This paper presents a method designed to build species-tailored diatom-environment models. Using a pruning algorithm of artificial neural networks, powerful species-tailored models constrained to water temperature, water depth, and dissolved organic carbon were developed from a 109-lake training set from northwestern Canada and Alaska. The reasoning behind the approach is that the implementation of a single, gradient-based, organism-environment relationship should only use species that are comprehensively influenced by the variable of interest. By pruning species according to their relevance to each of the three studied variables, the cross-validated performances of all three models were significantly increased,...