Modelling of stream fishes in the Great Plains, USA

 –  Predicting species distributions has important implications for the conservation and management of freshwater fishes, particularly in areas such as the Great Plains, USA where human impacts have resulted in extirpations and declines for numerous native species. There are a number of statistical approaches for constructing distributional models; the accuracy of each is likely dependent on the nature of the environmental gradients, species responses to those gradients and the spatial extent of the modelling. Thus, it is important to compare multiple approaches across species and habitats to identify the most effective modelling approach. Using geographical information system (GIS) derived characteristics of stream segments as predictors, we tested the model performance of three methodologies – linear discriminant function analysis, classification trees and artificial neural networks (ANN) – for predicting the occurrence of 38 fish species in a Great Plains river basin. Results showed that all approaches predicted species occurrences with relatively high success. ANN generally were the best models, in that they generated the most significant models (35 of 38 species) and most accurately predicted species presence for the greatest number of species (average correct classification = 81.1%). The importance of GIS variables for predicting stream fish occurrences varied among species and modelling techniques, but were generally strong predictors of species distributions, including the federally endangered Topeka shiner Notropis topeka . In summary, predictive models should be viewed as both competitive and complementary methodologies for establishing quantitative linkages between fish species and their environment. Our study demonstrates the potential utility of such an approach for guiding conservation efforts for stream fishes of the Great Plains, USA.