[Post-doc completed] Feriel Bouhadjera: LADi D: Long-term Analysis of Dieback Drivers. Analysis of the determinants of multi-year vineyard dieback trajectories

LADi D: Long-term Analysis of Dieback Drivers. Analysis of the determinants of multi-year vineyard dieback trajectories by Bayesian linear regression for functional data

I did a Master's degree in Probability and Statistics, followed by a thesis in Mathematical Statistics. After my thesis, I moved on to an ATER position at the University of Lille and finally, since last September, I've been a postdoctoral researcher at the Mistea laboratory, Institut Agro, Montpellier. My thesis focuses on the development of regression function estimation methods for incomplete data with (or without) some form of dependency.
My tasks as a post-doctoral researcher are :
(i) a diagnosis of the main determinants of vine decline (yield and mortality) and of the temporality of their interventions, in the regional context of Cognac
(ii) the evaluation of innovative data analysis methodologies for processing the same type of databases produced by professional organizations or companies in other regions, in viticulture or in other agricultural sectors
(iii) the extension of an existing R package (BLISS) with the generic statistical models to be developed in the project.
I have chosen this theme in order to contribute to the development of methodologies and tools to address real-life issues arising from the life sciences.

• Starting date:  1st September 2021
• Scientific field: Statistics/data sciences
• Funding: Numev (12 months) – #DigitAg (9 months)
• Post-doc supervisors: Baragatti Meili (Institut Agro, Mistea) et Nadine Hilgert (Inrae, AbSyS)
• #DigitAg : Axe 5 : Fouille de données, analyse de données, extraction de connaissances, Challenge 1 : Le challenge agroécologique, Challenge 3 : La protection des cultures

Keywords: Bayesian statistics, Regression function, Functional data, Bliss method

Abstract: In recent years, the wine industry has been complaining about a drop in yields and vine longevity. Vineyard dieback is a syndrome that results in a drop in productivity over several years and the premature death of certain vines. A particular feature of vine dieback is that it occurs over long periods of time, from a few years to a few decades. Few data are available over such long periods. The Bureau National Interprofessionnel du Cognac (BNIC) has monitored 55 plots since 1977. This unique database offers the opportunity to analyse the determinants of multi-year vine dieback trajectories, notably by undertaking a "dynamic" characterisation of vine yield and mortality (i.e. with explanatory variables that evolve over time). The objective of this post-doc project is to identify (1) the factors and interactions of biotic, abiotic and technical factors that contribute to the decline in plot yield and to the mortality of individual grapevines and (2) the time period over which these factors have an impact, both in the short term at the scale of the crop cycle and in the long term since the plot was planted. Exploratory statistical analyses will be required, as well as methodological developments in linear regression for functional data. Beyond the analysis of this case study, the post-doc will produce a generic protocol for the analysis of temporal data on agricultural systems that will fill a gap in statistical tools for the analysis of agroecological transition trajectories.

Contact : meili.baragatti [AT] supagro.fr - Tél: 07.52.23.33.97

Social networks: ResearchGate – LinkedIn

Communications & Papers:

Thesis manuscript (lien vers texte intégral en libre-accès) : https://tel.archives-ouvertes.fr/tel-03134914

Bouhadjera F. , Ould SaïdE. & Remita M.R. (2021). Strong consistency of the nonparametric local linear regression estimation under censorship model. Communication in statistics : Theory and Practice.

Benrabah O. , Bouhadjera F. & Ould Saïd E. (2021). Local linear estimation of the regression function for twice censored data. Statistical Papers.

Bouhadjera F. & Ould SaïdE.(2021). Asymptotic normality of the relative error regression function estimator for censored time series data. Dependence Modeling.

Bouhadjera F. & Ould Saïd E. (2021). Nonparametric local linear estimation of the relative error regression function for twice censored data. Statistics and Probability Letters

Bouhadjera F. & Ould Saïd  E. (2021). Relative error prediction: Strong uniform consistency for censoring time series model. Communication in Statistics : Theory and Methods

Oral presentations 

• Consortuim Phenodyn, November 29th 2021, Inrae, Montpellier (France).
• Team seminar (IMAG), October 18th 2021, Montpellier (France).
• Team seminar (LAMAV), May 6th 2021, Valencienne (France).
• Team seminar (Labo. Paul Painlevé), March 2021, Lille (France).
• Team seminar (Labo. Statistique et Optimisation), March 2021, Toulouse (France).
• Team seminar (Labo. Probab. Statist. et Appli.), Febrary 2021, Poitiers (France).
• Team seminar (Labo. Statist. ), Febrary 2021, Strasbourg (France).
• Team seminar (Labo. Probab. Statist. et Appli.), January 2021, Alger (Algérie).
• Seminar of Ph.D students (IMAG), October 2019, Montpellier (France).
• 13th Nord-Pas-de-Calais doctoral meeting in mathematics, September 2019, Lens (France).
• International Conference on Mathematical Modeling and Applications, April 2019, Rabat (Maroc).
• Team seminar (LMPA), March 2019, Calais (France).
• Young researchers’ days (LaPS),December 2018, Annaba (Algérie).
• Team seminar (Labo. Probab. Statist. et Appli.), December 2018, Alger (Algérie).

Seminar of Ph.D students (LMPA), November 2018, Calais (France).