Learn R Programming

aqp (version 1.18)

profileApply-methods: Apply a function to soil profiles within a SoilProfileCollection object.

Description

Apply a function to soil profiles within a SoilProfileCollection object, each iteration has access to a SoilProfileCollection object.

Usage

# method for SoilProfileCollection objects
profileApply(object, FUN, simplify=TRUE, ...)

Arguments

object

a SoilProfileCollection

FUN

a function to be applied to each profile within the collection

simplify

logical, should the result be simplified to a vector? see examples

further arguments passsed to FUN

Value

When simplify is TRUE, a vector of length nrow(object) (horizon data) or of length length(object) (site data). When simplify is FALSE, a list is returned.

Methods

signature(object = "SoilProfileCollection")

See Also

slab, estimateSoilDepth

Examples

Run this code
# NOT RUN {
data(sp1)
depths(sp1) <- id ~ top + bottom

# estimate soil depth using horizon designations
profileApply(sp1, estimateSoilDepth, name='name', top='top', bottom='bottom')

# scale properties within each profile
# scaled = (x - mean(x)) / sd(x)
sp1$d <- profileApply(sp1, FUN=function(x) round(scale(x$prop), 2))
plot(sp1, name='d')


# compute depth-wise differencing by profile
# note that our function expects that the column 'prop' exists
f <- function(x) { c(x$prop[1], diff(x$prop)) }
sp1$d <- profileApply(sp1, FUN=f)
plot(sp1, name='d')

# compute depth-wise cumulative sum by profile
# note the use of an anonymous function
sp1$d <- profileApply(sp1, FUN=function(x) cumsum(x$prop))
plot(sp1, name='d')


# compute profile-means, and save to @site
# there must be some data in @site for this to work
site(sp1) <- ~ group
sp1$mean_prop <- profileApply(sp1, FUN=function(x) mean(x$prop, na.rm=TRUE))

# re-plot using ranks defined by computed summaries (in @site)
plot(sp1, plot.order=rank(sp1$mean_prop))


## iterate over profiles, subsetting horizon data

# example data
data(sp1)

# promote to SoilProfileCollection
depths(sp1) <- id ~ top + bottom
site(sp1) <- ~ group

# make some fake site data related to a depth of some importance
sp1$dep <- profileApply(sp1, function(i) {round(rnorm(n=1, mean=mean(i$top)))})

# custom function for subsetting horizon data, by profile
# keep horizons with lower boundary < site-level attribute 'dep'
fun <- function(i) {
  # extract horizons
  h <- horizons(i)
  # make an expression to subset horizons
  exp <- paste('bottom < ', i$dep, sep='')
  # subset horizons, and write-back into current SPC
  horizons(i) <- subset(h, subset=eval(parse(text=exp)))
  # return modified SPC
  return(i)
}

# list of modified SoilProfileCollection objects
l <- profileApply(sp1, fun, simplify=FALSE)

# re-combine list of SoilProfileCollection objects into a single SoilProfileCollection
sp1.sub <- union(l)

# graphically check
par(mfrow=c(2,1), mar=c(0,0,1,0))
plot(sp1)
points(1:length(sp1), sp1$dep, col='red', pch=7)
plot(sp1.sub)


# }
# NOT RUN {
##
## helper functions: these must be modified to suit your own data
##

# compute the weighted-mean of some property within a given diagnostic horizon
# note that this requires conditional eval of data that may contain NA
# see ?slab for details on the syntax
# note that function expects certain columns within 'x'
f.diag.wt.prop <- function(x, d.hz, prop) {
  # extract diagnostic horizon data
  d <- diagnostic_hz(x)
  # subset to the requested diagnostic hz
  d <- d[d$diag_kind == d.hz, ]
  # if missing return NA
  if(nrow(d) == 0)
    return(NA)
  
  # extract depths and check for missing
  sv <- c(d$featdept, d$featdepb)
  if(any(is.na(sv)))
    return(NA)
  
  # create formula from named property
  fm <- as.formula(paste('~', prop))
  # return just the (weighted) mean, accessed from @horizons
  s <- slab(x, fm, slab.structure=sv, slab.fun=mean)$value
  return(s)
}


# conditional eval of thickness of some diagnostic feature or horizon
# will return a vector of length(x), you can save to @site
f.diag.thickness <- function(x, d.hz) {
  # extract diagnostic horizon data
  d <- diagnostic_hz(x)
  # subset to the requested diagnostic hz
  d <- d[d$diag_kind == d.hz, ]
  # if missing return NA
  if(nrow(d) == 0)
    return(NA)
  
  # compute thickness
  thick <- d$featdepb - d$featdept
  return(thick)
}


# conditional eval of property within particle size control section
# makes assumptions about the SPC that is passed-in
f.psc.prop <- function(x, prop) {
  # these are accessed from @site
  sv <- c(x$psctopdepth, x$pscbotdepth)
  # test for missing PCS data
  if(any(is.na(sv)))
    return(NA) 
  
  # this should never happen... unless someone made a mistake
  # check to make sure that the lower PSC boundary is shallower than the depth
  if(sv[2] > max(x))
    return(NA)
  
  # create formula from named property
  fm <- as.formula(paste('~', prop))
  # return just the (weighted) mean, accessed from @horizons
  s <- slab(x, fm, slab.structure=sv, slab.fun=mean)$value
  return(s)
}

# try with some sample data
data(loafercreek, package='soilDB')

profileApply(loafercreek, f.diag.wt.prop, d.hz='argillic horizon', prop='clay')
profileApply(loafercreek, f.diag.thickness, d.hz='argillic horizon')
profileApply(loafercreek, f.psc.prop, prop='clay')

# }

Run the code above in your browser using DataLab