R/invert.lsq.R

In ashiklom/PEcAnRTM: PEcAn functions used for radiative transfer modeling

#' Least squares model inversion
#' 
#' Performs a least-squares inversion of an arbitrary radiative transfer model 
#' (passed as an R function). The inversion attempts to minimize the sum of 
#' residual least squares between modeled and observed spectra via the 
#' Levenberg-Marquardt algorithm (`nls.lm` function from the `minpack.lm` 
#' package).
#' @author Alexey Shiklomanov
#' @param observed Vector of observations (e.g. a reflectance spectrum).
#' @param inits Vector of initial conditions for the parameters.
#' @param model An R function that calls the RTM and returns the error to be 
#' minimized. Be sure to include constants here.
#' @param lower Lower bounds on parameters (default=NULL, which means -Inf).
#' @param uppper Upper bounds on parameters (default=NULL, which means +Inf).
#' @export

invert.lsq <- function(observed, inits, model, lower = NULL, upper = NULL) {
  testForPackage("minpack.lm")
  observed <- as.matrix(observed)
  merit <- function(params) {
    spec <- model(params)
    error <- spec - observed
    return(error)
  }
  fit <- minpack.lm::nls.lm(par = inits, lower = lower, upper = upper, fn = merit)
  return(fit)
} # invert.lsq