Geoscience Reference
In-Depth Information
Table 10.1.
Overview of the main available chemometric techniques used for exploring organic matter fluorescence data sets
Category
Methods
Input data
Goals
Exploratory -
visualization
and clustering
Cluster analysis, e.g., hierarchical
clustering,
k
-means clustering, self
organizing maps (SOM)
Multivariate data set consisting of
training and validation samples
Assign observations into groups containing
similar samples.
Principal component analysis (PCA)
Multivariate data set consisting of
training and validation samples
Reduce dimensionality, explore and visualize
linear gradients of variability in the data set,
identify clustering among samples
Exploratory -
spectral
decomposition
Multivariate curve resolution (MCR)
Multivariate data set (e.g.,
synchronous scan spectra)
consisting of training and
validation samples
Determine the number, amount and spectral
shapes of underlying components in
mixtures, explore variability in the data set,
visualization, clustering
PARAFAC, PARAFAC2, Tucker3
Multiway data set consisting of
training and validation samples
Determine the number, amounts, and spectral
shapes of underlying components in
mixtures, explore variability in the dataset,
visualization, clustering. PARAFAC allows
no spectral shifts, PARAFAC2 allows explicit
spectral shifts along one axis, Tucker3 allows
implicit spectral shifts along all axes but has
no second-order advantage.
Exploratory -
time series
Principal filters analysis (PFA)
Time series of fluorescence EEMs
Identify periods of time associated with high
fluorescence variability.
Calibration
Principal components regression (PCR),
partial least squares regression (PLS),
and their multiway versions (N-PCR
and N-PLS)
Multivariate or multiway data set
consisting of independent and
dependent variables. Training
and validation samples
Predict dependent variables from independent
variables
Classification
PLS discriminant analysis (PLS-DA), soft
independent modeling of class analogy
(SIMCA), multiway N-PLS-DA
Multivariate or multiway data set,
training samples identified by
category, test samples
Classify training and test samples into one of
several possible categories
