Predicting species occurrence is central to ecology and numerous modeling approaches have been applied, including deductive and inductive. We conducted two modeling projects, one in Northwest U.S. and the other in Alaska, to evaluate the potential of improving species distribution models by combining deductive and inductive modeling approaches for over 1,000 taxa. Our objectives were to map each species' range, produce an inductive distribution model based on species occurrence records using Maxent and a deductive distribution model using expert knowledge of species-habitat associations, combine their output, and compare their relative strengths and limitations. While the two projects were independent, but related, in general, our methods included compiling species occurrence records, environmental variables, habitat associations, and assessing model accuracy. We had 342 deductive and 213 combined models in Northwest U.S. and 222 deductive, 74 inductive, and 37 combined models in Alaska selected as final models. Despite extensive efforts to filter occurrence data for our inductive models, our deductive models were often selected as the final models and were comparable to combined models. How well a deductive, inductive, or combined model represented a species distribution depended on the quality and quantity of input data. We found that collecting high-quality occurrence data, assembling accurate environmental data, obtaining rigorous expert input, and addressing sampling bias were vital. Our modeling efforts have improved the existing data, modeling, and understanding of many taxa. All species ranges and distribution models are available online.
