Can a Random-Fields Approach for Estimating Abundance Indices be Improved upon with Auxiliary Habitat Information?

The parameterization of ecological and population models with random effects (e.g., hierarchical modeling; state-space time series modeling; spatial modeling) has been conducive for advancing the  bounds of statistical inference, distinguishing sources of error variation, and establishing a framework for continued emerging research.  One area of development is the equivalent of a multivariate extension of random effects to account for unobserved random stochastic processes across continuous space, such as those arising from the general, collective influence of environmental and biological variation/covariation across the landscape.  In this context, recent advances in the application of Gaussian Markov random fields for estimating parameters in geostatistical models have shown promise for improving survey abundance indices used in stock assessment.  Here, we use simulations to explore whether information on explicit habitat-related covariates improves the performance of the random-fields approach to more accurately/precisely depict species distribution and density.  We then compare resulting trawl survey abundance indices estimated from the geostatistical delta general linear mixed model with and without auxiliary habitat information for several west coast groundfish species.