Chapter 18 . Identification of Marine Microalgae
359
18.2
Materials and Methods
18.2.1
Pulse Shape Extraction
CytoBuoy pulse shape data for 36 small unicellular phytoplankton species in pure
culture (Table 18.1) were collected during the course of the AIMS project (CEC
grant no. MAS3-CT97-0080). These species were selected from a much larger
database of species as the only Each species was represented by four data files
containing 64kB of 8-bit sampie values. Pulse shape data for 1000 particles per
species were extracted from these files by purpose-written software. The length of
each pulse, log peak value and log mean value were found, and particles for which
the FSC pulse was less than four sampie values in length (21.lm) were rejected. The
pulses were normalised using a cubic spline interpolation procedure (Press et al.
1992) to a standard number of sampies and to unit integral value. This
normalisation procedure effectively separates out information describing pulse
shape from global pulse characteristics such as length and area, allowing
independent assessment of the contribution of each type of information to species
discrimination. The number of sampies used to represent each pulse was chosen to
be apower of two plus one, for subsequent compatibility with the frequency-space
decomposition methods used; 33 sampies was found to be an adequate description
of the pulse shapes for the species in this study, although other species may
require more (e.g. chain-formers; Fig. 18.1a).
18.2.2
Data Filtering
The data for each species were plotted and pulses corresponding to particles with
low integral red fluorescence (a measure of chlorophyll content) were removed;
such partieles are frequently present in phytoplankton cultures and typically
represent cellular debris, dead cells or bacterial contamination of the culture.
18.2.3
Data Transformation
A discrete eosine transformation (DCT) procedure (Press et al. 1992) was used to
decompose all pulses into a frequency space representation, expressing the
original pulse as the superposition (sum) of a number of separate components
taking the form of eosine waves with differing amplitude and frequency. The
reason for this is twofold. Such a representation effectively separates out elements
of structure due to the presence of features with different length seal es in the
359
18.2
Materials and Methods
18.2.1
Pulse Shape Extraction
CytoBuoy pulse shape data for 36 small unicellular phytoplankton species in pure
culture (Table 18.1) were collected during the course of the AIMS project (CEC
grant no. MAS3-CT97-0080). These species were selected from a much larger
database of species as the only Each species was represented by four data files
containing 64kB of 8-bit sampie values. Pulse shape data for 1000 particles per
species were extracted from these files by purpose-written software. The length of
each pulse, log peak value and log mean value were found, and particles for which
the FSC pulse was less than four sampie values in length (21.lm) were rejected. The
pulses were normalised using a cubic spline interpolation procedure (Press et al.
1992) to a standard number of sampies and to unit integral value. This
normalisation procedure effectively separates out information describing pulse
shape from global pulse characteristics such as length and area, allowing
independent assessment of the contribution of each type of information to species
discrimination. The number of sampies used to represent each pulse was chosen to
be apower of two plus one, for subsequent compatibility with the frequency-space
decomposition methods used; 33 sampies was found to be an adequate description
of the pulse shapes for the species in this study, although other species may
require more (e.g. chain-formers; Fig. 18.1a).
18.2.2
Data Filtering
The data for each species were plotted and pulses corresponding to particles with
low integral red fluorescence (a measure of chlorophyll content) were removed;
such partieles are frequently present in phytoplankton cultures and typically
represent cellular debris, dead cells or bacterial contamination of the culture.
18.2.3
Data Transformation
A discrete eosine transformation (DCT) procedure (Press et al. 1992) was used to
decompose all pulses into a frequency space representation, expressing the
original pulse as the superposition (sum) of a number of separate components
taking the form of eosine waves with differing amplitude and frequency. The
reason for this is twofold. Such a representation effectively separates out elements
of structure due to the presence of features with different length seal es in the