Training course: Numerical methods for weather prediction

This course describes contemporary methods, recent advancements, and the challenges in numerical techniques which are used across a wide spectrum of weather prediction models for solving the governing equations of atmospheric dynamics on High-Performance Computing (HPC) systems. The covered methods have relevance in short-term and medium-term forecasting, seasonal forecasting, climate prediction, as well as global and limited-area modelling.

Topics

• Governing equations and hydrostatic/non-hydrostatic dynamics
• The spectral transform, semi-implicit, semi-Lagrangian integration method of the ECMWF model dynamical core
• Horizontal and vertical discretizations
• Advection schemes, time-integration methods, semi-implicit methods and elliptic solvers
• Finite volume and finite element methods, high-order discontinuous spectral element discretization methods
• Structured and quasi-uniform grids, unstructured meshes and mesh generation
• Massively parallel computing
• Contrasts and synergies between established numerical weather prediction models and currently explored machine learning models for weather prediction

Requirements

Participants should have a good mathematical background and are expected to be familiar with the contents of standard meteorological and mathematical textbooks. A familiarity with python is useful for some of the practical sessions.

Other complementary material can be found in our lecture note series.

Some general knowledge or experience in numerical weather prediction is an advantage.

All lectures will be given in English.