Temporal trends in the spatial bias of species occurrence records

Abstract

Large-scale biodiversity databases have great potential for quantifying long-term trends of species, but they also bring many methodological challenges. Spatial bias of species occurrence records is well recognized. Yet, the dynamic nature of this spatial biashow spatial bias has changed over time -has been largely overlooked. We examined the spatial bias of species occurrence records within multiple biodiversity databases in Germany and tested whether spatial bias in relation to land cover or land use (urban and protected areas) has changed over time. We focused our analyses on urban and protected areas as these represent two well-known correlates of sampling bias in biodiversity datasets. We found that the proportion of annual records from urban areas has increased over time while the proportion of annual records within protected areas has not consistently changed. Using simulations, we examined the implications of this changing sampling bias for estimation of long-term trends of species' distributions. When assessing biodiversity change, our findings suggest that the effects of spatial bias depend on how it affects sampling of the underlying land-use change drivers affecting species. Oversampling of regions undergoing the greatest degree of change, for instance near human settlements, might lead to overestimation of the trends of specialist species. For robust estimation of the long-term trends in species' distributions, analyses using species occurrence records may need to consider not only spatial bias, but also changes in the spatial bias through time.

Identifying the Most Probable Mammal Reservoir Hosts for Monkeypox Virus Based on Ecological Niche Comparisons

“…To reconstruct the ecological niche of mammal species, we obtained species occurrence data from multiple databases, in an effort to reduce sampling biases [ 61 , 62 , 63 ]: the Global Biodiversity Information Facility (GBIF; , accessed on 22 June 2022), Integrated Digitized Biocollections (IdIgBio), VertNet, and INaturalist, using the R package spocc [ 64 ]. All of these databases carry occurrence records for mammal species, but all have their specificity: GBIF is the largest database of species occurrence records; IdIgBio is the United States digitised collection of specimens [ 65 ]; VertNet is a publicly accessible database on vertebrate specimen records from natural history collections around the world [ 66 ]; finally, iNaturalist is a citizen-science database with research-grade observations of multiple taxa.…”

Rodent trapping studies as an overlooked information source for understanding endemic and novel zoonotic spillover

“…Despite the importance of understanding these complex systems, current evidence on host-pathogen associations is considerably affected by taxonomic and geographical sampling biases [ 11 , 17 ]. Curated biodiversity datasets such as the Global Biodiversity Information Facility (GBIF) and resources produced by the International Union for Conservation of Nature (IUCN) suffer from well described spatial and temporal sampling biases [ 18 , 19 ]. These data are typically obtained from museum specimen collections and non-governmental organisation surveys.…”

A Double machine learning trend model for citizen science data

“…Studies of citizen science data have also shown how interannual changes in the observation process bias trend estimates. Bowler et al (2022) showed how changes in the spatial site selection produce species-specific biases in trend estimates. Zhang et al (2021) showed how bias can even arise despite interannual survey structure, documenting how unexpected changes in survey censoring due to urbanization biased trends in species richness.…”