Thesis

Giovanni Poggiato. Integrating ecological dependencies in biodiversity modelling

Abstract: Understanding the ecological processes driving the distribution of life on Earth has always been a central goal of Ecology.
Nowadays, this knowledge is also crucial to project how biodiversity from various ecosystems will respond to global changes in order to propose adaptation and mitigation measures to safeguard biodiversity and associated ecosystem functions.
Statistical ecology has arisen as a discipline that moves away from describing biodiversity patterns towards attempting to model the output of the ecological processes that generate these patterns. However, many statistical models apply to ecology, assumed independence between the modelled entities. For example, Species Distribution Models (SDMs) predict the distribution of each species in an ecosystem as a function of some environmental covariates, but independently of other species. However, we know that species interact, and how these biotic interactions shape biodiversity patterns even at large scale. Trait distribution models (TDMs) suffer from the same lacks in the context of functional traits, that are strongly characterised by trade-offs and synergies. These are just two ecological dependencies that drive biodiversity distribution and ultimately ecosystem functioning. The aim of the thesis is therefore to integrate these ecological dependencies in biodiversity modelling.

Detail:

Clément Vallé. Adapting and developing new biodiversity indicators from improved Joint Species Distribution Models (JSDM)

Abstract: Abrupt climate change, land-use change and other human-made disturbances are triggering species extinctions and range shifts, and are altering ecosystem functioning. Effective conservation strategies to mitigate these threats should build on a good understanding of how global change affects biodiversity distribution and ecosystem functions. Their definition and effectiveness assessment must rely on relevant indicators of biodiversity. However, despite the importance of interactions between species in structuring communities and characteristic trait distributions, current indicators do not take them explicitly into account.
Recently, tremendous progress has been made in extending correlative species distribution models (SDMs) to Joint Species Distribution Models (JSDMs) to account for and estimate species interdependencies. As in SDMs, JSDMs can predict species (or traits) distributions based on environmental and spatial variables, but they can also discriminate their shared responses to these variables from additional ecological or evolutionary processes (e.g. biotic interactions). However, the practicality and efficiency of these recent methodological developments for conservation have not been properly tested and applied yet.
This thesis aims to investigate the use of JSDMs to adapt and develop biodiversity indicators able to account for these interactions in biodiversity states and trends assessments.

Detail:

Julien Gibaud. Extending the supervised component-based generalized linear regression to joint response modelling for species-rich ecosystems

Abstract: In this thesis, we propose to extend the SCGLR methodology, enabling it to identify clusters of responses sharing explanatory components. Originally, SCGLR was designed to find explanatory components in a large set of possibly highly redundant covariates, something much needed in a high-dimensional framework. These components are jointly supervised by all the responses. Henceforth, we aim at identifying clusters of responses sharing the same explanatory dimensions. In an ecological framework for instance, communities of species should be modeled by components which are characteristic of each community. An algorithm is developed in order to estimate the model.
This thesis aims to investigate the use of JSDMs to adapt and develop biodiversity indicators able to account for these interactions in biodiversity states and trends assessments.

Detail:

Samuel Valiquette. Multivariate regression models for zero-inflated and over-dispersed count data.

Abstract: The main objective of this PhD thesis is to provide a new class of distribution functions for count data with enough flexibility to fit zero-inflated and over-dispersed data in the uni- and multi-variate context, and encompassing classical ones such as Poisson, Negative Binomial, Poisson-log-Normal, Negative Binomial-log-Normal (Zhou et al. 2012) or Poisson-inverse Gaussian distributions (Dean et al. 1989).

Detail:

Souléman Traoré. Title

GAMBAS