Species communities can accurately predict the occurrence of an imperilled fish

• Published in: Canadian journal of fisheries and aquatic sciences 2024-10, Vol.81 (10), p.1358-1368
• Main Authors: Brownscombe, Jacob W., Bzonek, Paul, Drake, D. Andrew R.
• Format: Article
• Language: English
• Subjects: Decision trees; Fish; Geographical distribution; Introduced species; Machine learning; Notropis photogenis; Population studies; Rare species; Sampling; Tributaries
• ISSN: 0706-652X; 1205-7533
• DOI: 10.1139/cjfas-2023-0168

Species distribution information is essential for conservation. However, sampling the full range of a species’ potential distribution is rarely feasible, necessitating the development of models to predict distributions, as well as relevant environmental and biotic drivers. We applied a novel approach to model the distribution of a species at risk in Canada, silver shiner (SS; Notropis photogenis) in tributaries of Lake Ontario using the fish community as a predictor of SS occurrence. Associative rule learning (ARL) identified simple species combinations that provided strong insight into SS distribution, which may be particularly useful for identifying new occupied locations, including making sampling decisions in real time. The species with the most positive or negative associations with SS identified by ARL were included in a random forests model, which predicted SS distribution with high accuracy in test data from the study tributary system and in a neighbouring system where SS is exceedingly rare. Predicting species distributions based on biotic associations presents opportunities for discovering new populations, identifying critical habitat, and evaluating the suitability of sites for re-introduction potential.