Cerfacs Enter the world of high performance ...

DDM- Data Assimilation and Optimisation

Data assimilation deals with the problem of estimating model parameters or model state variables by combining prior estimates and observations, together with information about their uncertainties, in a statistically optimal manner. CERFACS has well-established data assimilation activities in oceanography, atmospheric chemistry, hydrology/hydraulics and wildfires, which are conducted in close collaboration with CERFACS partners, and leading operational centres and research institutes in forecasting and retrospective analysis.

Optimization concerns the development of numerical algorithms for solving diverse linear and nonlinear problems. It involves studying the properties of the algorithms, such as convergence and complexity, as well as improving their performance through preconditioning techniques. CERFACS develops derivative-based and derivative-free algorithms for a variety of applications including data assimilation, Earth imaging and aerodynamics design.

Accuracy, efficiency, scalability, robustness and practicability are important considerations when designing methods for data assimilation and optimization. To develop effective methods requires a good understanding of the underlying application and problem characteristics.

The current actions in Data Assimilation at CERFACS are organized in actions :

  • Develop advanced algorithms for variational and ensemble-variational DA
  • Develop advanced methods for modelling and estimating error covariances
  • Develop ensemble generation and surrogate model methodologies for DA
  • Improve and extend the use of in situ and remotely-sensed data in DA

These actions are implemented on Use Cases

  • Use Case #1: Extending methodologies in variational and ensemble-variational DA: Accounting for nonlinearity; model error estimation & representation; preconditioning; developing robust minimization for extreme-valued observations. Implementation & evaluation in OOPS-based and other systems.
  • Use Case #2: Development of flexible and efficient covariance models for B: multi-variate; accounting for rotational anisotropy in correlations; multi-scale; ensembles; localization; SPDEs; preconditioning; hybrid; coupled; estimation procedures; MLMC. Implementation & evaluation in OOPS-based systems including NEMOVAR for ocean DA, and other systems.
  • Use Case #3: Development of flexible and efficient covariance models for R: unstructured meshes; SPDEs; FEMs; 2D, 3D and 4D (space + inter-channel + time); R-1; preconditioning; estimation procedures. Implementation & evaluation in existing Matlab framework and OOPS-based systems.
  • Use Case #4: Development of efficient ensemble-based DA: Ensemble generation & validation methodologies; surrogate models for stochastic covariance estimation. Development for hydraulics, wildfire & pollutant dispersion.
  • Use Case #5: Assimilation of satellite and image data: high-resolution altimeter (SWOT) and SST; high-resolution & hyper-spectral sounders (IASI); patterns/fronts. Developments for hydraulics & hydrology, ocean, atmospheric chemistry, history matching and wildfire applications.

CALENDAR

Monday

09

October

2023

Machine learning for data science

From Monday 9 October 2023 to Thursday 12 October 2023

  Training    

Tuesday

10

October

2023

PhD Defense: ANTOINE MOUZE-MORNETTAS

Tuesday 10 October 2023 at 14h00

  CNRS ICARE – Orléans - France     Organized by Nathalie BROUSSET    

Monday

23

October

2023

Fundamentals to understand and analyse high fidelity compressible Large Eddy Simulations

From Monday 23 October 2023 to Wednesday 25 October 2023

  Training    

ALL EVENTS