Workshop on model uncertainty

ECMWF | Reading | 9-12 May 2022


In order to correctly describe the growth of forecast uncertainties, ensembles require representations of model uncertainties. These matter on all time scales from hours in data assimilation to months in seasonal forecasts. Developing these representations of model uncertainties is a major scientific challenge as random model errors are generally not directly observable and there are sources of model errors in all Earth system components. This workshop was an opportunity to share recent advances in the field, to identify common interests and open research questions. While acknowledging the diversity of approaches that has emerged over the past two decades, the workshop primarily focussed on stochastic representations of model uncertainties and the following five main themes:

  • Uncertainty representations in convection-permitting ensembles
  • Physical consistency of perturbations
  • Stochastic parametrisation closures
  • Surface interface uncertainties (atmosphere-land/ocean/wave/sea-ice)
  • Evaluation of model uncertainty representations

ECMWF sought input and recommendations from the workshop participants regarding future work on these topics and especially encouraged contributions describing research and development for improving model uncertainty representations in operational convection permitting ensembles.


The workshop took place at the ECMWF Headquarters in Reading, 9-12 May 2022, and adopted a hybrid format with online and physical participation.