Bayesian hierarchical space-time models to improve multispecies assessment by combining observations from disparate fish surveys (2012.02196v1)

Abstract: Many wild species affected by human activities require multiple surveys with differing designs to capture behavioural response to wide ranging habitat conditions and map and quantify them. While data from for example intersecting but disparate fish surveys using different gear, are widely available, differences in design and methodology often limit their integration. Novel statistical approaches which can draw on observations from diverse sources could enhance our understanding of multiple species distributions simultaneously and thus provide vital evidence needed to conserve their populations and biodiversity at large. Using a novel Bayesian hierarchical binomial-lognormal hurdle modelling approach within the INLA-SPDE framework, we combined and analysed acoustic and bottom trawl survey data for herring, sprat and northeast Atlantic mackerel in the North Sea. These models were implemented using INLA-SPDE techniques. By accounting for gear-specific efficiencies across surveys in addition to increased spatial coverage, we gained larger statistical power with greatly minimised uncertainties in estimation. Our statistical approach provides a methodological development to improve the evidence base for multispecies assessment and marine ecosystem-based management. And on a broader scale, it could be readily applied where disparate biological surveys and sampling methods intersect, e.g. to provide information on biodiversity patterns using global datasets of species distributions.