HR: 11:15h
AN: NB42B-04    [Abstracts]
TI: Predicting Diatom Assemblages in Minimally-Impacted Streams Using a New Hybrid Modelling Approach
AU: * Rollins, S L
EM: rollins8@msu.edu
AF: Department of Zoology, Michigan State University, 203 Natural Science Building, East Lansing, MI 48824-1115 United States
AU: Stevenson, R J
EM: rjstev@msu.edu
AF: Department of Zoology, Michigan State University, 203 Natural Science Building, East Lansing, MI 48824-1115 United States
AU: Hawkins, C P
EM: chuck.hawkins@usu.edu
AF: Department of Aquatic, Watershed, and Earth Resources, Utah State University, 5210 Old Main Hill, Logan, UT 84322-5210 United States
AU: Manoylov, K M
EM: manoylov@msu.edu
AF: Department of Zoology, Michigan State University, 203 Natural Science Building, East Lansing, MI 48824-1115 United States
AU: Olson, J R
EM: jrolson@cc.usu.edu
AF: Department of Aquatic, Watershed, and Earth Resources, Utah State University, 5210 Old Main Hill, Logan, UT 84322-5210 United States
AU: Hill, R A
EM: ryanhill@cc.usu.edu
AF: Department of Aquatic, Watershed, and Earth Resources, Utah State University, 5210 Old Main Hill, Logan, UT 84322-5210 United States
AB: The expected taxonomic composition in minimally-impacted streams can be predicted using linear discriminant analysis (LDA) and environmental variables that are relatively independent of human activities. However, many biological responses to ecological gradients are non-linear. A hybrid approach combining LDA with nonlinear predictions made by recursive partitioning was applied using a Bayesian methodology. Diatom assemblages were used to classify reference sites throughout the western United States using a self-organizing map clustering technique. Predictive models for the classes were then developed using LDA, recursive partitioning, and the new hybrid method. Predictive models were evaluated using the ratio of correct-to-incorrect classifications, weighted by the confidence of each prediction. The hybrid method outperformed both LDA and recursive partitioning. Confidence-weighted ratios were 1.39 for the hybrid method, 1.15 for recursive partitioning, and 1.05 for LDA. Temperature was the most important linear predictor of diatom assemblages. Stream flashiness and precipitation were among the most important nonlinear predictors. Our results suggest that nonlinear determinants of diatom assemblages are important and that our hybrid method improves predictions of the expected conditions in reference habitats that are used in environmental assessments.
DE: 1871 Surface water quality
DE: 1899 General or miscellaneous
DE: 9820 Techniques applicable in three or more fields
DE: 9903 NABS Student Award Methodology
SC: North American Benthological Society [NB]
MN: 2005 Joint Assembly