Fishes of lower Michigan rivers: Distribution patterns, abundance models, and causal relationships.

Abstract

Lower Michigan has a diverse array of rivers, many of which have fish assemblages that are poorly understood. I developed tools to facilitate description, understanding, and prediction of patterns in fish assemblage structure across this region. I used cluster analysis to group the 69 most common fishes based on patterns of co-occurrence. The seventeen clusters explained about 39% of the variation in fish assemblage structure and provided a reasonable, albeit simplified picture of general associations of fishes in lower Michigan streams. Ordination of these clusters on landscape-scale axes of 90% exceedence flow yield and catchment area provided an empirically derived framework for comparing Michigan streams and for assessing the physical and biological potential of different river reaches. I also developed multiple linear regression models for the 68 most common fishes to determine the extent to which spatial patterns in standing crops of fishes in lower Michigan rivers could be predicted from other variables. Significant regression models were developed for each species based on: (1) all sites with assemblage-level data and; (2) sites where the species occurred. The latter set of models explained more spatial variation in fish abundance than the former set (average of 43% vs. 26% of variance explained), included fewer variables, and had lower estimation error. Catchment area, July mean temperature, channel gradient, total phosphorus, and substrate variables were significant most frequently in both sets of models. Variables characterizing human land use and habitat connectivity were also important to many fishes. These findings indicated that prediction of fish assemblage structure, at a coarse-scale of resolution, appears feasible. Using covariance structure analysis, I found that landscape-scale variables (catchment area, channel gradient, land use, and coarse-textured geologic deposits) explained a 16--84% of spatial variation in, and had significant direct and indirect effects on, local habitat features in lower Michigan rivers. Additional covariance structure analyses with brook trout <italic>Salvelinus fontinalis</italic>, brown trout <italic>Salmo trutta</italic>, and smallmouth bass <italic>Micropterus dolomieu</italic> demonstrated the over-riding influence of sample set selection on pattern-process relationships, and helped reconcile issues regarding the relative importance of habitat variables and biotic interactions to standing crops of these species.