Geochemical Fingerprinting - Application of Geochemical Tracers to Fluvial Sediment - page 25

estimates the concentration of the sediment if it was entirely composed of fine-
grained, chemically reactive material, and is performed using the following equa-
tion:

NC = DF · BC (2.1)

where BC is the bulk concentration and the dilution factor, DF , is calculated as:

100
RS

DF = (2.2)

where RS is the percentage of reactive sediment of a given size range.
A significant disadvantage of the approach is that the normalized data do not
necessarily reflect the actual chemical concentrations within the sampled sediments
for the selected size range, particularly when the samples contain < 50% silt and

clay (Horowitz 1991 ).
A slightly different approach is to normalize bulk elemental concentrations by
the concentration of a conservative element such as Al, Ti, or Li. In contrast to the
methods used for grain size, normalization is performed by dividing the concentration
of the potential tracer by the concentration of the conservative element.
The third method commonly used to deal with the transport invariant problem
is to incorporate a correction factor into the mixing model (Collins et al. 1998 ;
He and Owens 1995 ; Russell et al. 2001 ; Motha et al. 2003 , 2004 ; Juracek and
Ziegler 2009 ). This approach has been widely used to account for the effects of
both grain size and organic matter. However, Mukundan et al. ( 2012 ) point out that
the relationship between a specific fingerprinting parameter and grain size and/or
organic matter content may vary between the other parameters used in the composite
fingerprint; thus, the incorporation of a single, universally applicable correction factor
into the model may not be appropriate. In addition, it has been argued that the use
of multiple correction factors, such as one for grain size and one for organic matter,
may result in over correction of the parameter values, a problem which is difficult to
test (Mukundan et al. 2012 ).
Once the geochemical parameters that exhibit non-conservative behavior have
been removed from the list of potential fingerprints, a statistical test is generally
used to identify geochemical properties that are good at discriminating between
sediment from various sources. The most commonly used statistical method is the
Kruskal-Wallis H-test (Collins et al. 1998 , 2001 ; Walling et al. 1999 ), but a wide
range of other methods have also been applied, including the Mann-Whitney U-test
(Carter et al. 2003 ; Porto et al. 2005 ), theWilcoxon rank-sum test (Juracek andZiegler
2009 ), and the Tukey test (Motha et al. 2003 ). A subset of parameters identified
during this step is then selected to define the fingerprint that is assumed to represent
the optimum combination of parameters for discriminating between the sediment
sources or source types (Walling andWoodward 1995 ; Collins et al. 1998 ; Mukundan
et al. 2012 ). This last step often relies on the use of a step-wise discriminant function
analysis (e.g., Evrard et al. 2013 ; Miller et al. 2013 ), although other data reduction
techniques (e.g., Principle Component Analysis) have also been used.