R/auto_PCA.R

In sdmvspecies: Create Virtual Species for Species Distribution Modelling

#!/usr/bin/env Rscript

#' autoPCA
#'
#' easily used PCA analysis
#'
#' This method implemented an easily used PCA analysis method
#'
#' @param env.stack a \code{rasterStack} object that contain the environment variable
#' @param nfactors Number of factors to extract
#' @return \code{rasterStack} object
#' @encoding utf-8
#' @export
autoPCA <- function(env.stack, nfactors) {
    env.data <- values(env.stack)

    pc <- principal(env.data, nfactors=nfactors)
    weights <- pc$weights

    PC_list <- list()
    for (index in 1:nfactors) {
        env.name.list <- rownames(weights)
        component_list <- NULL
        for (env.name in env.name.list) {
            # env.name <- env.name.list[1]
            weights.data <- weights[env.name, index]
            env.value <- env.data[,env.name]
            component <- weights.data * env.value
            if (!is.null(component_list)) {
                component_list <- cbind(component_list, component)
            } else {
                component_list <- component
                component_list <- as.data.frame(component_list, byrow=TRUE, ncol=1)
            }
        }
        colnames(component_list) <- env.name.list
        raster.value <- apply(component_list, FUN=sum, MARGIN=1)
        raster.model <- env.stack[[1]]
        values(raster.model) <- raster.value
        PC_list[[index]] <- raster.model
    }
    name.vector <- paste("PC", 1:nfactors, sep="")
    names(PC_list) <- name.vector
    result.stack <- stack(PC_list)
    return(result.stack)
}