The dominant environmental determinants of aquatic communities have been a persistent topic for many years. Interactions between natural and anthropogenic characteristics within the aquatic environment influence fish communities in complex ways that make the effect of a single characteristic difficult to ascertain. Researchers are faced with the question of how to deal with a large number of variables and complex interrelationships. This study utilized multiple approaches to identify key environmental variables to fish communities of the Fox River Basin in Illinois: Pearson and Spearman correlations, an algorithm based on information theory called mutual information, and a measure of variable importance built into the machine learning algorithm Random Forest. The results are based on a dataset developed for this study, which uses a fish index of biological integrity (IBI) and its ten component metrics as response variables and a range of environmental variables describing geomorphology, stream flow statistics, climate, and both reach-scale and watershed-scale land use as independent variables. Agricultural land use and the magnitude and duration of low flow events were ranked by the algorithms as key factors for the study area. Reach-scale characteristics were dominant for native sunfish, and stream flow metrics were rated highly for native suckers. Regression tree analyses of environmental variables on fish IBI identified breakpoints in percent agricultural land in the watershed (~64%), duration of low flow pulses (~12 days), and 90-day minimum flow (~0.13 cms). The findings should be useful for building predictive models and design of more effective monitoring systems and restoration plans.
Keywords: Fish IBI; Land use; Mutual information; Regression tree; Stream fish; Stream flow.