Spatial and spatio-temporal models for use in the marine environment with applications to the Scotian Shelf

Abstract

The marine environment is a particularly challenging place for anyone interested in the animals which inhabit it. Unlike on land, where animals go and what they do is usually unobservable. Efforts to learn often rely upon a tag, a device attached to an individual animal that records or transmits information about the animal, where it goes, and perhaps what it does (often via some ancillary information that is also recorded). Alternatively, locations may be sampled and animals captured, and counted, at these locations, so as to learn about numbers and distribution. Such studies are usually expensive, and the number of tagged or captured animals small, such that a large gap exists between knowing where that small sample of animals went or was found, and knowing where animals of that population go, and what they do, in their habitat. One means of bridging this gap is to develop models that use the small set of discrete locations generated by the tagged or captured animals to model, or predict, at all locations in the ecosystem, the number, or the behaviour, expected at those unobserved locations. This thesis explores the use of Bayesian hierarchical spatio-temporal random field models in the marine environment, using several different forms of available marine data. Models discussed include: methods appropriate to novel data forms being produced by deployed acoustic tags; applications in population distribution and stock structure modelling based upon research trawl data; and integrated models, which combine several different data types (physical, environmental, biological and acoustic tracking), into a single modelling framework needed to examine interdependencies between species, habitat wide. The results of these developments clearly demonstrate that random field models are a useful and practical modelling approach. They are reasonably easy to fit, are able to capture spatio-temporal trends when present, have a parameter set that is interpretable in ways that are biologically meaningful, and, provide new means of looking at interspecies relationships. The ready interpretability of the parameters also lends them to direct practical use when applied to populations or species under commercial exploitation and/or protective management. This same easy interpretability of parameters, and the flexibility of the modelling format, allow simultaneous modelling of predator and prey to view interspecies spatial relationships otherwise hidden beneath the sea.