue on Jul 06, 2020, when work started:
currently, from the code, it appears that the monthly fractions are sampled from a distribution once for each realization of annual data. So every year in realization 1 is the same kind of year, every year in realization 2 is the same kind of year (but different from the kind of year in realization 1).

• verify that is what the code is doing
• add option for using just the mode estimate from the fit for the fractions from the fit multivariable beta distribution, for all years and realizations
• add option for sampling across years (year 1 is the same kind of year in every realization, but a different kind of year than year 2)
• add option for sampling across realizations and years

Need to add documentation to functions and package data!

I just wanted to keep track of the cdo functions used in the monthly fraction processing code.

cdo documentation from https://code.mpimet.mpg.de/projects/cdo/embedded/index.html#x1-4610002.8.23.

• Instead of hard coding the extra time step to the end of the time series as 01 make it flexible
• Make digits into a user defined argument with a default setting

Set up raw-data processing code to calculate the average monthly fraction of precipitation that falls within a year. Output from the raw-data processing should be stored as internal package data as some sort of documented Rdata object like an array or something.

validating that what we see are reasonable.

brainstorm for metrics to look at?

@claudiatebaldi @kdorheim In trying to work through the code in more depth for doing this enhancement #16,

I've done more careful, line by line combing through the nested functions in data_raw/L3_fit_dirichlet_params.R and data_raw/jobrun.zsh. I think that the code is estimating the parameters of a multivariable beta distribution for the temperature data, and a separate set of parameters for the precipitation data. At least I think.

I didn't catch it in my initial trying to learn the an2month package, I think because of how the functions are nested. And because I think that approach of treating T and P separately is different from the very early notes I had contributing to figuring out what the sampling should look like (around Dec 2018) and then I wasn't involved in the actual work. And then so many issues came up with how fldgen was being called in the pipeline, I didn't return to this until last week/this week.

So do we want to keep T and P separate the way they're implemented, or do we want to estimate 24 parameters together (like I initially thought was happening)? Also thoughts on continuing to use a multivariate beta distribution?

• Check if the data files on PIC used to estimate fractions:
  https://github.com/JGCRI/an2month/blob/master/data-raw/L0.find_and_preprocess_isimip.R#L22
  ISIMIP_DIR <- "/pic/projects/GCAM/leng569/data/Input_GCM_bced" # The location of the isimip files on pic
  agree with the files ultimatly used elsewhere in the experiment.

• Check if the fractions are only from RCP4.5 as the code
  https://github.com/JGCRI/an2month/blob/master/data-raw/L0.find_and_preprocess_isimip.R#L38 and
  https://github.com/JGCRI/an2month/blob/master/data-raw/L0.find_and_preprocess_isimip.R#L61-L90 and
  https://github.com/JGCRI/an2month/blob/master/data-raw/run_L0_isimip.sh
  suggest - the .sh file appears to just be running the .R scripts once, with the default argument, suggesting only the RCP45 data is the only data being used to estimate fractions. Possible that what got run on pic differs from what's here in the code; at any rate we'll need to check through and document better at least for the paper

FYI to @crvernon @FeralFlows

Do we want to incorporate the Xanthos unit conversions into downscaling_component_api or should it be done in a separate step?

Monthly precipitation values are average rates, not total accumulation. The monthly downscaling function assumes the latter, and so needs to be changed.

Package data names will be changing to alpha_<model> instead of frac_<model>. Update relevant code to reflect this.

Documentation says the input alpha to the function monthly_downscaling should be

#' @param alpha Matrix[ngrid, 12] of the alpha parameters for the Dirichlet
#' distribution.  These values are included as package data.  For more details
#' on how they are calculated see data-raw.

But the code is actually expecting alpha to be a list, one entry of which is that matrix (under the appropriate variable name).

Will probably address this under the PR for #16, and will just add any other additional documentation corrections that come up here.