Debris flow hazard quantification with meta-models

Back to jobs listing

INRAE presentation

The French National Research Institute for Agriculture, Food, and Environment (INRAE) is a major player in research and innovation. It is a community of 12,000 people with 272 research, experimental research, and support units located in 18 regional centres throughout France. Internationally, INRAE is among the top research organisations in the agricultural and food sciences, plant and animal sciences, as well as in ecology and environmental science. It is the world’s leading research organisation specialising in agriculture, food and the environment. INRAE’s goal is to be a key player in the transitions necessary to address major global challenges. Faced with a growing world population, climate change, resource scarcity, and declining biodiversity, the Institute has a major role to play in building solutions and supporting the necessary acceleration of agricultural, food and environmental transitions.

Work environment, missions and activities

You will be hosted in the UMR IGE (Institute of Environmental Geosciences, Grenoble), which carries out research on climate, anthropization, and environmental risks in polar regions, the intertropical zone, and mountainous areas. You will join the ECRINS team working on the assessment and prevention of natural risks in mountainous areas (avalanches, torrential floods, landslides, rockfalls, glacier-related risks, etc.). This thesis is part of a collaboration with the Risk Direction (DRIS) of the BRGM, and in particular the RMVT unit (gravitational risks). As such, you will be required to share your time between the two sites.

Your work will focus on the modeling of hydro-gravitational flows such as debris flows and mudflows. More specifically, this project aims to develop new efficient and reliable methods for hazard mapping at both site-specific and regional scales. Current methods rely either on empirical approaches that are quick to implement but relatively simplistic, or on physically-based numerical models that better account for the complexity of processes but are computationally expensive. Meta-modeling, which involves developing statistical approximations trained from a database of simulations, offers a promising alternative to preserve the predictive capabilities of physically-based models while reducing computational costs.

The objective of the thesis is to develop and test a meta-modeling approach applied to thin-layer models used to simulate the propagation of debris flows. This approach will allow for the propagation of uncertainties in model parameters and initial conditions, thereby establishing probabilistic hazard maps. Several scientific and methodological challenges will need to be addressed, related in particular to the high dimensionality of the variables to be estimated (spatial predictions) and the non-linear responses of the model. This exploratory work could serve as a basis for similar studies on other phenomena such as snow and rock avalanches.

 

The study will consist of three main stages:

• Training a simplified meta-model (scalar outputs with or without temporal dependence) to reproduce field observations and infer the distribution of rheological parameters associated with the observed flows;
• Training a meta-model that takes rheological parameters and the initial condition provided in the form of an input hydrograph to estimate spatial outputs (flood maps);
• Using the meta-model to propagate the variability in initial conditions and rheological parameters to obtain probabilistic outputs.

The work will initially focus on the scale of a watershed and its alluvial fan. The possibility of extending the approach to a regional scale will then be explored. You will use the flow models developed by the host teams, namely Shaltop and Lave2D/SaVal-2D. You will adapt and/or extend meta-modeling approaches already proven in other contexts, as well as dimensionality reduction and adaptive sampling techniques.

You will be specifically responsible for:

• Conducting an in-depth bibliographic study of the state of the art in the field;
• Defining and testing the meta-modeling methodology for each stage of the study;
• Building the training datasets to train the meta-models;
• Validating the developed tools on study sites to be defined;
• Ensuring the dissemination and archiving (GitHub repository) of the developed codes;
• Regularly presenting and discussing the results with the supervision team;
• Writing and publishing at least two articles in international journals;
• Presenting the study results in at least one international conference.

• Thesis located at both IGE (Grenoble) and BRGM (Orléans): to enable optimal collaboration between the two teams, you will spend at least one year at BRGM.
• Occasional field work in mountainous areas.

Training and skills

• Recommended education: Master's Degree in mechanical modeling / Master's Degree in natural risks / Master's Degree in statistics / Engineering school.
• Expected knowledge: Proficiency in programming (Python/R), numerical modeling and statistics is essential, along with a good command of English (written and spoken). Knowledge in hydraulic modeling, risk science, or meta-modeling will be appreciated.
• Preferred experience: Research internship on the topic of natural risks or hydraulic modeling.
• Desired skills: autonomy, rigor, analytical and writing abilities.

INRAE's life quality

By joining our teams, you benefit from (depending on the type of contract and its duration):

- up to 30 days of annual leave + 15 days "Reduction of Working Time" (for a full time);
- parenting support: CESU childcare, leisure services;
- skills development systems: training, career advise;
- social support: advice and listening, social assistance and loans;
- holiday and leisure services: holiday vouchers, accommodation at preferential rates;
- sports and cultural activities;
- collective catering.

How to apply

I send my CV and my motivation letter