Comparison of Coarse versus Fine Scale Sampling on Statistical Modeling of Landscape Effects and Assessment of Fish Assemblages of the Muskegon River, Michigan

摘要

We used data sets of differing geographic extents and sampling intensities to examine how data structure affects the outcome of biological assessment. An intensive sampling (n = 97) of the Muskegon River basin provided our example of fine scale data,while two regional and statewide data sets (n = 276,310) represented data sets of coarser geographic scales. We constructed significant multiple linear regression models (R2 from 21% to 79%) to predict expected fish assemblage metrics (total fish, game fish, intolerant fish, and benthic fish species richness) and to regionally normalize Muskegon basin samples. We then examined the sensitivity of assessments based on each of five data sets with differing geographic extents to landscape stressors (urban and agricultural land use, dam density, and point source discharges). Assessment scores generated from the different data extents were significantly correlated and suggested that the Muskegon basin was generally in good condition. However, using coarser scale data extents to determine reference conditions resulted in greater sensitivity to land-use stressors (urban and agricultural land use). This was due in part to significant covariance between land use and drainage area in the fine scale data set. Ourresults show that the scale of data used to determine reference condition can significantly influence the results of a biological assessment. The training data sets with broader spatial range appeared to produce the most sensitive and accurate catchmentassessment. A covariance structure analysis using a data set with broad spatial range suggested that impounded channels and point source discharges have the strongest negative effects on intolerant fish richness in the Muskegon River basin, which provides a focus for conservation, mitigation, and rehabilitation opportunities.