Development and application of otolith-based methods to infer demographic connections in a marine metapopulation

Connectivity between local populations is critical if these are to function as a metapopulation and sustain locally open sink populations. Assessing whether such connections between local populations exist is thus an important step towards understanding coastal metapopulation dynamics as well as assessing the efficacy of spatial management tools such as marine reserve networks. For this thesis, I investigate population connectivity of the common triplefin (Forsterygion lapillum) in Cook Strait, New Zealand, using chemical signatures contained within fish otoliths (ear stones). I concentrate on likely connections between three local marine reserves: Kapiti Island (Kapiti coast), Long Island (Marlborough Sounds) and Taputeranga Marine Reserve (Wellington south coast). To this end I develop and implement new statistical methods to enable stronger inferences from otolith chemistry based approaches.

In chapter 2, I evaluate otolith core chemistry as a potential tool (i.e. an environmental fingerprint) for identification of natal source populations of the common triplefin. I sampled otolith chemistry from hatchling fish across a range of hierarchical scales: obtained from individual egg masses within a site; sites within different regions; and regions distributed on the two main islands of New Zealand (North and South Island). This sampling enabled me to construct an “atlas” (or baseline) of otolith core chemistry. I developed and applied a set of novel statistical approaches to examine the characteristics of this natal atlas and optimize its spatial resolution. These analyses allowed me to assess the utility of otolith chemistry as a potential tool to infer patterns of population connectivity in the vicinity of Cook Strait.

Chapter 3 develops a new Bayesian approach to facilitate improved clustering and classification of dispersing fish to putative natal populations based on their otolith chemistry. Otolith-based approaches used to infer natal origins of fishes routinely suffer from the (unrealized) requirement to sample all potential natal source populations. An incomplete baseline atlas has greatly limited the application of otolith chemistry as a tool for assessments of connectivity in the marine environment. In this chapter, I develop, evaluate, and implement statistical solutions to this problem. Specifically, I present a clustering model, based on infinite mixtures, which does not require the specification of a potential number of sources. In a second step, I embed this clustering model in a large-scale classification model that allows for classification on scales encompassing a number of potential sources, where recruits are clustered with observations from the baseline or a separate cluster within these regions. This opens the potential for fish that came from an identifiable source other than those sampled to not be assigned to a sampled source. I evaluate the strength of this approach using the well-known weakfish (Cynoscion regalis) dataset.

In chapter 4, I apply the statistical methods developed in chapter 3 to the common triplefin. I sampled recent recruits of the common triplefin within each of three marine reserves (Kapiti, Long Island, and Taputeranga) and used otolith chemistry to infer probable natal origins. I then compare these inferred patterns of connectivity with those predicted by a set of hydrodynamic simulations. This comparison enabled me to (qualitatively) assess the likelihood of connectivity (as predicted by otolith chemistry) given local hydrodynamic conditions.

For chapter 5, I extend the Bayesian modelling approaches developed in previous chapters to incorporate otolith chemistry data sampled from throughout the life-history of dispersers. As in chapter 3, I develop and evaluate the utility of this approach using a previously published data set (Chinook salmon), and I apply the approach to the common triplefin in a subsequent chapter. Specifically, I propose flexible formulations based on latent state models, and compare these in a series of illustrative simulations and an application to Chinook salmon contingent analysis.

In chapter 6, I apply the Bayesian framework (developed in chapter 5) to the common triplefin data set. Specifically, I formulate a model based on putative chemical distinctions between inshore and offshore water-masses. This model allows me to compare dispersal histories among recruits to a set of reserves (evaluated initially in chapter 4), and the approach reveals patterns that appear to be common to all successful recruits. I examine these findings in the light of results obtained in chapter 4 as well as local hydrodynamic conditions.

Finally, I conclude my thesis in chapter 7 by discussing the relevance of my findings for the functioning of networks of sub-populations, both in a metapopulation and a reserve network context.