Data Assimilation and Machine Learning: Evolution or revolution?

Machine Learning has proved to be an innovative, disruptive set of technologies whose deployment has revolutionised best practices in diverse fields of applied science and engineering.

ECMWF and other Weather Forecast Centres have recently started on the path to introducing ML components in its DA, NWP and Climate workflows, with some promising results which are already in or close to operational implementation. However, the speed of evolution of ML applications in NWP/Climate prediction and the appearance on the scene of big technology companies have caused waves in the previously tranquil operational NWP/Climate ecosystem and posed some difficult questions on the future sustainability of current R&D and operational practices in Data Assimilation, NWP and Climate.

In this talk I will give an overview of current ECMWF efforts in hybridising our NWP workflow with ML technologies with a focus on data assimilation applications, discuss potential and limitations of these technologies and share my perspective on the likely evolution of the field.