Search

Species distribution models are key to spatially explicit conservation and management plans. Modeling distributions using data collected by the public has become increasingly popular; however, it is vital to evaluate their advantages and limitations. We investigated whether distribution models developed using opportunistic data (public reports) are congruent with those developed using professional data, while evaluating two strategies to remove spatial biases: data filtering and covariate removal. We compared GPS telemetry data (598 locations, 2010–2018) with public reports (1384 locations, 2016–2019) of American black bears (Ursus americanus) in Missouri (USA). Reports were subdivided into unfiltered reports, filtered reports (more reliable), and verified reports (confirmed by state agency staff). Each data set was modeled twice, with and without road covariates. Each distribution model was an ensemble of three machine‐learning methods. Models using filtered or verified reports and excluding road covariates were most similar to the telemetry model, both in spatial extent and spatial suitability patterns. However, the report‐derived distributions included several times more agriculture and developed land. The inclusion of roads resulted in spurious patterns and consistently reduced the predicted distribution area by ~10%. Elevation and vegetation productivity were consistently selected as the most important variables, regardless of data source and covariates included in models, but the report‐derived distributions included more agriculture and developed lands. Filtering data for reliability and removing covariates related to spatial biases resulted in models suitable to estimate broad black bear distribution, though differences remained that could represent different segments of the bear population (e.g., dispersing vs. established). Verification was not essential to develop suitable models if reports are filtered for reliability, important for areas that lack professionally collected data to model species distributions. [ABSTRACT FROM AUTHOR]