A fundamental aspect of biodiversity is the distribution of species' abundance across the landscape. Presence-only observations are perhaps the most commonly available type of information about species' distributions, yet they are notoriously difficult to analyze. For example, widely used approaches require manufacturing pseudo-absence data prior to analysis, a statistically indefensible activity in other disciplines, but one necessary to fit models of binary data to presence-only data. Further, these approaches generally estimate relative abundance or probability of occurrence, rather than the more information-rich quantity of absolute abundance. To resolve these and other issues, we have developed a joint point-processes model of species occurrence and observation. The model estimates sampling effort and detection probability for the presence-only observation process by integrating other information sources, such as presence-only observations of other taxa and presence-absence observations of the focal taxon.
Results/Conclusions
Our model clarifies when it is possible to infer absolute species abundance from presence-only observations and eliminates the need for pseudo-absence data. In addition to spatially explicit inference of species abundance, the model also provides spatially explicit quantification of uncertainty, which can guide interpretation of model results, as well as future sampling efforts and management decisions. Our analysis of invasive species in New England demonstrates the effectiveness of our approach with large citizen-science datasets and identifies natural extensions for improving the model's biological usefulness. Adopting our mechanistic approach to modeling species' occurrence data along with species' abundance will make better use of large datasets, improve our understanding of biodiversity, better guide policy and management decisions, and enable detection of responses to environmental change.