HR: 16:00h
AN: NB44B-03    [Abstracts]
TI: Taxonomic Specificity or Indices? Aquatic Community Data or Water Chemistry? Applying Artificial Neural Network Models to a Variety of Aquatic Monitoring Approaches
AU: * Zaradic, P
EM: pzaradic@sas.upenn.edu
AF: Stroud Water Research Center, 970 Spencer Road, Avondale, PA 19311 United States
AU: Jackson, J K
EM: jkjackson@stroudcenter.org
AF: Stroud Water Research Center, 970 Spencer Road, Avondale, PA 19311 United States
AB: This study evaluates the potential use of artificial neural network (ANN) modeling for freshwater conservation planning based on stream macroinvertebrate and water chemistry data from three years (2000-2002) at 60 sites in the New York City drinking water watershed. ANNs can learn from examples, generalize to novel situations, tolerate noisy data and approximate any continuous function, making ANNs ideal for modeling the complexities of aquatic community data. ANNs based on several levels of taxonomic resolution are compared to determine the gain in predictive capacity with increasing taxonomic specificity. In addition an ANN model using water chemistry data is compared to the effectiveness of using aquatic macroinvertebrate community data. The ANN models were trained to independently identify the impact of percent agriculture, impervious surface and forested cover. Set aside datasets validated the capacity of the ANN models to correctly predict the three landcover impacts. Models using species indices such as EPT richness, HBI, or water quality scores had relatively low predictive power. A combination of 73 baseline water chemistry variables demonstrated greater predictive power than family level data but by far the best model used species level taxonomy.
DE: 1803 Anthropogenic effects
DE: 4815 Ecosystems, structure and dynamics
SC: North American Benthological Society [NB]
MN: 2005 Joint Assembly