Fish and habitat data were collected from 84 wadeable stream reaches in the Mississippi River drainage of Iowa to predict the occurrences of seven fish species of greatest conservation need and to identify the relative importance of habitat variables measured at small (e.g., depth, velocity, and substrate) and large (e.g., stream order, elevation, and gradient) scales in terms of their influence on species occurrences. Multiple logistic regression analysis was used to predict fish species occurrences, starting with all possible combinations of variables (5 large-scale variables, 13 small-scale variables, and all 18 variables) but limiting the final models to a maximum of five variables. Akaike's information criterion was used to rank candidate models, weight model parameters, and calculate model-averaged predictions. On average, the correct classification rate (CCR = 80%) and Cohen's kappa (κ = 0.59) were greatest for multiple-scale models (i.e., those including both largescale and small-scale variables), intermediate for small-scale models (CCR = 75%; κ = 0.49), and lowest for large-scale models (CCR = 73%; κ = 0.44). The occurrence of each species was associated with a unique combination of large-scale and small-scale variables. Our results support the necessity of understanding factors that constrain the distribution of fishes across spatial scales to ensure that management decisions and actions occur at the appropriate scale.
RightsWorks produced by employees of the U.S. Government as part of their official duties are not copyrighted within the U.S. The content of this document is not copyrighted.
AbstractFish and habitat data were collected from 84 wadeable stream reaches in the Mississippi River drainage of Iowa to predict the occurrences of seven fish species of greatest conservation need and to identify the relative importance of habitat variables measured at small (e.g., depth, velocity, and substrate) and large (e.g., stream order, elevation, and gradient) scales in terms of their influence on species occurrences. Multiple logistic regression analysis was used to predict fish species occurrences, starting with all possible combinations of variables (5 large-scale variables, 13 small-scale variables, and all 18 variables) but limiting the final models to a maximum of five variables. Akaike's information criterion was used to rank candidate models, weight model parameters, and calculate model-averaged predictions. On average, the correct classification rate (CCR = 80%) and Cohen's kappa (κ = 0.59) were greatest for multiple-scale models (i.e., those including both large-scale and small-scale variables), intermediate for small-scale models (CCR = 75%; κ = 0.49), and lowest for large-scale models (CCR = 73%; κ = 0.44). The occurrence of each species was associated with a unique combination of large-scale and small-scale variables. Our results support the necessity of understanding factors that constrain the distribution of fishes across spatial scales to ensure that management decisions and actio...