Geoscience Reference
In-Depth Information
10
Chemometric Analysis of Organic Matter Fluorescence
Kathleen R. Murphy, Rasmus Bro, and Colin A. Stedmon
10.1 Introduction
In multivariate data analysis (MVA), statistical and mathematical modeling techniques are
used to analyze data that result from measuring multiple variables on several samples.
Important variables influencing the fluorescence characteristics of a dissolved organic mat-
ter sample might include, among others, its composition, temperature, pH, and the concen-
tration and nature of any quenching agents that may be present. In
Chapter 9
, top-down
(researcher-driven) univariate techniques for exploring fluorescence data were introduced.
Such techniques can reduce the complexity of fluorescence data and aid visualization and
interpretation and are rapidly implemented and easy to understand. However, because they
require the researcher to identify which features of a data set deserve to be included in
models, they presuppose a good understanding of the system being studied. In addition,
with this type of approach there is a danger that unanticipated events remain undetected.
The aim of
exploratory
analysis is to avoid this, by presenting a model and visualization of
the data not focused on prior hypotheses. In fact, the basic idea in exploratory analysis is to
use the data analysis to get ideas for hypotheses
from the data
. With advances in technology
it is a simple matter these days to acquire comprehensive fluorescence data sets, even in
environments that were previously difficult to sample. Thus, data sets can quickly become
large or complex, masking subtle but potentially important differences between samples. In
cases where there is little existing experience of the sampling environment, methodologies
are needed that ensure that in the process of condensing the original data set, all relevant
information is retained and detected.
Extracting chemical information from a multivariate chemical data set falls within a
field of science known as chemometrics. It uses statistical and mathematical methods com-
bined with chemical and physical insight. Often the analyses are visually driven by the data
rather than imposed upon them under a theoretical or statistical framework. Chemometrics
has proved particularly useful for exploring and interpreting complex data sets involving
large numbers of variables that relate to one another in ways that are poorly understood. A
number of reviews of chemometric applications in chemistry and spectroscopy are avail-
able (Mobley et al.,
1996
; Workman et al.,
1996
; Bro et al., 1997; Bro,
2006
; Lavine and
Workman,
2010
).
339
