Efficient spatial models for predicting the occurrence of subarctic estuarine-associated fishes: implications for management

Abstract

In many of the nearshore areas where development is most likely to occur, essential fish habitat data are incomplete and there is little information on species occurrence that can be used to inform management decisions. This research investigated the use of multivariate remotely sensed geomorphic and landscape data to develop accurate predictive models of subarctic, estuarine-associated fishes. The random forest algorithm was used to predict the occurrence of 26 fish species captured in 49 estuaries in Southeast Alaska. Model prediction accuracy ranged from 100 to 42% for species presence and 87 to 15% for species absence. Model goodness of fit and accuracy were assessed by comparing the number of species occurrences predicted by the model against the observed presences and absences of species in an independent data set. Sixty percent of the models were able to predict species presence with an accuracy of 70% or better. The models were used to predict species occurrence for 521 unsampled Southeast Alaskan estuaries to provide a regional map of predicted species distributions.