Ecosystems are composed of many species interacting with each other and with their abiotic environment. These complex interdependencies are crucial for understanding ecosystems’ ability to respond to perturbations. Networks are powerful tools to describe such systems of species linked by ecological interactions and study how their architecture relates to ecosystem functioning and response to perturbation [e.g. 1].
In particular, plant-pollinator interactions are key for the reproduction of most wild plants and the production most crops in agroecosystems. In the past decades, many plant-pollinator communities have been sampled allowing to identify the architectural characteristics of these pollination networks. This involves computation of metrics describing the network structures such as the modularity or the nestedness  and the coclustering of plants and pollinators by using Latent Block Models  which are latent variable model accounting for heterogeneity of connections in a network. At the same time, the development of theoretical approaches highlighted links between these architectural characteristics and their dynamical stability or robustness to species extinction. However, the sampling of such networks is labor-intensive and many datasets only reveal a subset of the existing interactions. Obviously, the sampling process can induce enormous biases in the statistical analyses of the networks which have not being taken into account in most papers concerned with the analysis of plant-pollinator network structure. This raises doubts regarding the current understanding of the structure of pollination networks and their predicted response to perturbation.
The aim of this internship is to develop statistical tools to tackle sampling effects in these analyses. This means accounting for different species detectability and abundances in the inference of statistical model (Latent Block Models) and in the computation of the metrics (modularity and nestedness). The observed networks will be then considered as incomplete observations of « true » plant-pollinator networks. Statistical models for these observation processes should be proposed and the inference of latent block models should incorporate this additional layer of uncertainty on the data. Moreover, the computation of the metrics will come with an uncertainty assessment derived from observation processes.
These tools will be used to re-analyze a set of about 100 plant-pollinator networks and to investigate how accounting for the sampling effect provides a new insight in the structure of these networks. More specifically, studies often show that plant-pollinator networks are nested but this structure simply arises from differences in species abundance and sampling effort. This hypothesis could then be tested with this new tool.
Moreover, these methods could be used on data from a citizen science program (Spipoll ) aiming at monitoring plant-pollinator interactions in France, with more than 300,000 records across seasons.
M2 in Ecology or Applied Statistics. Interest in statistical modeling, in biodiversity and in community ecology. R programming.
- Labs: Institut d’Écologie des Sciences de l’Environnement (IEES), équipe Écologie et Évolution des Réseaux d’interaction (Sorbonne Université, Campus de Jussieu). Centre d’Écologie et des Sciences de la Conservation (CESCO), Museum National d’Histoire Naturelle (MNHN). MIA-Paris (UMR AgroParisTech / INRA).
- Supervision: Élisa Thébault (IEES), Colin Fontaine (CESCO), Pierre Barbillon (MIA Paris).
- Location: Either IEES (Jussieu) or AgroParisTech.
- Duration: 6 months - Starting Date: from Early 2020 to April 2020.
- Contacts: email@example.com, firstname.lastname@example.org, email@example.com.
- Allowance: 540 euros per month.