Models that Predict Standing Crop of Stream Fish from Habitat Variables: 1950-85

Kurt D. Fausch

Jan 1988 · U.S. Department of Agriculture, Forest Service, Pacific Northwest Research Station

About this ebook

We reviewed mathematical models that predict standing crop of stream fish (number or biomass per unit area or length of stream) from measurable habitat variables and classified them by the types of independent habitat variables found significant, by mathematical structure, and by model quality. Habitat variables were of three types and were measured on different scales in relation to stream channels: variables of drainage basins were measured on the coarsest scale from topographic maps; channel-morphometry and flow variables were measured in the field along transects perpendicular to flow; and habitat-structure, biological, physical, and chemical variables were measured on the finest scale in the field. We grouped the 99 reviewed models by the types of independent variables found significant during model development: (A) primarily drainage basin (5 models), (8) primarily channel morphometry and flow (16 models), (C) primarily habitat structure, biological, physical, and chemical (25 models), (D) a combination of several types of variables (39 models), and (E) tests of weighted usable area as a habitat model (14 models. Most models were linear or multiple linear regressions, or correlations, but a few were curvilinear functions (exponential or power). Some used multivariate techniques (principal components or factor analysis), and some combined independent variables into one or more indices. We judged model quality based on simple criteria of precision and generality: coefficient of determination, sample size, and degrees of freedom. Most models were based on data sets of fewer than 20 observations and, thus, also had fewer than 20 degrees of freedom. Most models with coefficients of determination of greater than 0.75 had fewer than 20 degrees of freedom, which led us to conclude that relatively precise models often lacked generality. We found that sound statistical procedures were often overlooked or were minimized during development of many models. Frequent problems were too small a sample size, possible bias caused by error in measuring habitat variables, using poor methods for choosing the best model, not testing models, using models based on observational data to predict standing crop, and making unrealistic assumptions about capture probabilities when estimating standing crop. The major biological assumptionthat the fish population was limited by habitat rather than fishing mortality, interspecific competition, or predationusually was not addressed. We found five main ways stream-fish-habitat models are used in fishery management. To be useful for analyzing land management alternatives, models must include variables affected by management and be specific for a homogeneous area of land.