Sample size and spatial configuration of volunteered geographic information affect effectiveness of spatial bias mitigation

Volunteered geographic information (VGI) can provide field samples for predictively mapping geographic phenomena. Yet the biased spatial coverage of VGI observations often undermines the fitness of use of VGI samples for predictive mapping. Although methods have been developed to mitigate spatial bias in VGI samples to improve predictive model performance, there exist limited investigations into the impacts of VGI sample size and spatial distribution characteristics on the effectiveness of the methods. This article presents an empirical evaluation on how the two factors affect the effectiveness of bias mitigation methods with a case study of mapping habitat suitability of the red-tailed hawk (Buteo jamaicensis) using eBird data. Results reveal positive correlations between model performance improvement and sample size, given samples of similar spatial configuration. VGI samples with more spread-out spatial coverage (i.e., more representative) are more amenable to bias mitigation. However, performance improvement plateaued beyond a certain sample size and sample representativeness thresholds.