Fast Bayesian inference for large occupancy datasets.

Area of Science:

• Ecology and Conservation Biology
• Computational Statistics
• Biodiversity Informatics

Background:

• Citizen science initiatives provide extensive, long-term species occurrence data, crucial for ecological research.
• Opportunistic data, though vast, present computational challenges for traditional occupancy models, especially within Bayesian frameworks.
• Existing methods struggle to effectively incorporate spatial and temporal dependencies common in ecological datasets.

Purpose of the Study:

• To develop a unified Bayesian framework for occupancy models handling spatial and temporal autocorrelation.
• To enable efficient analysis of large-scale citizen science datasets for biodiversity change assessment.
• To provide robust tools for measuring species' occurrence variations over extended periods.

Main Methods:

• Developed a Bayesian inference framework for occupancy models incorporating spatial and temporal autocorrelation.
• Utilized the Pólya-Gamma scheme for accelerated Bayesian inference.
• Integrated spatio-temporal random effects using Gaussian processes with subset of regressors and nearest neighbor approximations.

Main Results:

• Successfully applied the framework to 45 years of UK butterfly occurrence data from the Butterflies for the New Millennium database.
• Generated reliable occupancy indices for both common and rare butterfly species.
• Demonstrated the computational efficiency and applicability of the proposed Bayesian approach.

Conclusions:

• The developed framework offers a computationally efficient solution for analyzing large-scale, spatio-temporally autocorrelated species occurrence data.
• This approach enhances the utility of citizen science data for monitoring biodiversity trends and informing conservation strategies.
• The model is adaptable for diverse taxa, facilitating broader applications in ecological research and biodiversity assessment.