predict.SpectraPoints-method: Predict soil properties from an object of class `"SpectraPoints"`

Description

Predicts soil properties from an object of class "SpectraPoints" (or a "data.frame" containing absorbances and sample IDs) either by using a referent "SpectraModel" (comes with the package) or by using the user created model. Extends, for example, the "predict.mvr" function available from the pls package.

Usage

"predict"(object, idcol = "SAMPLEID", model,  variable = model@variable,  output, validate = FALSE, model.class = "mvr", confidence.band = TRUE, prob. = .90,  signif.digits = 3,  st.wavenumbers=wavenumbers,  instr.range = c("ten-mir", "alp-mir", "mpa-nir")[1], ...)
"predict"(object, idcol = "SAMPLEID", ...)

Arguments

object

object of type "SpectraPoints"

idcol

"character"; ID column name

model

"SpectraModel"; prediction model

variable

"character"; variable name (see 'attributes in spec.env(show.env = TRUE))

output

"SoilProfileCollection"; optional output object where the predicted soil properties can be written

validate

"logical"; runs validation of overlap of points in feature space (see details)

model.class

"character"; prediction model class

confidence.band

"logical"; specifies whether confidence bands should be derived for each prediction

prob.

"numeric"; probability for confidence bands

signif.digits

"integer"; significant digits (all output numbers are typically rounded to three significant digits to save space)

st.wavenumbers

"data.frame"; standard wavenumbers with band names and upper / lower limits (usually predefined)

instr.range

"character"; instrument range

...

other optional arguments

Details

When predicting from a data.frame, column names must contain wavenumbers and first column must contain sample IDs. Validation of overlap in feature space is done using PCs 1-3 derived from absorbance bands. This method basically compares for each new sample shortest distance (in 3D space; derived using spatstat::nncross.pp3) to the cloud of referent points and then prints warning if the distance is larger than two times the largest nndist (distance) for the referent point cloud.

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.

Examples

Run this code

data(afspec)
## predict pH using absorbances directly:
data(m.PHIHOX)
x <- predict(afspec$ab, model=m.PHIHOX)
## Not run: 
# ## how good is this model?
# ## we look at the calibration data used to produce "m.PHIHOX"
# pr <- m.PHIHOX@model$fitted.values
# library(plyr)
# library(scales)
# data(afss)
# afss.tbl <- join(afss$sites, afss$horizons)
# tbl <- merge(data.frame(predicted=pr, SAMPLEID=attr(pr, "names")), 
#    afss.tbl[,c("SAMPLEID","PHIHOX")])
# v.r <- range(tbl$PHIHOX)
# r.square <- summary(lm(as.formula(paste("PHIHOX ~ predicted - 1")),
#    tbl))$adj.r.squared
# plot(x=tbl$predicted, y=tbl$PHIHOX, asp=1, xlim=v.r, ylim=v.r, 
#    col=alpha("darkgrey", 0.5), xlab="Predicted", ylab="Measured", 
#    main=paste("R-square:", formatC(r.square, digits=3), sep=""))
# abline(a=0, b=1, lw=2, col="black")
# ## End(Not run)