Accounting for preferential sampling in species distribution models.

Area of Science:

• Ecology
• Spatial Statistics
• Conservation Biology

Background:

• Species distribution models (SDMs) are crucial for ecological management and conservation.
• Standard geostatistical models assume sampling locations are independent of species presence, an assumption often violated in practice.
• Opportunistic sampling leads to preferential sampling, biasing distribution predictions.

Purpose of the Study:

• To address bias in SDMs caused by preferential sampling.
• To present a computationally efficient alternative to existing Markov Chain Monte Carlo (MCMC) methods.
• To improve the accuracy of species distribution predictions.

Main Methods:

• Interpreting data as a marked point pattern.
• Employing a Bayesian approach for inference and prediction.
• Utilizing Integrated Nested Laplace Approximation (INLA) for efficient model fitting.

Main Results:

• Unaccounted preferential sampling significantly overestimates species abundance at low-density locations.
• The proposed model effectively corrects for bias in both simulated and real-world fishery data.
• Demonstrated the impact of non-randomized sampling on ecological predictions.

Conclusions:

• Ecologists must recognize and account for preferential sampling bias in SDMs.
• The developed INLA-based Bayesian SDM offers a computationally efficient solution.
• Accurate species distribution modeling requires addressing non-systematic sampling strategies.