An Evaluation of the Relative Influence of Spatial, Statistical, and Biological Factors on the Accuracy of Stream Fish Species Presence Models

Abstract

AbstractModels relating fish species presence to landscape and stream features are increasingly being used by natural resource managers. The accuracy of these models directly influences the ability to make sound stream management decisions. To evaluate the effect of biotic and abiotic factors on model accuracy, we fit parametric (logistic regression) and nonparametric (k‐nearest neighbor) models of species presence at two spatial scales using subwatershed and stream reach characteristics. We then evaluated the influence of model type, spatial scale, and species‐specific characteristics on species presence omission and commission errors for the best‐fitting scale‐specific parametric and nonparametric models (total of four per species). We found that error rates were higher within species than among species and varied among species by 3.9% (omission) and 8.7% (commission). Within‐species variation in error rates was primarily related to model type and spatial scale, whereas among‐species variation was due to the species‐specific characteristic‐habitat specialization. The relationship between these factors and the omission and commission error rates, however, were relatively complex. The k‐nearest neighbor models were generally more accurate (produced lower errors) at predicting species presence at larger, subwatershed scales, whereas the logistic regression models were more accurate at predicting species presence at smaller, stream reach scales. Similarly, commission error rates were lower for models predicting the presence of habitat generalists at larger scales, whereas commission error rates were lowest for models predicting the presence of habitat specialists at smaller, stream reach scales. The findings from this study suggest that species‐specific characteristics and spatial autocorrelation can influence the accuracy of species presence models and that biologists should consider these effects when modeling the presence or absence of fish species.