CHAPTER 22 . (hemometric Applications to Seawater Analysis
of variables and their possible interactions seems the most suitable for a better knowledge of the marine system.
Literature offers examples of chemometric treatments that make it possible to obtain information from chemical data on the whole behaviour of sea waters and related
systems (sediments, particulate and atmosphere). In fact the chemometric approach
allows researchers to make a screening of the most significant variables and to find out
their latent correlation, thus leading in general to a better knowledge of the ecosystem.
The data evaluated from samples of sea water, aerosol, particulate, sediments and
marine flora and fauna collected all around the world and concerning both natural and
anthropogenic components present in the sea system have been treated by pattern recognition methods such as: principal components analysis (PCA), factor analysis (FA),
poly topic vector analysis (PVA), hierarchical (HCA) and non hierarchical (NHCA) clustering methods, fuzzy clustering methods (FCA), soft independent modelling of class
analogy (SIM CA) classification method and non parametric k * -nearest neighbour (k * -NN).
Beside these kinds of applications, chemometric treatments are largely used also
during the development of analytical methods. In particular techniques of experimental design and partial least squares (PLS) have been employed for the optimization of
new methods for the determination of pollutants in complex matrices, as seawaters
and sediments. Examples concerning the analysis of pesticides, PAHs, PCBs and other
chlorinated compounds are reported.
22.2
Chemometric Applications to Marine Samples
22.2.1
Classification Techniques
22.2.1.1
Sea Water
Pollution from polychlorinated biphenyls (PCBs) was studied in seawater and particulate samples collected in the San Francisco Estuary (Jarman et a1.1997), the largest one
on the coast of California: its drainage basin area having an inflow of 600 m 2 sec-I,
mainly from the Sacramento and San Joaquin River. In addition to fresh-water inflow,
ocean tides and winds bring ocean waters into the bay through the Golden Gate. The
northern bay is a partially mixed estuary and the southern bay is a tidal lagoon: this
distinction results in dramatically different biogeochemical processes in the two sections of the bay. The San Francisco Estuary is also a highly impacted area and several
areas of the bay are highly industrialised. The matrix of the experimental data is formed
by 14 samples and 32 PCB congeners for each of the six cruises performed over the years
1993-1995. Poly topic vector analysis (PVA) was designed to determine three important parameters in the analysis of mixtures: (i) the number of sources (end-members
or chemical fingerprints) contributing to the mixture, (ii) the chemical composition
of each-end member and (iii) the relative contribution of each-end member in each
sample (Ehrich and Full 1987). PYA method is relatively new in environmental and
chemometric literature but extensively used in geological sciences. It is a self-training
classification method, because no training data set of known or suspected sources is
of variables and their possible interactions seems the most suitable for a better knowledge of the marine system.
Literature offers examples of chemometric treatments that make it possible to obtain information from chemical data on the whole behaviour of sea waters and related
systems (sediments, particulate and atmosphere). In fact the chemometric approach
allows researchers to make a screening of the most significant variables and to find out
their latent correlation, thus leading in general to a better knowledge of the ecosystem.
The data evaluated from samples of sea water, aerosol, particulate, sediments and
marine flora and fauna collected all around the world and concerning both natural and
anthropogenic components present in the sea system have been treated by pattern recognition methods such as: principal components analysis (PCA), factor analysis (FA),
poly topic vector analysis (PVA), hierarchical (HCA) and non hierarchical (NHCA) clustering methods, fuzzy clustering methods (FCA), soft independent modelling of class
analogy (SIM CA) classification method and non parametric k * -nearest neighbour (k * -NN).
Beside these kinds of applications, chemometric treatments are largely used also
during the development of analytical methods. In particular techniques of experimental design and partial least squares (PLS) have been employed for the optimization of
new methods for the determination of pollutants in complex matrices, as seawaters
and sediments. Examples concerning the analysis of pesticides, PAHs, PCBs and other
chlorinated compounds are reported.
22.2
Chemometric Applications to Marine Samples
22.2.1
Classification Techniques
22.2.1.1
Sea Water
Pollution from polychlorinated biphenyls (PCBs) was studied in seawater and particulate samples collected in the San Francisco Estuary (Jarman et a1.1997), the largest one
on the coast of California: its drainage basin area having an inflow of 600 m 2 sec-I,
mainly from the Sacramento and San Joaquin River. In addition to fresh-water inflow,
ocean tides and winds bring ocean waters into the bay through the Golden Gate. The
northern bay is a partially mixed estuary and the southern bay is a tidal lagoon: this
distinction results in dramatically different biogeochemical processes in the two sections of the bay. The San Francisco Estuary is also a highly impacted area and several
areas of the bay are highly industrialised. The matrix of the experimental data is formed
by 14 samples and 32 PCB congeners for each of the six cruises performed over the years
1993-1995. Poly topic vector analysis (PVA) was designed to determine three important parameters in the analysis of mixtures: (i) the number of sources (end-members
or chemical fingerprints) contributing to the mixture, (ii) the chemical composition
of each-end member and (iii) the relative contribution of each-end member in each
sample (Ehrich and Full 1987). PYA method is relatively new in environmental and
chemometric literature but extensively used in geological sciences. It is a self-training
classification method, because no training data set of known or suspected sources is