Marie-Pierre Etienne

Assistant Professor UMR AgroParisTech/INRA 518 Team MORSE 16 rue Claude Bernard 75 231 Paris Cedex 05 Tel : +33 (0)1 44 08 16 72 fax : +33 (0)1 44 08 16 66 Currently at the Fisheries Center http://www.fisheries.ubc.ca/

Hierarchical modeling is a powerful framework which allows wide flexibility in model conception.
The studies related to environmental issues mostly occur in some uncontrolled design. Therefore classical statistical models are not pertinent to study the obtained data (spatial dependence, multi scales, zero inflated, large p small n ….). Furthermore, the ecological processes of interest are mostly hidden and data are just a truncated observation of those processes (tuna declared captures as an indicator of the tuna biomass, number of prey in stomach as a measure of tuna feeding, acceleration data for trajectories of fur seals …) Hierarchical modeling is a useful framework to propose adequate models to link observations and latent processes of interest. I work in collaboration with biologists and ecologists in order to develop and analyze statistical models which a strong interest in highlighting the links between biological and mathematical assumptions. Hierarchical Bayesian Model are then a powerful tool for modelling ecological process but the counterpart is the potential difficulty of estimations of such models. The limits of classical algorithms for the estimation of hidden variables models like EM, MCEM in a frequentist framework or Hastings-Metropolis, Gibbs sampler in a Bayesian framework are reached and put some limitations on the modelling process.

Current Focus on :

• Abundance indices : Managing a fish population requires to have accurate information of the actual level of this popupaltion (Bluefin Tuna is one of the most emblematic example), how can me help Fisheries Scientist to define reliable abundance indices ?
• Trajectories Analysis : A population is not only described by its size but also by the spatial and temporal movements. Telemetry is the measure of individual movements using various Tags. It's more and more common in ecology to study fine scale behaviour of animals, from cervids, bovine animals to fish using individuals tags. The resulting data can be very different in term of density (1 GPS position every day to one position every 15mn, measure of depth every second) but also the nature of the data highly depends of the kind of tags used. Available data can be GPS positions, 3D accelerations, 3D compas, depths, …

Publications List

Random effects compound Poisson model to represent data with extra zeros

Extracting abundance indices from longline surveys : method to account for hook competition and unbaited hooks

2010 June : TIES Conference, Venezuela : Accounting For Interspecific Competition And Empty Hooks When Building Relative Abundance Indices From Longline Catch.

2009 December : Pacific Halibut Commission Presentation : Abundance indices and empty hooks

2009 August : useR 2009 Coalition : a simple and useful tool to distribute R-works on a set of computers. Coalition is available on Google Code

• Coordinator of regular courses at AgroParistech (since 2011) on Linear Models

• Coordinator of regular courses in Master degree in Ecological science EBE AgroParisTech (since 2006)
  • Introduction to Statistics,
  • Advanced Statistics for Ecological Science
  • Random Model in ecology and environment

• Coordinator of Hierarchical and Bayesian Modeling in forest science using OpenBugs for Amazonian doctoral formation program (AgroParisTech-Kourou). Some notes on Bayesian modelling, R and openBugs in french

• Regular courses at AgroParisTech-ENGREF
  • Statistical Modeling for risk assessment and management
  • Simulation and Prediction using random modeling

• Fall 2007 : Hierarchical and Bayesian Modeling in marine science using OpenBugs, professional formation for Fisheries science center (Centre de Ressourece halieutique ,CRH, Sete)

Coalition : a Task Scheduler. This simple tool is very useful to schedule tasks with R (by instance simulations study) useR2009 presentation. Presentation of Coalition at AgroParisTech http://www.agroparistech.fr/mmip/maths/essaimia/accueil:intranet:coalition

INLA : Integrated Nested Laplace Approximation (INLA) is a method which approximates posterior distribution in a Gaussian Markovian Random Fields framework. The program inla and all the documentations are available at Havard Rue Homepage . How to install R-INLA under Ubuntu distribution for newbie.Installation Notes for R -INLA

Latex package Installation

Member of Environmental Groupe at SFDS

Leader of the groupe Statistics for trajectories (funded by INRA, MIA research division)