Continuous-space occupancy models

Area of Science:

• Ecology
• Spatial Statistics
• Computational Biology

Background:

• Occupancy models are crucial for inferring species distributions, but existing methods face limitations in modeling continuous spatial domains with discrete observed data.
• Current approaches struggle to account for species presence within only a fraction of a surveyed site.

Purpose of the Study:

• Develop a new class of spatial occupancy models capable of handling a change of spatial support between observed data and the underlying occurrence process.
• Enable more realistic modeling of species occurrence in continuous space at finer resolutions than observed data.
• Relate detection probabilities to within-site occurrence proportions.

Main Methods:

• Introduce a clipped Gaussian process to represent species occurrence in continuous space.
• Employ Bayesian methods for model fitting, including a computationally efficient Markov chain Monte Carlo (MCMC) algorithm.
• Utilize a Vecchia approximation for the spatial Gaussian process and a surrogate data approach for joint updates of spatial terms and covariance parameters.

Main Results:

• The developed model successfully embeds a change of spatial support, allowing for finer-resolution inferences.
• Demonstrated the model's efficacy using simulated data and comparison with alternative spatial occupancy models.
• Successfully analyzed ovenbird occurrence data from New Hampshire, USA.

Conclusions:

• The new spatial occupancy models provide a more realistic and flexible framework for species distribution modeling.
• The approach enhances the ability to model species occurrence across continuous spatial domains and account for imperfect detection within sites.
• This method offers significant advancements for ecological research and conservation efforts requiring fine-scale distribution mapping.