emulator_reducer: Subset a trained emulator.

In JGCRI/fieldgenr: Generate Temperature Fields with Spatial and Temporal Correlation

Description

A trained fldgen emulator features a large amount of data for both using the emulator and rigorously validating an emulator.

Usage

emulator_reducer(emulator)

Arguments

emulator

A trained fldgen emulator, with all entries needed for generating new residuals and for rigourously validating the quality of the trained emulator

Details

If one is just interested in the use of an emulator for generating felds, this function can be called to reduce a trained emulator to the bare essential list entries, which can then be saved and called the same as an unreduced emulator by generate.TP.resids and generate.TP.fullgrids

Note that with this reduced emulator, there is NO way to reconstruct the training data. A fully trained emulator contains a copy of the training data, in addition to the training regressor values (tgav), and the estimated linear model parameters and residuals (meanfieldT$b, w, r), which together can also reconstruct the data.

Even though the coordinate information stored in an emulator$griddataT is not needed directly to generate a new field of residuals or full data, it is often needed in downstream use of the fields. Therfore an entry reducedEmulator$griddataT$coord containg a matrix is saved in the reducedEmulator. Each is a matrix of coordinates for each grid cell, with cells in rows and latitude, longitude in the two columns. Keeping these coordinate matrices for T and P is negligible.

Finally, the reduced emulator produced by this function is specifically meant for temperature and precipitation only, and is not robust to extension to other variables.

Finally finally, if a user is interested in a different subset of list entries in a trained emulator, they are encouraged to subset and save themself, as appropriate for their project.

Value

reducedEmulator A trained fldgen emulator with only the list entries needed by generate.TP.resids and generate.TP.fullgrids for generating new fields:

griddataT

Only the coordinate ids and set information.

griddataP

Only the coordinate ids and set information, and the function to convert from logP to P.

\

tgav

The Tgav data from training.

meanfldT

the slope (w) and intercept (b) terms from the mean field fit.

meanfldP

the slope (w) and intercept (b) terms from the mean field fit.

tfuns

The empirical quantile functions for temperature, mapping N(0,1) to the native distribution in each grid cell.

pfuns

The empirical quantile functions for logP, mapping N(0,1) to the native distribution in each grid cell.

reof

The EOFs.

fx

Time coefficients for each EOF from training data.

infiles

The names of the files used for training the emulator.

Author(s)

ACS July 2020

JGCRI/fieldgenr documentation built on July 22, 2020, 3:17 a.m.