Chapter 14
A Generic Artificial Neural Network Model for
Short Term Predictions of Aigal Blooms in Lakes
and Reservoirs
H. Wilson . F. Recknagel
14.1
Introduction
Sudden growth of algal biomass, leading to the formation of water blooms, are a
world wide environmental problem causing detrimental effects on water quality
such as taste, odour and potential toxicity. Effective management of the algal
bloom problem requires that appropriate decisions be made with regards to the
implementation of strategic and operational control techniques in a timely manner.
However, decision making in these respects is hampered by the complexity of the
relationships between nutrients, primary production, and other physical,
biological, and chemical processes in freshwater ecosystems (Ferguson 1997).
Thus there is a need for predictive models that account for these complexities to
aid strategic and operational decision making (National Rivers Authority 1990;
Parliament of Australia 1993; Ferguson 1997). Whilst existing inductive and
deductive modelling approaches are useful decision making tools, they are limited
in terms of their ability to predict the timing, magnitude and algal species of
significant bloom events (Recknagel et al. 1997; Recknagel and Wilson 2000).
Since the mid 1990's, there has been growing interest in applications of
machine learning techniques such as artificial neural networks (ANNs) and
genetic algorithms (GAs) to algal bloom modelling. These techniques differ from
conventional process-based and statistical models in that no model is assumed apriori. Instead the models' decision boundaries are induced entirely from data.
Intuitively, this property heraids two key theoretical advantages over conventional
inductive and deductive modelling approaches in the context of limnological
modelling applications:
1. There is no requirement for the symbolic expression of domain knowledge and
associated parameter estimation. Therefore, ANNs are potentiaUy less resource
intensive in terms of domain expertise and data for realistic parameter
estimations than deductive approaches such as process-based deterministic
models.
2. The inherent non-linearity of ANNs means that they are not constrained by the
simplifications of conventional statistical approaches. Therefore, ANNs
A Generic Artificial Neural Network Model for
Short Term Predictions of Aigal Blooms in Lakes
and Reservoirs
H. Wilson . F. Recknagel
14.1
Introduction
Sudden growth of algal biomass, leading to the formation of water blooms, are a
world wide environmental problem causing detrimental effects on water quality
such as taste, odour and potential toxicity. Effective management of the algal
bloom problem requires that appropriate decisions be made with regards to the
implementation of strategic and operational control techniques in a timely manner.
However, decision making in these respects is hampered by the complexity of the
relationships between nutrients, primary production, and other physical,
biological, and chemical processes in freshwater ecosystems (Ferguson 1997).
Thus there is a need for predictive models that account for these complexities to
aid strategic and operational decision making (National Rivers Authority 1990;
Parliament of Australia 1993; Ferguson 1997). Whilst existing inductive and
deductive modelling approaches are useful decision making tools, they are limited
in terms of their ability to predict the timing, magnitude and algal species of
significant bloom events (Recknagel et al. 1997; Recknagel and Wilson 2000).
Since the mid 1990's, there has been growing interest in applications of
machine learning techniques such as artificial neural networks (ANNs) and
genetic algorithms (GAs) to algal bloom modelling. These techniques differ from
conventional process-based and statistical models in that no model is assumed apriori. Instead the models' decision boundaries are induced entirely from data.
Intuitively, this property heraids two key theoretical advantages over conventional
inductive and deductive modelling approaches in the context of limnological
modelling applications:
1. There is no requirement for the symbolic expression of domain knowledge and
associated parameter estimation. Therefore, ANNs are potentiaUy less resource
intensive in terms of domain expertise and data for realistic parameter
estimations than deductive approaches such as process-based deterministic
models.
2. The inherent non-linearity of ANNs means that they are not constrained by the
simplifications of conventional statistical approaches. Therefore, ANNs