Objectives

The activity on uncertainty quantification at CERFACS aims to estimate uncertainties for numerical models requiring large computational resources. We develop environmental risk assessments, data assimilation approaches, probabilistic optimization and coupling improvements for High Performance Computing applications.

At CERFACS, the activity on uncertainty quantification relates to the transversal axis “Data-Driven Modeling” and is at the crossroads of the Environment, Climate, Aerodynamics and Combustion thematic axes. Applications at CERFACS concern flood forecasting, prediction of atmospheric dispersion of pollutants, wildfire propagation, representation of climatic variability and combustion chamber ignition calculations.

The main actions are :

• Ensemble-based simulations with scalar and functional variables, including dimension reduction strategy
• Development and evaluation of reduced models for large scale problems
• Use of reduced models for sensitivity analysis, optimization and data assimilation
• Development of efficient algorithms for stochastic estimation with solvers of increasing complexity (multi-fidelity, multi-level Monte Carlo/MLMC)

These actions are deployed on the following applications:

• Development of uncertainty quantification algorithms for real scale computation with efficient, scalable and robust domain decomposition algorithms
• Development of reduced models for sensitivity analysis and ensemble data assimilation, application in hydraulics and aerodynamics for large uncertain variables
• Application of reduced models for atmospheric boundary layer simulations in the context of uncertainty quantification, application to micro-scale meteorology and in particular to pollutant dispersion
• Application of multi-fidelity and MLMC algorithms for industrial computational fluid mechanics, multidisciplinary systems and geosciences.
• Parametric sensitivity tests of global climate model projections to assess uncertainties in regional and global climate risks
• Simple climate models to assess data constraints on global climate projection uncertainty

Collaboration