Relations of Angler Catch and Per Unit Effort and Yield to Limnological Variables in Wisconsin Lakes
Potter, David F.
University of Wisconsin-Stevens Point, College of Natural Resources
MetadataShow full item record
Linear predictive models were developed for Wisconsin lakes that use easily-measured limnological variables to predict summer angler CPUE (number of fish caught/angler hour) and yield (number of fish harvested/ha) for six species of sport fish: muskellunge (Esox masquinongy), northern pike (Esox lucius), largemouth bass (Micropterus salmoides), smallmouth bass (Micropterus dolomieu), walleye (Stizostedion vitreum), and yellow perch (Perea flavescens). Creel and limnological data for 163 Wisconsin lakes were partitioned into two data sets to develop (108 lakes) and validate (86 lakes) predictive models (31 lakes overlapped). To reveal important relations, model development lakes were first classified according to similarities among seven limnological variables: area of adjacent wetlands, surface area, conductivity, growing season, mean depth, shoreline development factor, and percent of the surface area less than 1 meter deep. K-means cluster analysis produced five lake clusters that accounted for 42% of the variation in the limnological data. Discriminant analysis re-classified the lakes into five lake groups with an overall misclassification rate of 15%. These lake groups were different from each other: Group 1 contained lakes that were typically the smallest and shallowest; Group 2 were deepest; Group 3 had the highest conductivity, ·lowest sinuosity, and longest growing seasons; Group 4 were the largest with the greatest amount of adjacent wetlands and highest sinuosity; Group 5 had lakes with the shortest growing season. Indicator variable regression analysis developed, and tested for differences among, lake-group predictive models. In most cases, the initial predictive models for lake groups did not differ from each other. In all, 17 species-specific models explained from 6 to about 72% of the variation in CPUE and yield from five limnological variables: area, conductivity, growing season, mean depth, and percent shallow depth. Model validation lakes were classified with the discriminant functions, and the mean square of predictive errors (MSPR) was derived for comparison with the mean square of errors (MSE) of initial models. Four of the 17 models did not have significant differences between MSPR's and MSE's, and were thus validated. Data sets were then pooled to derive a second set of more robust models. Nine models were developed; however, they only accounted for 4 - 29% of the variation with the same five predictors, and did not detect differences in slopes and/or intercepts among lake groups. Although most of the regression models are of little practical use in predicting species-specific CPUE and yield, they provide a framework for further investigations that may involve use of different dependent variables, incorporating more elaborate independent variables, and classifying lakes differently. The lake classification delineated lake groups that were limnologically different, and therefore, may serve other management applications.