Simple models to quickly estimate the probable range of data‐limited riverine fishes

Abstract

AbstractSpecies distribution models provide biologists with insight into species ranges, ecological niches, and spatial predictions of suitable habitat, but require significant training, data, and time for biologists to obtain reliable results. Alternatively, ad hoc inferences from sparse occurrence records leave biologists to face conservation decisions with high uncertainty. Simple predictive models can fill the gap between these two alternatives. One method to predict the probable range of riverine fishes is to add the average distance between occurrences to the most distal occurrence (parametric estimator). However, there are drawbacks to this method when the distances between occurrences deviate from a non‐uniform distribution. Herein, estimators that are robust to non‐uniformity, a non‐parametric and an optimal linear estimator, were adapted and compared to the parametric estimator using simulations. The non‐parametric estimator showed the least bias and variance in bias for the probable range endpoint. The optimal linear estimator best balanced type I and type II errors in the maximum probable range estimate (i.e., upper CI). Application of the robust estimators were made to demonstrate how sampling can be planned based on available effort (Gulf Coast walleye, Sander vitreus) and where early detection sampling may be focused (silver carp, Hypophthalmichthys molitrix). This study provides biologists with a suite of simple options (R code provided as supplementary material) to bridge the gap between ad hoc inference and complex modeling over a range of scenarios. Future studies should focus on accounting for false‐positives, variable effort, and develop online apps for accessibility.