• intermittent streams;
  • north-western Great Plains, U.S.A.;
  • prairie ecosystems;
  • tree classifiers

The objectives of this study were (1) to determine whether the presence or absence of prairie fishes can be modelled using habitat and biotic characteristics measured at the reach and catchment scales and (2) to identify which scale (i.e. reach, catchment or a combination of variables measured at both scales) best explains the presence or absence of fishes. Reach and catchment information from 120 sites sampled from 1999 to 2004 were incorporated into tree classifiers for 20 prairie fish species, and multiple criteria were used to evaluate models. Fewer than six models were considered significant when modelling individual fish occurrences at the reach, catchment or combined scale, and only one species was successfully modelled at all three scales. The scarcity of significant models is probably related to the rigorous criteria by which these models were evaluated as well as the prevalence of tolerant, generalist fishes in these stochastic and intermittent streams. No significant differences in the amount of reduced deviance, mean misclassification error rates (MER), and mean improvement in MER metrics was detected among the three scales. Results from this study underscore the importance of continued habitat assessment at smaller scales to further understand prairie-fish occurrences as well as further evaluations of modelling methods to examine habitat relationships for tolerant, ubiquitous species. Incorporation of such suggestions in the future may help provide more accurate models that will allow for better management and conservation of prairie-fish species.