Proces-gebaseerde en statistische benaderingen van soortengemeenschapssamenstelling doorheen tijd en ruimte

Understanding how species assemble into communities across space and time is a central aim of ecology. Over the past decades, metacommunity ecology has primarily focused on niche-based and dispersal-related mechanisms. Quantitative attempts to link empirical patterns to these processes typically yield remarkably low explanatory power. This observation has led to the suggestion that community assembly might be largely driven by mere stochasticity. In this dissertation, I explore two alternative mechanisms that might underlie the low explanatory power that currently dominates community ecology. First I investigate how historical contingencies, and priority effects in particular, can influence the assembly process and constrain predictability. Priority effects occur when the order and timing of arrival of individuals and species impact the subsequent assembly process, and can be reinforced by contemporary evolution. Second, I develop novel Bayesian hierarchical modelling approaches to model metacommunity data. Through improved modelling of the data generating process, spanning the sampling, observation and biological processes, I attempt to improve explanatory power and to broaden the type of input data for metacommunity analyses. In Chapter 1, I use a spatially explicit individual-based metacommunity model characterised by serial propagule rains to quantify the impact of historical contingencies in community assembly. By tracking ancestral lineages through time, I show that the majority of individuals in a landscape descend from the first propagule rains, across a wide range of ecological and evolutionary conditions. These findings suggest that historical contingencies under the form of ecological and evolution-mediated priority effects may be common in nature, across multiple spatiotemporal scales. They also provide a potential explanation for the low explanatory power in metacommunity analyses. Further model explorations are, however, needed to verify the effects of alternative, more realistic model assumptions. Chapter 2 expands on this work by acknowledging that dispersal ability as a trait can evolve, like any other trait. I developed an individual-based mainland-island model to explore how the evolution of dispersal ability impacts the prevalence of evolution-mediated priority effects across multiple scenarios. Low species richness and low immigration rates from the mainland promoted the evolution of the highest dispersal abilities. Evolution-mediated priority effects were strongest when species pools were small. I also find that assuming a fixed dispersal ability that cannot evolve tends to lead to either under- or overestimating the importance of evolution-mediated priority effects. Overall, this chapter demonstrates that the evolution of dispersal abilities can shape evolution-mediated priority effects when provided with greater ecological opportunity in species-poor communities, and underscores the impact of considering the evolution of dispersal ability. Chapters 3-5 shift focus from process-based models to statistical approaches for analysing empirical biodiversity data. The overarching goal of these chapters is to reduce apparent patterns of stochasticity in observational metacommunity data through improved modelling of environmental relationships and by better acknowledging the sampling and observation process shaping such data. In Chapter 3, I introduce a comprehensive Bayesian site-occupancy model tailored to large, opportunistically collected citizen-science datasets, which are typically characterised by an imperfect and strongly heterogeneous detection process. By explicitly modelling imperfect detection and phenological as well as observer-specific variation thereon, I show that this modelling approach is able to retrospectively infer biodiversity dynamics through space and time. As such, this model enables fine-grained monitoring of species for which standardised monitoring is unavailable or unsuitable. I showcase the model through its application to over three million butterfly records in Belgium, collected through the online biodiversity portal Waarnemingen.be/Observations.be. In Chapter 4, I extend the framework outlined in Chapter 3 to the multi-species setting. Specifically, I fuse a comprehensive site-occupancy model with an extended version of the hierarchical modelling of species communities (HMSC) framework, a leading joint species distribution model. Through the efficient exchange of information across species, observers, space and time, this approach enables fine-grained, species-specific inference on a variety of biological parameters, revealing environmental relationships, phylogenetic and trait-based influences thereon, interannual trends, phenological shifts, and co-distributional patterns through space and time. I demonstrate the use of the model by applying it to odonate data derived from the online biodiversity portal Waarnemingen.be/Observations.be, containing over 100,000 pond visits conducted by hundreds of naturalists across Flanders. This chapter demonstrates that opportunistically collected data, typically regarded as too noisy for metacommunity analysis, can yield insights when analysed with methods that acknowledge the data generating process, unlocking the largest source of biodiversity information for metacommunity research. Chapter 5 addresses a different but pervasive challenge in ecological community data: temporal aggregation of count data. Many ecological survey techniques, such as trapping and migration counts, aggregate observations over heterogeneous time-intervals, masking underlying temporal dynamics. I developed a temporal disaggregation approach based on interval-integrated B-splines, enabling inference on phenology and long-term spatiotemporal trends from aggregated data. Applying this model to nearly fifty years of pitfall-trapping data in the Belgian province of Limburg, I reconstruct fine-grained activity patterns over time, showing that the phenology of ground-dwelling spider species advanced by up to 35 days over the past 50 years. The model also yields a wide variety of other quantities of interest, including spatial and interannual trends. Together, these five chapters illustrate how process uncertainty owing to biological complexity, such as historical contingencies caused by evolution-mediated priority effects, as well as data uncertainty, owing to statistical considerations related to the sampling and observation processes, can lead to low apparent levels of explanatory power in metacommunity analysis. In addition to addressing challenges underlying observational biodiversity data, the statistical approaches introduced in this dissertation also provide means for an improved modelling of biological processes, as demonstrated through a number of case studies. The data enrichment paradigm, which integrates augmented and incomplete data modelling by combining observed information with latent structures, serves as a unifying theme to jointly address these considerations when analysing metacommunity data. The present work provides several building blocks towards a more comprehensive understanding of biodiversity patterns, ranging from the sampling and observation process to the many complexities shaping the biological process.