Chon, Tae-Soo, Inn Sil Kwak and Young Seuk Park
Dept. of Biology, Pusan National University, Pusan
609-735, KOREA
To
community data, sampled in a regular interval on the long-term basis,
artificial nerual networks were implemented to extract information
characterizing community patterns.
The Kohonen network and Adaptive Resonance Theory were utilized in
combination in learning benthic macroinvertebrate communities in streams of the
Suyong River collected monthly for three years. In static manner, by regarding each monthly collection as a
separate sample unit, communities were grouped into similar patterns after the
training with the network.
Subsequently changes in communities in a sequenc of samplings (e.g.,
two-months, four-months, etc.) were also given as input to the networks to
train community dynamics. After
training it was possible, both in static and dynamic mode, to recognize new
data set on the on-time basis as sampling proceeded. Through the comparative study on benthic macroinvertebrates
with these learning processes, it was shown that patterns of commnity changes
in chironomids were diverging, being more sensitive to the impacts of internal
or external factors, while those of benthic macro-invertebrates in total
appeared to be more persistent.
Key
Words: Patterning community dynamics, Artificial neural network, Adaptive
Resonance Theory, Kohonen network, Benthic macroinvertebrates, Chironomids