Chapter 14 . Time-Series Prediction of Chlorophyll a in Lakes
factors for algal growth, and ii) they are widely available in water quality
monitoring databases.
Nitrogen and phosphorous are two of the most important nutrients for plant
growth. Phosphorous is frequently a limiting factor for phytoplankton growth in
freshwater ecosystems, and has been widely used to define the trophic conditions
of lakes and reservoirs (e.g. Vollenweider 1970). Nitrogen is not usually limiting
in temperate regions, but becomes increasingly important in tropical water bodies
(Harris 1986). The ratio NIP of these nutrients has been demonstrated to be linked
to patterns in cyanobacterial species succession and hence overal productivity (e.g.
Takamura and Aizaki 1991).
Temperature and water transparency are two obvious candidates for inclusion:
temperature drives the rates of chemical and biological processes, and water
transparency (measured by Secchi disc depth) has clear implications with overall
phytoplankton photosynthesis. Furthermore, the relative competitive advantages
of different phytoplankton species is affected by the underwater light availability
as a result of water transparency in a way that may influence overall productivity.
This model structure does not consider the effect of top down processes on
phytoplankton communities from grazing pressure by fish or zooplankton. Nor
does it consider the very important role that mixing regimes play in the
development of cyanobacterial blooms in particular. These two factors were
omitted from the structure because they were not well represented in all the
databases available to this study. For these reasons it was elected to limit the
model to predicting chlorophyll a as overall phytoplankton abundance rather than
individual species abundance.
The structures of the ANN models are depicted in Figure 14.2. Figure 14.2a
llustrates the same-day model structure, where algal abundance is predicted on the
same day as the independent variables were observed. Figure 14.2b shows the 30
day model which is a time delay structure where inputs lag the output by a single
month. Note that the 30 day model includes the aIgal abundance from the
previous month in the input layer.
Since sampling frequency in the databases available tended to vary
considerably from weekly to monthly it was decided to discretise the database into
monthly averages for the 30 day model (Figure 14.2b). Thus the model predicts
next month's average algal abundance from this month's average values of input
variables.
factors for algal growth, and ii) they are widely available in water quality
monitoring databases.
Nitrogen and phosphorous are two of the most important nutrients for plant
growth. Phosphorous is frequently a limiting factor for phytoplankton growth in
freshwater ecosystems, and has been widely used to define the trophic conditions
of lakes and reservoirs (e.g. Vollenweider 1970). Nitrogen is not usually limiting
in temperate regions, but becomes increasingly important in tropical water bodies
(Harris 1986). The ratio NIP of these nutrients has been demonstrated to be linked
to patterns in cyanobacterial species succession and hence overal productivity (e.g.
Takamura and Aizaki 1991).
Temperature and water transparency are two obvious candidates for inclusion:
temperature drives the rates of chemical and biological processes, and water
transparency (measured by Secchi disc depth) has clear implications with overall
phytoplankton photosynthesis. Furthermore, the relative competitive advantages
of different phytoplankton species is affected by the underwater light availability
as a result of water transparency in a way that may influence overall productivity.
This model structure does not consider the effect of top down processes on
phytoplankton communities from grazing pressure by fish or zooplankton. Nor
does it consider the very important role that mixing regimes play in the
development of cyanobacterial blooms in particular. These two factors were
omitted from the structure because they were not well represented in all the
databases available to this study. For these reasons it was elected to limit the
model to predicting chlorophyll a as overall phytoplankton abundance rather than
individual species abundance.
The structures of the ANN models are depicted in Figure 14.2. Figure 14.2a
llustrates the same-day model structure, where algal abundance is predicted on the
same day as the independent variables were observed. Figure 14.2b shows the 30
day model which is a time delay structure where inputs lag the output by a single
month. Note that the 30 day model includes the aIgal abundance from the
previous month in the input layer.
Since sampling frequency in the databases available tended to vary
considerably from weekly to monthly it was decided to discretise the database into
monthly averages for the 30 day model (Figure 14.2b). Thus the model predicts
next month's average algal abundance from this month's average values of input
variables.