Workshop on Machine Learning for Weather and Climate Modelling held

Members of the predictability group organised a workshop to discuss the value of machine learning to improve weather and climate forecasting. The four day workshop, held at Corpus Christi College (2-5 September), was organised by Tim Palmer, Hannah Christensen, Matthew Chantry and Philippa Towler from the predictability group alongside Peter Duben from ECMWF. The workshop welcomed attendees from all around the world with representatives from academia, operational forecasting centres and industry.

The workshop discussed recent developments in machine learning and the impact that these could have on improving forecasts on timescales from nowcasting to climate. Sessions discussed how these tools could aid all areas of forecasting, from data-assimilation to post-processing. Discussions were often framed in terms of soft to hard AI, where soft used machine learning as an acceleration tool for existing code and hard attempted to replace the entire dynamic model. A particular focus of the workshop was physical parameterisation schemes, one of the major uncertainties in weather and climate forecasting. Attendees heard about range of approaches to the parameterisation problem across the spectrum of soft to hard AI.

One highlight of the workshop was breakout groups which gathered together many experts in the field to discuss the next 5 years in this topic. These breakout groups discussed the role of AI in postprocessing, how to use AI in a high performance computing context, how to encode physical understanding into machine learning and how machine learning might be used to create improved parameterisation schemes.

The conference was possible thanks to financial support from Amazon, the Copernicus Atmosphere Monitoring Service (C3S), the Copernicus Climate Change Service (C3S), the ESiWACE Centre of Excellence, NVIDIA, the Office of Naval Research and Vulcan.