Design and Analysis of Biodiversity Experiments

Abstract

Biodiversity and ecosystem functioning (BEF) relationships define the ways in which the diversity of species in an ecosystem drive the quantity and quality of the goods and services provided. Species diversity may be defined in various ways, such as richness (species number) and/or evenness (relative proportions of species), and may or may not include species identity. Designing BEF experiments, where species diversity is manipulated across experimental units, can present unique challenges given that species diversity has multiple dimensions. In this thesis, using a simulation study, I test the best practices for designing a BEF experiment, with a focus on capturing the nature of how species interact within BEF relationships. The provision of simultaneous functions by an ecosystem is referred to as multifunctionality. In multifunctionality BEF studies which span across multiple time points, each experimental unit has a multivariate repeated measures response; thus the data that arises from each experimental unit potentially has multiple sources of correlation which must be accounted for in analysis of the data. The development of the Diversity-Interactions (DI) modelling framework builds upon the traditional usage of 'richness-only' models in the BEF literature; richness-only models equate species diversity to solely the number of species, which reduces the true dimensionality of species diversity and therefore may cause loss of information and confounding within an analysis. DI modelling is a regression-based modelling approach to quantifying the BEF relationship which has the ability to separate the effects of individual species and their interactions on ecosystem functioning and is highly generalisable. This approach has previously been extended to model multivariate data. In this thesis, I extend DI models for use with BEF data containing responses with multiple sources of correlation (multivariate repeated measures data). I develop new statistical software tools to fit the new methodologies that are user-friendly and aim to promote uptake of advanced statistical modelling techniques for multifunctionality BEF research. Finally, I develop a new modelling approach for jointly modelling continuous and proportional multivariate response measurements; this work is motivated by challenges identified in the BEF literature but has wider applicability.