EMMA {EMMAgeo}

End-member modelling analysis algorithm.

Description

A multivariate data set (m samples composed of n variables) is decomposed by eigenspace analysis and modelled with a given number of end-members (q). Several steps of scaling, transformation, normalisation, eigenspace decomposition, factor rotation, data modelling and evaluation are performed.

Usage

EMMA(
  X,
  q,
  l,
  c,
  Vqn,
  classunits,
  ID,
  EM.ID,
  rotation = "Varimax",
  plot = FALSE,
  ...
)

Arguments

X

Numeric matrix, input data set with m samples (rows) and n variables (columns).

q

Numeric scalar, number of end-members to be modelled.

l

Numeric scalar or vector, weight transformation limit, i.e. quantile. Set to zero if omitted.

c

Numeric scalar, constant sum scaling parameter, e.g. 1, 100, 1000. Set to 100 if omitted.

Vqn

Numeric matrix, optional unscaled user-defined end-member loadings. If provided, these are used instead of model-derived ones. See details.

classunits

Numeric vector, optional class units (e.g. micrometers or phi-units) of the same length as columns of X.

ID

Numeric or character vector, optional sample IDs of the same length as rows of X.

EM.ID

Character vector, end-member names. If present, these will be set as row-names of the output data set and used in the legend text.

rotation

Character scalar, rotation type, default is "Varimax". See details.

plot

Logical scalar, optional graphical output of the results, default is FALSE. If set to TRUE, end-member loadings and end-member scores are plotted.

...

Additional arguments passed to the plot function. Since the function returns two plots some additional graphical parameters must be specified as vector with the first element for the first plot and the second element for the second plot.

Details

The parameter Vqn is useful when EMMA shall be performed with a set of prior unscaled end-members, e.g. from other data sets that are to be used as reference or when modelling a data set with mean end-members, as in the output of robust.loadings.
The rotation type Varimax was used by Dietze et al. (2012). In this R package, one out of the rotations provided by the package GPArotation is possible, as well. However, tests showed that the rotation type has no dramatic consequences for the result.
The function values $loadings and $scores are redundant. They are essentially the same as $Vqsn and $Mqs. However, they are included for user convenience.

Value

A list with numeric matrix objects.

loadings

Normalised rescaled end-member loadings.

scores

Rescaled end-member scores.

Vqn

Normalised end-member loadings.

Vqsn

Normalised rescaled end-member loadings.

Mqs

Rescaled end-member scores.

Xm

Modelled data.

modes

Mode class of end-member loadings.

Mqs.var

Explained variance of end-members

Em

Absolute row-wise model error.

En

Absolute column-wise model error.

RMSEm

row-wise root mean square erroe

RMSEn

column-wise root mean square erroe

Rm

Row-wise (sample-wise) explained variance.

Rn

Column-wise (variable-wise) explained variance.

ol

Number of overlapping end-members.

Author(s)

Michael Dietze, Elisabeth Dietze

References

Dietze E, Hartmann K, Diekmann B, IJmker J, Lehmkuhl F, Opitz S, Stauch G, Wuennemann B, Borchers A. 2012. An end-member algorithm for deciphering modern detrital processes from lake sediments of Lake Donggi Cona, NE Tibetan Plateau, China. Sedimentary Geology 243-244: 169-180.
Klovan JE, Imbrie J. 1971. An Algorithm and FORTRAN-IV Program for Large-Scale Q-Mode Factor Analysis and Calculation of Factor Scores. Mathematical Geology 3: 61-77. Miesch AT. 1976. Q-Mode factor analysis of geochemical and petrologic data matrices with constant row sums. U.S. Geological Survey Professsional Papers 574.

See Also

test.parameters, rotations, eigen, nnls

Examples

## load example data and set phi-vector
data(example_X)
phi <- seq(from = 1, to = 10, length.out = ncol(X))

## perform EMMA with 5 end-members
EM <- EMMA(X = X, q = 5, l = 0.05, c = 100, plot = TRUE)

## perform EMMA with 4 end-members and more graphical settings
EM <- EMMA(X = X, q = 4, l = 0.05, c = 100, 
           plot = TRUE,
           EM.ID = c("EM 1", "EM 2", "EM 3", "EM 4"),
           classunits = phi,
           xlab = c(expression(paste("Class [", phi, "]")), "Sample ID"),
           cex = 0.7,
           col = rainbow(n = 4))