Diversity and Distributions Экология, эволюция, поведение и систематика
Assessing Strategies to Optimize Participatory Science Data in Species Distribution Models
Аннотация
ABSTRACT Aims Data from participatory science projects are widely used to understand species distributions. However, participatory science data often contain biases that can obscure real patterns. We address two sources of sampling bias: location uncertainty of individual records and spatially heterogeneous sampling effort. A variety of data processing and modelling strategies have been used to address these biases, without consensus on their relative effectiveness. We test several strategies using data from iNaturalist, a large repository of participatory science data. Location Twenty‐four amphibian ranges of varying sizes and locations within the continental United States. Method We fit species distribution models with iNaturalist data for 24 amphibian species to compare the effectiveness of strategies intended to improve distribution estimates. We examined three categories of strategies: (1) filtering out records with high location uncertainty, (2) thinning to reduce heterogeneous effort, and (3) observation models that account for heterogeneous effort without removing data. We evaluated the predictive accuracy of each approach using validation data from systematic surveys with constant effort and known locations. Results The strategies we tested generally had little effect on predictive accuracy as measured by AUC, with no consistent pattern across species. Instead, we found that the minimal change in performance was influenced by the species‐specific search process, the distribution of records, and the resolution of the analysis. Main Conclusion None of the strategies we tested consistently improve model performance, but our results provide a foundation for a species‐ and analysis‐specific approach. Further improvement in model performance may require additional information or novel methods.