fit.SpectraModel-method: Fits a calibration model for soil spectroscopy

Description

Fits a calibration model that can be used to convert absorbances to soil properties. Implements the "plsr" function available from the pls package.

Usage

"fit"(formulaString, sampled, reference,  idcol = "SAMPLEID", ab.r = get("MIR", envir=spec.opts),  CO2.band = get("CO2.band", envir=spec.opts), ncomp, ncomp.max=15,  repl = 5, segment.type = "interleaved", prefix = "X", ...)
"fit"(formulaString, sampled, reference, ...)

Arguments

formulaString

"formula"; formula string

sampled

"SpectraPoints"; absorbances

reference

"data.frame"; reference values estimated in laboratory using at the order of magnitude more precise technique

idcol

"character"; column with sample IDs

ab.r

"numeric"; natural range of wavenumbers of interest

CO2.band

"numeric"; range of wavenumbers for the CO2 band

ncomp

"integer"; number of components (if not specified will be determined using cross-validation)

ncomp.max

"integer"; maximum number of components to be used for model fitting

repl

"integer"; number of repetitions for "pls::plsr"

segment.type

"character"; segment type for "pls::plsr"

prefix

"character"; default prefix to be added to the band names

...

other optional arguments

Details

Returns an object of class "SpectraModel", which contains slots: (1) "variable" (variable name according to the global soil data registry), (2) "Space" (3D space defined by the 1--3 principal components for absorbances), and (3) "model" (fitted model; usually of type pls::plsr and as an output from the partial least squares regression). Fitting models for large data sets can be time and memory consuming.

References

Mevik, B.-H., Wehrens, R. (2007) The pls Package: Principal Component and Partial Least Squares Regression in R. Journal of Statistical Software 18(2), 1-24.
Africa Soil Information Service (http://africasoils.net)

Examples

Run this code

## Not run: 
# ## build a callibration model using AfSIS data:
# library(plyr)
# library(sp)
# library(rgdal)
# 
# data(afss)
# afss.tbl <- join(afss$sites, afss$horizons)
# con <- url("http://gsif.isric.org/lib/exe/fetch.php?media=afss.spec.rda")
# load(con)
# afss.spec.ab <- Spectra(samples=data.frame(SAMPLEID=afss.spec$SAMPLEID, 
#    MID="ICR_SOIL", DateTime=Sys.time()), ab=afss.spec)
# sel.r <- !afss.tbl$LONWGS84==100
# sp <- data.frame(afss.tbl[sel.r,c("LONWGS84","LATWGS84")])
# coordinates(sp) <- ~LONWGS84+LATWGS84
# proj4string(sp) <- CRS("+proj=longlat")
# attr(sp@coords, "dimnames")[[1]] <- afss.tbl$SAMPLEID[sel.r]
# ## extend to SprectraPoints:
# afss.spec.sp <- SpectraPoints(Spectra=afss.spec.ab, sp=sp)
# ## select bands of interest:
# cutspec <- c(600,2350.8,2379.8,4000)
# col.no <- sapply(names(afss.spec.sp@data@ab)[-1], 
#    function(x){as.numeric(strsplit(x, "X")[[1]][2])})
# sel <- col.no>cutspec[1] & col.no<cutspec[2] | 
#        col.no>cutspec[3] & col.no<cutspec[4]
# summary(sel)
# formulaString <- as.formula(paste("na.omit(log(ORCDRC)) ~ ", 
#    paste(names(afss.spec.sp@data@ab[-1])[sel], collapse="+")))
# m.ORC <- fit(formulaString, sampled=afss.spec.sp, reference=afss.tbl) 
# str(m.ORC)
# ## End(Not run)
## this model, in fact, is included in the package:
data(m.ORCCNS)
str(m.ORCCNS)