debias.met.regression.Rd
This script debiases one dataset (e.g. GCM, re-analysis product) given another higher resolution product or empirical observations. It assumes input are in annual CF standard files that are generate from the pecan extract or download funcitons.
debias.met.regression(train.data, source.data, n.ens, vars.debias = NULL, CRUNCEP = FALSE, pair.anoms = TRUE, pair.ens = FALSE, uncert.prop = "mean", resids = FALSE, seed = Sys.Date(), outfolder, yrs.save = NULL, ens.name, ens.mems = NULL, force.sanity = TRUE, sanity.tries = 25, lat.in, lon.in, save.diagnostics = TRUE, path.diagnostics = NULL, parallel = FALSE, n.cores = NULL, overwrite = TRUE, verbose = FALSE)
train.data | - training data coming out of align.met |
---|---|
source.data | - data to be bias-corrected aligned with training data (from align.met) |
n.ens | - number of ensemble members to generate and save for EACH source ensemble member |
vars.debias | - which met variables should be debiased? if NULL, all variables in train.data |
CRUNCEP | - flag for if the dataset being downscaled is CRUNCEP; if TRUE, special cases triggered for met variables that have been naively gapfilled for certain time periods |
pair.anoms | - logical stating whether anomalies from the same year should be matched or not |
pair.ens | - logical stating whether ensembles from train and source data need to be paired together (for uncertainty propogation) |
uncert.prop | - method for error propogation if only 1 ensemble member; options=c(random, mean); *Not Implemented yet |
resids | - logical stating whether to pass on residual data or not *Not implemented yet |
seed | - specify seed so that random draws can be reproduced |
outfolder | - directory where the data should go |
yrs.save | - what years from the source data should be saved; if NULL all years of the source data will be saved |
ens.name | - what is the name that should be attached to the debiased ensemble |
ens.mems | - what labels/numbers to attach to the ensemble members so we can gradually build bigger ensembles without having to do do giant runs at once; if NULL will be numbered 1:n.ens |
force.sanity | - (logical) do we force the data to meet sanity checks? |
sanity.tries | - how many time should we try to predict a reasonable value before giving up? We don't want to end up in an infinite loop |
lat.in | - latitude of site |
lon.in | - longitude of site |
save.diagnostics | - logical; save diagnostic plots of output? |
path.diagnostics | - path to where the diagnostic graphs should be saved |
parallel | - (experimental) logical stating whether to run temporal_downscale_functions.R in parallel *Not Implemented yet |
n.cores | - (experimental) how many cores to use in parallelization *Not implemented yet |
overwrite | - overwrite existing files? Currently ignored |
verbose | logical: should |
Debias Meteorology using Multiple Linear Regression Statistically debias met datasets and generate ensembles based on the observed uncertainty
Other debias - Debias & Align Meteorology Datasets into continuous time series: align.met