Chapter 9
Elucidation of Hypothetical Relationships
between Habitat Conditions and
Macroinvertebrate Assemblages in Freshwater
Streams by Artificial Neural Networks
H. Hoang· F. Recknagel· J. Marshall . S. Choy
9.1
Introduction
It has been widely demonstrated that interactions among chemical and physical
processes create environmental conditions at a range of sc ales that strongly
influence the distribution and abundance of lotic biota, and thus the composition
of macroinvertebrate assemblages (e.g. Hynes 1970). Many studies have
identified substrate composition, complexity and heterogeneity as major
determinants of in-stream biota (e.g. Downes et al. 1998). Other abiotic factors
such as flow velocity (e.g. Barmuta 1990) and water chemistry (e.g. Bunn et al.
1986) have also been found to influence biotic composition.
The insight that local species assemblages are a reflection of local
environmental conditions is fundamental to biomonitoring programmes
increasingly incorporated into water resource management practices throughout
the world. Statistical models have been developed to predict the occurrence of
macroinvertebrate taxa based on their association with environmental variables
(e.g. Wright 1995; Reynoldson et al. 1997; Simpson et al. 1997). Machine
leaming techniques, such as artificial neural networks (ANN), have recently been
applied to this problem and show promise to provide greater predictive capacity
than statistical modelling techniques (Chon et al. 1996; Walley and Fontama 1998;
Pudmenzky et al. 1998; Schleiter et al. 1999).
In the context of this chapter aseries of ANN models were developed based on
both the 'clean water' (Huong et al. 2001) and the 'dirty water' approach (Huong
2001) which accurately predicted the presence and absence of most common
macroinvertebrate taxa in the stream system of Queensland, Australia. The
referential 'clean water' approach (Reynoldson et al. 1997) aimed at the prediction
of fauna at impacted sites assuming they were unimpacted. The 'dirty water'
Elucidation of Hypothetical Relationships
between Habitat Conditions and
Macroinvertebrate Assemblages in Freshwater
Streams by Artificial Neural Networks
H. Hoang· F. Recknagel· J. Marshall . S. Choy
9.1
Introduction
It has been widely demonstrated that interactions among chemical and physical
processes create environmental conditions at a range of sc ales that strongly
influence the distribution and abundance of lotic biota, and thus the composition
of macroinvertebrate assemblages (e.g. Hynes 1970). Many studies have
identified substrate composition, complexity and heterogeneity as major
determinants of in-stream biota (e.g. Downes et al. 1998). Other abiotic factors
such as flow velocity (e.g. Barmuta 1990) and water chemistry (e.g. Bunn et al.
1986) have also been found to influence biotic composition.
The insight that local species assemblages are a reflection of local
environmental conditions is fundamental to biomonitoring programmes
increasingly incorporated into water resource management practices throughout
the world. Statistical models have been developed to predict the occurrence of
macroinvertebrate taxa based on their association with environmental variables
(e.g. Wright 1995; Reynoldson et al. 1997; Simpson et al. 1997). Machine
leaming techniques, such as artificial neural networks (ANN), have recently been
applied to this problem and show promise to provide greater predictive capacity
than statistical modelling techniques (Chon et al. 1996; Walley and Fontama 1998;
Pudmenzky et al. 1998; Schleiter et al. 1999).
In the context of this chapter aseries of ANN models were developed based on
both the 'clean water' (Huong et al. 2001) and the 'dirty water' approach (Huong
2001) which accurately predicted the presence and absence of most common
macroinvertebrate taxa in the stream system of Queensland, Australia. The
referential 'clean water' approach (Reynoldson et al. 1997) aimed at the prediction
of fauna at impacted sites assuming they were unimpacted. The 'dirty water'