Work

Accounting for physical barriers in species distribution modeling with non-stationary spatial random effects

Year: 2016
Type: preprint
Abstract: This paper is motivated by a study on the spatial distribution of fish larvae in reproduction habitats of three commercially important fish species in the Finnish archipelago. Species distribution modeling (SDM) requires realistic models that account for spatial dependence in observation data that cannot be explained by environmental factors alone. Gaussian latent variable models are increasingly used in this context, where a stationary Gaussian random field accounts, for instance, for the spatial correlation driven by unmeasured covariates. In this paper we challenge the assumption of stationarity in cases where there are physical barriers in the study area, such as when our interest is in fish populations near the coast. In these situations, the common assumption of stationary second order spatial behaviour is likely to be unrealistic, as spatial dependencies should be considered around the barriers rather than across them. We develop a non-stationary Gaussian field that treats, e.g., land as a barrier to spatial correlation. We define the field as a continuous solution to a differential equation with spatially varying coefficients, and achieve a computational cost similar to that of the stationary model. We demonstrate the benefits of the new model for a fish larvae application in the Finnish archipelago.
Language: English
Open Access status: green
OpenAlex ID: W2509455001
Cites:
Cited by: 14
Related to: 10