Predicting the occurrence of rare species is challenging, especially outside of their known ranges. However, this information is critical for assessing conservation status, guiding monitoring efforts, and directing conservation actions for species threatened with extinction. Furthermore, frequent updates are needed under changing climate and landscape and thus, shifting species' distribution. Data collected by community science projects may provide additional information about species occurrence, complementing standard surveys. We used data from a community science project, eBird, and species-specific environmental variables with Bayesian Networks (BNs) to predict the occurrence of 12 federally-protected birds in Canada outside archived known ranges. Our results indicate the number of presence records in the eBird database in the previous three years is a useful variable to include in models predicting species occurrence outside known ranges for 7 out of 12 species. Land cover variables were important predictors, but the distance between a given site and the known range of species was not. While low number and proportion of detections in input datasets were negatively correlated with model performance, our BNs correctly predicted more than half of the presences in independent datasets for 6 species. For these 6 species, BNs predicted additional 0.07 to 68.22% (mean 20.93%) area of occurrence beyond known ranges. This study suggests that community science data may help to fill some information gaps in standard surveys and could be used to update species occurrence information for rare species, particularly in combination with models (e.g., BNs) capable of making and updating predictions with limited data.
Read full abstract