420
M.e. Gennaro· S. Angelino
the diagonal elements of the correlation matrix by the square communalities obtained
from the factor loadings and continuing the process until the difference in the traces
of two successive correlation matrices were smaller than a prefixed threshold. Nine
factors were identified as the most reasonable estimate of the data. Most of the loadings are unspecific: so for instance chlorinated species exhibit a distribution behaviour
that is roughly independent on their origin. In general, however, source inputs are better
differentiated by FA than by PCA. However, since the data set used to describe the system has unsuitable dimension (102 variables for 40 samples), the data set was divided
between biogenic and anthropogenic markers and each group was again evaluated with
PCA and FA. PCA analysis did not lead to higher definitions of geochemical sources,
while FA analysis for the biogenic subgroup evidenced four factors as the most likely
to estimate the reduced data set: two main biogenic factors correspond to algae origins, the first related to diatom sources and the second to green algae and dinoflagellates, while two factors differentiate between river and irrigation channel samples. The
characterisation of anthropogenic data are more difficult, also because the diversity
of pollutant discharge sites in the Ebre Delta.
SIMCA, a principal component chemometric modelling program, was used to examine complex mixtures of PCB residues in fish and turtles (Stalling et al. 1987) obtained from Tinicum National Environmental Center in Philadelphia and to compare
the results with Aroclor mixtures. The data set consists of three classes: Aroclor mixture (14 samples), turtles (7 samples) and fish (6 samples). The chemometric treatment
of SIMCA gave three well differentiated classes, and PCA individuated 3 PCs able to
explain more than 93% of the total variance. These results confirm that PCB residues
in turtles and fish are markedly different from Aroclor standards or their mixtures,
even though fish residue profiles were closer to Aroclor than the turtle residue.
More than 150 samples of particulate, sediments, isopods, zoo- and phytoplankton,
mysidis, perch, fourhorned sculpin, herring, salmon, salmon trout and whitefish were
collected in the autumn 1991 and spring 1992 in the Gulf of Bothnia, between the coasts
of Sweden and Finland, to study the accumulation of PCBs and other organochlorine
contaminants in benthic and pelagic marine food chains (Van Bavel et al.1994). In the
treatment of the data of PCBs for sediments and for the amphipod Monoporeia affinis,
PCA was employed. The first two PCs describe 85% of the total variance: PC1 is able to
describe for the same station a concentration gradient from spring (lower concentration) to autumn (higher concentration) and a gradually higher level from Hornslamnder station to the others. This difference was explained as an indication of different sources of pollution of one sampling site with respect to the other three: this site is
located in a sub arctic region where run off water from the arctic region is collected.
Large seasonal fluctuations in PCB levels are then generally observed from the autumn
(high concentration) and spring (low) for both sediment and amphipod samples.
22.2.2
Optimization Methods
Chemometric methods can be applied to collect from the chemical data more information about the marine system, as in the examples reported above, as well as for the
development and the optimization of adequate methods for sample pretreatment and
analyte recovery. The large number of species present in sea water plays in fact a com-
M.e. Gennaro· S. Angelino
the diagonal elements of the correlation matrix by the square communalities obtained
from the factor loadings and continuing the process until the difference in the traces
of two successive correlation matrices were smaller than a prefixed threshold. Nine
factors were identified as the most reasonable estimate of the data. Most of the loadings are unspecific: so for instance chlorinated species exhibit a distribution behaviour
that is roughly independent on their origin. In general, however, source inputs are better
differentiated by FA than by PCA. However, since the data set used to describe the system has unsuitable dimension (102 variables for 40 samples), the data set was divided
between biogenic and anthropogenic markers and each group was again evaluated with
PCA and FA. PCA analysis did not lead to higher definitions of geochemical sources,
while FA analysis for the biogenic subgroup evidenced four factors as the most likely
to estimate the reduced data set: two main biogenic factors correspond to algae origins, the first related to diatom sources and the second to green algae and dinoflagellates, while two factors differentiate between river and irrigation channel samples. The
characterisation of anthropogenic data are more difficult, also because the diversity
of pollutant discharge sites in the Ebre Delta.
SIMCA, a principal component chemometric modelling program, was used to examine complex mixtures of PCB residues in fish and turtles (Stalling et al. 1987) obtained from Tinicum National Environmental Center in Philadelphia and to compare
the results with Aroclor mixtures. The data set consists of three classes: Aroclor mixture (14 samples), turtles (7 samples) and fish (6 samples). The chemometric treatment
of SIMCA gave three well differentiated classes, and PCA individuated 3 PCs able to
explain more than 93% of the total variance. These results confirm that PCB residues
in turtles and fish are markedly different from Aroclor standards or their mixtures,
even though fish residue profiles were closer to Aroclor than the turtle residue.
More than 150 samples of particulate, sediments, isopods, zoo- and phytoplankton,
mysidis, perch, fourhorned sculpin, herring, salmon, salmon trout and whitefish were
collected in the autumn 1991 and spring 1992 in the Gulf of Bothnia, between the coasts
of Sweden and Finland, to study the accumulation of PCBs and other organochlorine
contaminants in benthic and pelagic marine food chains (Van Bavel et al.1994). In the
treatment of the data of PCBs for sediments and for the amphipod Monoporeia affinis,
PCA was employed. The first two PCs describe 85% of the total variance: PC1 is able to
describe for the same station a concentration gradient from spring (lower concentration) to autumn (higher concentration) and a gradually higher level from Hornslamnder station to the others. This difference was explained as an indication of different sources of pollution of one sampling site with respect to the other three: this site is
located in a sub arctic region where run off water from the arctic region is collected.
Large seasonal fluctuations in PCB levels are then generally observed from the autumn
(high concentration) and spring (low) for both sediment and amphipod samples.
22.2.2
Optimization Methods
Chemometric methods can be applied to collect from the chemical data more information about the marine system, as in the examples reported above, as well as for the
development and the optimization of adequate methods for sample pretreatment and
analyte recovery. The large number of species present in sea water plays in fact a com-