paleocar_models: Fit PaleoCAR models to a (potentially large) set of...

In bocinsky/PaleoCAR: Paleoclimate Reconstruction from Tree Rings using Correlation Adjusted corRelation

Fit PaleoCAR models to a (potentially large) set of predictands

Description

This is the primary function for fitting PaleoCAR models to a large set of predictands using a uniform set of predictors (tree-ring chronologies). For each predictand, carscores are calculated for each predictor, and models are calculated by adding predictors stepwise to an ordinary least-squares linear model using the lm function. Model selection is performed by minimizing corrected AIC. This occurs for every unique set of available predictors through time.

Usage

paleocar_models(
  chronologies,
  predictands,
  calibration.years,
  prediction.years = NULL,
  min.width = NULL,
  verbose = F,
  ...
)

Arguments

chronologies	A matrix of tree ring chronologies, indexed annually. Each chronology is a column. The first column must be labeled "YEAR" and is the calendar year.
predictands	A RasterBrick, RasterStack, matrix, or vector of the numeric predictand (response) variable.
calibration.years	An integer vector of years corresponding to the layers in the predictands brick.
prediction.years	An optional integer vector of years for the reconstruction. If missing, defaults to the total years present in chronologies.
min.width	integer, indicating the minimum number of tree-ring samples allowed for that year of a chronology to be valid.
verbose	Logical, display status messages during run.
...	Further arguments to be passed to other functions.

Details

See paleocar_models for a simplified algorithm that fits a PaleoCAR model for a single predictand. The batch algorithm gains efficiencies by minimizing the number of unique model calculations (using multiple-response models) and by halting calculations for individual predictands once they stabilize.

Value

A named list containing

• models A data.table giving the fitted coefficients, LOOCV errors, and AICc values through time for each predictand.

• predictands A vector, matrix, or RasterBrick of predictands, as provided.

• predictor.matrix A matrix of predictors for calibration; chronologies cropped to calibration.years.

• reconstruction.matrix A matrix of predictors for reconstruction; chronologies cropped to prediction.years, or all of chronologies if prediction.years==NULL.

• carscores A list of carscore matrices, one for each predictand.

bocinsky/PaleoCAR documentation built on Feb. 23, 2023, 12:14 p.m.