This project focuses on the study and development of generative models capable of reproducing heavy-tailed distributions, particularly those derived from radar-observed rainfall fields. Current generative models—VAEs, GANs, and diffusion models—excel on complex data such as images or videos, but their theoretical guarantees generally rely on assumptions incompatible with heavy distributions, which characterize extreme rainfall events. However, the realistic modeling and generation of these extremes are essential, especially in the context of climate change where their frequency is expected to increase.
Projects
This project focuses on improving the computational efficiency of parameter estimation in complex spatiotemporal geostatistical models. Geostatistics aims to model natural phenomena evolving in space and time, and Gaussian processes are widely used to predict these phenomena in unobserved locations, while quantifying the uncertainty of the predictions. While classical models rely on simple parametric covariance functions, recent advances have introduced models based on stochastic partial differential equations (SPDEs), which are better suited to representing physical mechanisms such as advection or diffusion.
The objective of this project is to develop a joint model of headwater flow and water quality using a generative AI approach. To achieve this, a diffusion model is fitted to multi-site and multi-variable time series and adapted to allow training on incomplete data and simulation in the following contexts: gap-filling, nowcasting, synthetic time series generation, and conditional simulation with covariates. The outputs of this diffusion model will be used to study the impact of different climate change and/or anthropogenic scenarios on headwater hydrosystems and to emulate certain hydro-meteorological parameters used as inputs for impact models (hydrogeology, stream ecology, etc.).
The aim of this project is to develop statistical methods for interpolation and prediction of spatio-temporal data distributed on surfaces. These methods will be applied to data from temperature and deformation sensors placed at certain points on the surface of a cell in order to monitor its evolution.
This project aims to merge probabilistic methods for marked point processes, extreme events, and spatial graphs to propose a new framework for better understanding and simulation of spatial and temporal interactions between different types of extreme events in the climate system.
The objective of this project is to develop and implement models for multivariate extreme events assuming a (known) graph structure representing the links between the different variables. The model will be applied to extreme climatic events (temperatures, precipitation, etc.) occurring in different geographic areas (departments, regions, etc.), for example by choosing a graph representing the structure of the geographic neighborhoods of the regions.
In recent years, hydrological extremes, in particular intense rains and long droughts, have been at the heart of concerns in many sectors (agronomy, energy, insurance, etc.). However, the great variability of these phenomena in time and space requires stochastic modeling in order to incorporate several hazards, uncertainties and heterogeneous information (measurements, climate models, reanalyses). The theme of the thesis is therefore centered around stochastic spatio-temporal simulation models of intense precipitation and long droughts.
The objective of this project is to implement a stochastic precipitation generator using a generative AI type approach in order to simulate high-resolution spatio-temporal rain fields resembling as much as possible those observed by weather radars.
To perform simulations of precipitation fields over large domains, as is the case here with the simulation of precipitation fields at the national scale, non-stationary approaches that are efficient enough to deal with large grids must be considered.
Some of the most devastating climatic events of recent decades involve the combination of several climatic variables on fairly large scales and over a period of time. Sometimes combinations of elementary climatic events which are not necessarily devastating when isolated can have devastating effects when they occur simultaneously or successively within a short period.