New insights in bayesian survival analysis in ecology

Résumé

Wildlife is under siege. And this is not only a fancy sentence to start a thesis, sadly, it is a fact. Over the last century, many wildlife species have seriously declined and many others face their extinction due to rapid and large-scale changes in habitats and ecosystems. Climate change, invasive species, illegal hunting and overfishing are only some of the main threats affecting animal populations nowadays. To address these concerns will require commitment at all levels, from local communities to governments, as well as experts, education and research. Indeed, research is a fundamental tool for conservation. Understanding the factors affecting wildlife populations allow us to improve the management of animal populations and therefore, their conservation. The recently increase in the amount (and variety) of data collected on ecological systems has led to the development of more complex statistical models. This complexity has made the inferential process challenging to perform. The Bayesian approach arises as an alternative to address these issues due to the computational advances occurred in the last decades. Further, prior information (if available) can be easily incorporated as well as this approach takes fuller account of the uncertainties related to models and parameters. In this work we investigate survival, recapture, recovery and migration probabilities in the context of capture-recapture(-recovery) models. These models account for imperfect detection, a common issue in ecological systems. Indeed, if imperfect detection is not taken into account, it may cause biases in estimated demographic parameters of interest. The context of this work is a real one, in particular, a seabird species, the common guillemot (Uria aalge). Seabirds are sentinels of the sea. Their populations reflect conditions over large spatial and long term scales, making them bioindicators of environmental change. Further, the estimation of juvenile survival probabilities in seabirds is far from simple, mainly due to their ecological characteristics. Therefore, this thesis supposes not only a challenge from statistical perspective but also from ecological one. With all this in mind, this thesis is structured as follows. Chapter 1 is devoted to provide the motivation along with an overview of the capture-recapture methods and associated statistical models. After that, we explain in detail the two models used: Cormack-Jolly-Seber (CJS) models and mark-recapture-recovery (MRR) models. Furthermore, we give an introduction of Bayesian inference, before describing the Markov chain Monte Carlo (MCMC) methods. Lastly, we conclude with a brief explanation of the two basic Markov chain simulation algorithms (Metropolis-Hastings and Gibbs sampler), as well as we provide useful software packages and web pages to implement a large variety of capture-recapture(-recovery) models. Chapter 2 takes a brief look at the ecological context of this thesis. To do so, we introduce the species focus of this research, the common guillemot, as well as the study colony, Stora Karlsö (Gotland, Sweden). We briefly provide some of its main ecological characteristics that will be necessary to know in order to fit statistical models biologically motivated. The main part of this Chapter consists in a detailed description of the two data sets that motivated the methodological developments performed in this thesis, a capture-recapture and a mark-recapture- recovery database. Chapters 3, 4 and 5 are dedicated to display the studies performed along this thesis. In particular, Chapter 3 is devoted to provide reliable juvenile survival estimates for common guillemots. The novelty of this work lies in the difficulty on the assessment of juvenile survival due to in this species (and in general, in most seabird species) young birds spend large periods at sea, remaining hence unobservable. However, the study colony has an special feature: a big proportion of immature birds are resighted allowing so to provide those reliable juvenile survival estimates. Further, this work represents a first approximation to the problem of partial monitoring that causes bias in parameter estimates. In particular, we adopt a subjective Bayesian approach so that we incorporate prior information corresponding to the areas where the partial monitoring is affecting. Chapter 4 incorporates a methodology commonly performed in medical survival studies in the context of ecological capture-recapture framework. In particular, we show how differently the capture histories are presented depending on the selected temporal scale. Further, the use of the alternative temporal scale presented (the age) may allow a better interpretation of model parameters when age is the primary interest. Finally, in Chapter 5 we provide an integrated mark-recapture-recovery framework for partially monitored studies. The information gathered by the ring-recovery data allows to correct the bias in survival estimates obtained with only capture-recapture data due to partial monitoring. Moreover, due to the (big) size of the database, we present it in multinomial formulation, so that we provide the explicit efficient likelihood expression along with the associated sufficient statistics of the integrated model proposed. Both the correction of partial monitoring problem (widespread in colonial species) and the construction of the integrated m-arrays along with the efficient likelihood suppose an step forward on this area, either from a practical or methodological perspective. Chapter 6 provides some conclusions and future lines of research, and finally a generic bibliography used along this work is presented.