128
T.-S. Chon . Y.S. Park' I.-S. Kwak . E.Y. Cha
Hawkes 1979; Sladecek 1979; Tittizer and Kothe 1979; HeIlaweIl 1986). Many
useful biological indicators based on benthic macroinvertebrates such as TBI,
BMWP have been developed (e.g., SpeIlerberg 1991).
Benthic macroinvertebrates are generally cosmopolitan and diverse. The
parameters on community structure such as diversity and dominance could be
effectively used for indicating water quality as weIl as for expressing ecological
status. At the same time, each different group of macroinvertebrates could be the
indicator to specific toxic effects. The physicochemical analyses are specific and
accurate, however they are sometimes not integrative, would provide only local
information, and are generally expensive. Monitoring by biological communities
could be a good compensator for the physicochemical methods for indicating
water quality (HeIlaweIlI986; Tittizer and Kothe 1979; Welch and LindeIlI992).
Artificial Neural Networks and Non-Linear Data
Through field survey, data for community dynamics are usually obtained from
various sampie sites on the regular basis, and are accumulated during the survey
period Ce.g., Fig. 8.1). Since communities consist of many species and vary in
nonlinear fashions, they are complex and difficult to analyze. There have been
numerous accounts of statistical analyses on communities through conventional
multivariate analyses Ce.g., Bunn et al. 1986; Legendre and Legendre 1987;
Ludwig and Reynolds 1988; Quinn et al. 1991). The researches have been usually
directed to classification of communities to ordination of multivariate data through
eigen analyses. However, conventional statistical methods are mainly limited to
linear data (Ludwig and Reynolds 1988), and are not flexible in many aspects, for
instance, data handling (e.g., missing sampies) and predicting dynamics.
Artificial neural networks solve this problem of complexity in community data.
Artificial neural networks are parallel and distributed information extraction
processors, have adaptive and self-organizing properties, and are consequently
feasible in handling non linear data (Lippmann 1987; Hecht-Nielsen 1990;
Zuradal992; Haykin 1994). Since the neural computation system was proposed
by McCulloch and Pitts (1943) in the forties, artificial neural networks have been
rapidly developed in extracting information of complex and nonlinear phenomena
in a wide spectrum in the field of machine intelligence in the eighties (e.g.,
Lippmann 1987; Wasserman 1989; Hecht-Nielsen 1990; Zurada 1992; Haykin
1994». In ecology, artificial neural networks have been used for classifying
groups (e.g., Chon et al. 1996; Levine et al. 1996), and for patteming complex
relationships (e.g., Lek et al. 1996; Huntingford and Cox 1996; Tuma et al. 1996).
On macro-invertebrates in aquatic ecosystems, training with artificial neural
networks have been conducted on grouping and community dynamics (Chon et al.
1996, 2000a, 2000b, 2000c, 2001; Park et al. 2001a, 2001b; Brosse et al. 2001).
Implementations of artificial neural networks to ecology have been extensively
reviewed by Lek and Guegan (1999, 2000).
T.-S. Chon . Y.S. Park' I.-S. Kwak . E.Y. Cha
Hawkes 1979; Sladecek 1979; Tittizer and Kothe 1979; HeIlaweIl 1986). Many
useful biological indicators based on benthic macroinvertebrates such as TBI,
BMWP have been developed (e.g., SpeIlerberg 1991).
Benthic macroinvertebrates are generally cosmopolitan and diverse. The
parameters on community structure such as diversity and dominance could be
effectively used for indicating water quality as weIl as for expressing ecological
status. At the same time, each different group of macroinvertebrates could be the
indicator to specific toxic effects. The physicochemical analyses are specific and
accurate, however they are sometimes not integrative, would provide only local
information, and are generally expensive. Monitoring by biological communities
could be a good compensator for the physicochemical methods for indicating
water quality (HeIlaweIlI986; Tittizer and Kothe 1979; Welch and LindeIlI992).
Artificial Neural Networks and Non-Linear Data
Through field survey, data for community dynamics are usually obtained from
various sampie sites on the regular basis, and are accumulated during the survey
period Ce.g., Fig. 8.1). Since communities consist of many species and vary in
nonlinear fashions, they are complex and difficult to analyze. There have been
numerous accounts of statistical analyses on communities through conventional
multivariate analyses Ce.g., Bunn et al. 1986; Legendre and Legendre 1987;
Ludwig and Reynolds 1988; Quinn et al. 1991). The researches have been usually
directed to classification of communities to ordination of multivariate data through
eigen analyses. However, conventional statistical methods are mainly limited to
linear data (Ludwig and Reynolds 1988), and are not flexible in many aspects, for
instance, data handling (e.g., missing sampies) and predicting dynamics.
Artificial neural networks solve this problem of complexity in community data.
Artificial neural networks are parallel and distributed information extraction
processors, have adaptive and self-organizing properties, and are consequently
feasible in handling non linear data (Lippmann 1987; Hecht-Nielsen 1990;
Zuradal992; Haykin 1994). Since the neural computation system was proposed
by McCulloch and Pitts (1943) in the forties, artificial neural networks have been
rapidly developed in extracting information of complex and nonlinear phenomena
in a wide spectrum in the field of machine intelligence in the eighties (e.g.,
Lippmann 1987; Wasserman 1989; Hecht-Nielsen 1990; Zurada 1992; Haykin
1994». In ecology, artificial neural networks have been used for classifying
groups (e.g., Chon et al. 1996; Levine et al. 1996), and for patteming complex
relationships (e.g., Lek et al. 1996; Huntingford and Cox 1996; Tuma et al. 1996).
On macro-invertebrates in aquatic ecosystems, training with artificial neural
networks have been conducted on grouping and community dynamics (Chon et al.
1996, 2000a, 2000b, 2000c, 2001; Park et al. 2001a, 2001b; Brosse et al. 2001).
Implementations of artificial neural networks to ecology have been extensively
reviewed by Lek and Guegan (1999, 2000).