Skip to contents

Determine ML Horizon Boundaries

Source: R/get.ml.hz.R
get.ml.hz.Rd

This function accepts input from slab() (a data.frame) along with a vector of horizon names, and returns a data.frame of the most likely horizon boundaries.

This function expects that x is a data.frame generated by slab(). If x was not generated by slab(), then o.names is required.

Usage

get.ml.hz(x, o.names = attr(x, which = "original.levels"))

Arguments

x

data.frame, output from slab()

o.names

an optional character vector of horizon designations that will be used in the final table

Value

A data.frame with the following columns:

  • hz: horizon names

  • top: horizon top depth

  • bottom: horizon bottom depth

  • confidence: integrated probability over thickness of each ML horizon, rounded to the nearest integer

  • pseudo.brier: A "pseudo"" Brier Score for a multi-class prediction, where the most-likely horizon label is treated as the "correct" outcome. Details on the calculation for traditional Brier Scores here: https://en.wikipedia.org/wiki/Brier_score. Lower values suggest better agreement between ML horizon label and class-wise probabilities.

  • mean.H: mean Shannon entropy (bits), derived from probabilities within each most-likely horizon. Larger values suggest more confusion within each ML.

References

Beaudette, D. E., Roudier, P., & Skovlin, J. (2016). Probabilistic representation of genetic soil horizons. Digital soil morphometrics, 281-293.

See also

slab()

Author

D.E. Beaudette

Examples


# init SPC
data(sp1)
depths(sp1) <- id ~ top + bottom

# set horizon designation metadata
hzdesgnname(sp1) <- 'name'

# generalize horizon designations from character vector
# result is an ordered factor
sp1$genhz <- generalizeHz(
  sp1$name,
  new = c('O','A','B','C'),
  pat = c('O', '^A','^B','C'),
  ordered = TRUE
)

# compute slice-wise GHL probability
# so that it sums to contributing fraction
# from 0-150cm
a <- slab(sp1, fm = ~ genhz, cpm = 1, slab.structure = 0:150)
#> horizons with zero thickness have been omitted from results

# note original GHL names are set by slab()
attr(a, 'original.levels')
#> [1] "O"        "A"        "B"        "C"        "not-used"

# generate table of ML horizonation
get.ml.hz(a)
#>   hz top bottom confidence pseudo.brier    mean.H
#> 1  O   0      2         37    0.3950617 0.9910761
#> 2  A   2     32         75    0.1547325 0.7922828
#> 3  B  32    145         57    0.3574667 1.0813045
#> 4  C 145    150         71    0.1250000 0.8112781