Species distribution modelling with spatio-temporal nearest neighbour Gaussian processes

TL;DR

This paper introduces a flexible, computationally efficient spatio-temporal Gaussian process model for ecological data analysis, implemented in the staRVe package, enabling easier and more insightful species distribution studies.

Contribution

It develops a generalized linear mixed model with a nearest neighbour Gaussian process, facilitating analysis of large ecological datasets without advanced coding skills.

Findings

Model provides accurate predictions and forecasts.

Tutorial demonstrates effective workflow for Carolina wren data.

Showcases analysis of haddock survey data.

Abstract

1.) Spatio-temporal datasets that are difficult to analyze are common in ecological surveys. There are software packages available to analyze these datasets, but many of them require advanced coding skills. There is a growing need for easy to use packages that researchers can use to analyze common ecological datasets 2.) We develop a particular generalized linear mixed model for spatio-temporal point-referenced data that is flexible enough to accommodate data from most ecological surveys while being structured enough to facilitate analyses without advanced coding. Our implementation in the staRVe package uses a computationally efficient version of a nearest neighbour Gaussian process enabling analysis of relatively large datasets. 3.) A brief simulation study shows our model produces accurate predictions and forecasts, while a tutorial analysis of a Carolina wren survey suggests a recommended workflow for analyses. We also analyze a more complicated scientific survey of haddock to showcase the capabilities of our model. 4.) Our model and package are tools that can easily be added to researchers' workflow to help make sense of data from ecological surveys. We emphasize the ability of our model to create useful visualisations of data which can then lead to identification of important trends in species distributions.