Spatial models of carbon, nitrogen and sulphur stable isotope distributions (isoscapes) across a shelf sea: An INLA approach

Spatial models of variation in the isotopic composition of structural nutrients across habitats (isoscapes) offer information on physical, biogeochemical and anthropogenic processes occurring across space, and provide a tool for retrospective assignment of animals or animal products to their foraging area or geographic origin. The isotopic differences among reference samples used to construct isoscapes may vary spatially and according to non-spatial terms (e.g. sampling date, or among individual or species effects). Partitioning variance between spatially dependent and spatially independent terms is a critical but overlooked aspect of isoscape creation with important consequences for the design of studies collecting reference data for isoscape creation and the accuracy and precision of isoscape models. We introduce the use of integrated nested Laplace approximation (INLA) to construct isoscape models. Integrated nested Laplace approximation provides a computationally efficient framework to construct spatial models of isotopic variability explicitly addressing additional variation introduced by including multiple reference species (or other recognized sources of variance). We present carbon, nitrogen and sulphur isoscape models extending over c. 1 million km ² of the UK shelf seas. Models were built using seven different species of jellyfish as spatial reference data and a suite of environmental correlates. Compared to alternative isoscape prediction methods, INLA-spatial isotope models show high spatial precision and reduced variance. We briefly discuss the likely biogeochemical explanations for the observed spatial isotope distributions. We show for the first time that sulphur isotopes display systematic spatial variation across open marine shelf seas and may therefore be a useful additional tool for marine spatial ecology. The INLA technique provides a promising tool for generating isoscape models and associated uncertainty surfaces where reference data are accompanied by multiple, quantifiable sources of uncertainty.