Testing the accuracy of species distribution models based on community science data

Abstract

While traditional methods of tracking species, collecting specimens, and performing surveys are known to be accurate, additional opportunities to broaden the data pool are evolving. Community science data5 has emerged as a new way of gathering large amounts of data, but little research has been done on its reliability for making models for novel locations. The goal of this project was to test the reliability of eBird data as the primary dataset for ecological niche modeling by determining the accuracy of models derived from the citizen-science based eBird dataset. I made species distribution models of 676 bird species in Costa Rica based on eBird observations to predict which species would be found in two localities in Costa Rica that were surveyed. I compared the predictions with these field surveys to determine the prediction success and Sorensen index of the models. Overall, I found that while spatio-temporal factors can affect the accuracy of ecological models, eBird data have great potential as data for species distribution modeling. The models more accurately predicted the community composition in the rural locality as opposed to the more urban locality, and the accuracy of the models increased when compared with data that covered two month as opposed to one month time periods. I tested to see how the number of observations per species influenced the predictive ability of the models and determined that an intermediate number of observations led to better models. These are important metrics to understand because modeling can be an informative and cost effective way to monitor inaccessible areas and can be used in conservation efforts.