Bayesian hierarchical methods for species distribution modelling under imperfect detection

Monitoring the distribution of wildlife populations has become essential for the understanding of how species are affected by environmental changes and to provide adequate management plans and effective strategies for the conservation of biodiversity. The growing concern about biodiversity loss has led to a rapid development of sampling methods and data collection schemes that enables data of the distributions for multiple species to be obtained at different temporal and spatial scales. Nowadays, biodiversity conservation involves monitoring programs that target multiple species within a community where individual species responses vary widely. This high variability makes the task of identifying the ecological processes that drive species distributions challenging and complex. This complexity has led to the development of a wide range of species distribution models that allow the identification of the most important areas for biodiversity conservation. However, describing such processes is no easy task due to the sources of uncertainty that occur at different spatial and temporal scales and that are induced by imperfect detectability. Thus, modern methods in statistics are increasingly being used to analyse the distribution and abundance of wildlife populations while accounting for the multiple sources of error associated with both, the ecological process of interest and the data collection process. The present work extends some of the well-established species distribution modelling techniques that address imperfect detection and propose new methods to describe different attributes of biological communities (e.g. species rarity) and their relationship with the environment. Computer simulations were used to assess models performance. Then, the proposed methods were applied to a data set of Odonata occurrence records in water bodies across the UK that were partially observed due to imperfect detection. The data for this research were provided by the Hydroscape project (www.hydroscapeblog.wordpress.com), a project that aim to determine how different connectivity metrics interact with environmental stressors to affect species diversity in UK freshwaters. Chapter 1 gives a background of the ecological concepts and statistical principles that are commonly used in species distribution modelling, introduces the questions of interest and the aims of this research. It also shows an overview of the data and presents an exploratory analysis that will be necessary to take into account for the analysis in subsequent chapters. Chapter 2 reviews some of the existing models that have been developed to investigate species distributions under imperfect detection and apply such methods to the Odonata case study, discusses the importance of accounting for imperfect species detection through a simulation study, and compares different software and approaches that have been developed to fit such models. Chapter 3 proposes a new method to quantify species rarity in a community when species are detected imperfectly. Then, a two-step modeling framework is proposed as an approach that enables for a complex hierarchical model to be analyzed in different stages to provide a pragmatic computationally efficient method for choosing the most relevant predictors affecting an ecological response of interest while propagating the uncertainty associated with the estimation of this quantity on a second analysis. In chapter 4, a new method that accounts for non-linear relationships between species distribution and environmental conditions is developed. A simulation study is presented to assess the model performance under different scenarios and different methodological consideration practices are discussed. Chapter 5 provides an application of the model developed in chapter 4 to investigate Odonata species distribution temporal patterns and discusses how these results can be used for biodiversity conservation and management. Finally, chapter 6 summarizes the main outcomes of this research and discusses the methodological innovations and challenges of the proposed methods with a final discussion on possible future work