General information
Offer title : PhD in High-dimensional model calibration with machine learning methods to aid sustainable fisheries management M/F (H/F)
Reference : UMR9190-LAUVIC-012
Number of position : 1
Workplace : SETE
Date of publication : 30 June 2025
Type of Contract : FTC PhD student / Offer for thesis
Contract Period : 36 months
Start date of the thesis : 1 October 2025
Proportion of work : Full Time
Remuneration : 2200 gross monthly
Section(s) CN : 51 - Data and biological systems modelling and analysis: computer, mathematical and physical approaches
Description of the thesis topic
High-dimensional model calibration with machine learning methods to support sustainable fisheries management. The project is motivated by the need for operational decision-making tools to halt the decline in marine biodiversity caused by overfishing. To date, there is no rigorous method for high-dimensional calibration of fisheries management models. The thesis (scholarship and environment) is funded by the Institut Mathématiques Pour la Terre (IMPT). The thesis will have the following objectives: 1) Develop a generic method for calibrating complex models, 2) disseminate it as open source code, and 3) propose a concrete application for decision support with the ISIS-Fish model. The objective of this thesis is to propose an innovative approach to improve the high-dimensional calibration of fisheries management models. It is motivated by the need for operational decision-making tools to halt the decline in marine biodiversity caused by overfishing.
The focus is on methodological innovation by exploring advanced dimension reduction techniques such as variational autoencoders, integrating uncertainty quantification essential for assessing model reliability and optimizing this reduction for calibration. The operational objectives are 1) to develop a generic calibration method applicable to complex models with high dimension in its inputs (parameters to be calibrated)-outputs (decision variables), 2) to disseminate it in the form of open source code and 3) to validate these developments on a concrete application for decision support with the ISIS-Fish model recalibrated to simulate fisheries management scenarios. The valorization of the results will be done via scientific publications and dissemination to decision-makers. The work axes will be
Axis 1: Appropriation of the state of the art of calibration methods in high dimension and handling of the ISIS-Fish model. Axis 2: Genericity of the approach applicable to other models having parameters to be calibrated in high dimension. Dissemination of the method in the form of a free and documented code. Axis 3: Validation of the new calibration method on ISIS-Fish and studies of fisheries management scenarios. The candidate must have an academic background in statistics or applied mathematics with a strong interest in applications in environmental sciences. He / She must understand or have experienced mechanistic modeling approaches, statistical methods classically used for the exploration of these models (sensitivity analysis and calibration) and computer tools such as R, github, ... A good command of English is essential. An interest in fisheries will be appreciated.
Work Context
The recruited person will join the UMR MARBEC. The Joint Research Unit (UMR) MARBEC, MARine Biodiversity, Exploitation and Conservation, was created on January 1, 2015. Its supervisory authorities are the IRD, Ifremer, the University of Montpellier and the CNRS. MARBEC is one of the largest laboratories working on marine biodiversity and its uses in France with approximately 300 staff, including 90 researchers and lecturer-researchers. The unit is located in Sète, Montpellier and Palavas-les-flots, as well as in the Indian Ocean, Asia, Africa and South America. It studies the marine biodiversity of lagoon, coastal and offshore ecosystems, mainly Mediterranean and tropical. Its research focuses on different levels of integration, from molecular, individual, population and community aspects, to the uses of this biodiversity by humans. Material scientific conditions: The work will be carried out mainly in Sète, hosted by the UMR MARBEC at the Ifremer station in Sète. Regular missions to Paris-Saclay will be carried out for the work with Pierre Barbillon, Lecturer and co-supervisor at AgroParisTech. The supervisor and co-supervisor have the technical skills to achieve the project objectives, skills that will be passed on to the recruited person.
-Financial conditions: The thesis (scholarship and environment) is financed by the Institute of Mathematics for the Earth (IMPT). This funding will cover the travel expenses of the recruited person to go to Paris. As a member of the UMR MARBEC, he/she will have a personal amount of €1,500 for 3 years that he/she can use for his/her activity (conference fees, specific training, etc.). This project will mainly be the subject of a thesis under the co-supervision of Pierre Barbillon and Stéphanie Mahévas. The doctoral student will be hosted at AgroParisTech, UMR MIA Paris-Saclay and at Ifremer Sete UMR MARBEC. To ensure eco-responsible management of this project, we are committed to carrying out our missions by train, to favoring videoconferencing, to participating in an international conference in Europe accessible by train and to reasoning the use of supercomputers by being vigilant about the number of launches of calculation codes and backups of model outputs.
The recruited person will have a computer, access to IFREMER's DATARMOR supercomputer, an office and the necessary equipment to carry out their work. This project will be an integral part of the discussions and work carried out within the framework of two national research networks, the INRAE Mexico network (https://reseau-mexico.fr/) and the RT-UQ thematic network (https://uq.math.cnrs.fr/). The generic dimension of the approach, more particularly highlighted in axis 2 of the project, will have direct repercussions in the community of these two networks. Possibility of supervising interns.
Constraints and risks
No risk identified.