Chapter 15
Predictive Rules for Phytoplankton Dynamics in
Freshwater Lakes Discovered by Evolutionary
Aigorithms
J. Bobbin . F. Recknagel
15.1
Introduction
Blue-green algae are also called cyanobacteria as they are in many ways like
bacteria: prokaryotic, smaller than 2 JLm and rapidly reproducing by cell division.
At low abundance they are an important component of aquatic food webs but tend
to form outbreaks when favored by still, warm waters rieh in nutrients. Blooms of
blue-green algae can contaminate freshwater by toxins, anoxia, taste and odour
causing high treatment costs in water works, risks for public health and
degradation of freshwater ecosystems.
The problems associated with blue-green algae have prompted long-term
monitoring of physical, chemical and biological parameters of freshwater lakes
and rivers worldwide. Accumulated data from these monitoring activities contain
a wealth of information about properties and behaviors of lake and river
ecosystems that is rarely fully explored.
Machine learning techniques are recognized for being able to discover patterns
in complex databases. They have recently been applied for modeling of ecological
time-series (e.g. Recknagel et al. 1997; Jeong et al. 2001; Whigharn and
Recknagel 2001; Bobbin and Recknagel 2001; Chon et al. 2001; Wilson and
RecknageI2001).
The strength of these techniques is to inductively acquire knowledge on
ecosystem behavior that can be used for forecasting and elucidation. Ca se studies
show that machine learning models can be superior to deductive, process based
models in terms of accuracy of short-term predictions (e.g. Walter et al. 2001;
Barciella et al. 1999). Moreover knowledge acquired by machine learning can
provide unique insights into the causality of ecosystem behavior by means of
scenario and sensitivity analyses (e.g. Recknagel and Wilson 2000; Jeong et al.
2001) or by its direct representation as classification rules (e.g. Bobbin and
Recknagel2001, Recknagel et al. 2002).
Predictive Rules for Phytoplankton Dynamics in
Freshwater Lakes Discovered by Evolutionary
Aigorithms
J. Bobbin . F. Recknagel
15.1
Introduction
Blue-green algae are also called cyanobacteria as they are in many ways like
bacteria: prokaryotic, smaller than 2 JLm and rapidly reproducing by cell division.
At low abundance they are an important component of aquatic food webs but tend
to form outbreaks when favored by still, warm waters rieh in nutrients. Blooms of
blue-green algae can contaminate freshwater by toxins, anoxia, taste and odour
causing high treatment costs in water works, risks for public health and
degradation of freshwater ecosystems.
The problems associated with blue-green algae have prompted long-term
monitoring of physical, chemical and biological parameters of freshwater lakes
and rivers worldwide. Accumulated data from these monitoring activities contain
a wealth of information about properties and behaviors of lake and river
ecosystems that is rarely fully explored.
Machine learning techniques are recognized for being able to discover patterns
in complex databases. They have recently been applied for modeling of ecological
time-series (e.g. Recknagel et al. 1997; Jeong et al. 2001; Whigharn and
Recknagel 2001; Bobbin and Recknagel 2001; Chon et al. 2001; Wilson and
RecknageI2001).
The strength of these techniques is to inductively acquire knowledge on
ecosystem behavior that can be used for forecasting and elucidation. Ca se studies
show that machine learning models can be superior to deductive, process based
models in terms of accuracy of short-term predictions (e.g. Walter et al. 2001;
Barciella et al. 1999). Moreover knowledge acquired by machine learning can
provide unique insights into the causality of ecosystem behavior by means of
scenario and sensitivity analyses (e.g. Recknagel and Wilson 2000; Jeong et al.
2001) or by its direct representation as classification rules (e.g. Bobbin and
Recknagel2001, Recknagel et al. 2002).