Response of the Physical Aspects Section of the Research Department of ECMWF to the WCRP consultation on Model Evaluation and Improvement

This note has been prepared by the Physical Aspects Section of the Research Department at ECMWF. ECMWF is an operational centre for medium range weather forecasting, monthly forecasting, seasonal forecasting and ocean wave forecasting. The operational system is also used for re-analysis projects (ERA-15, ERA-40 and ERA-Interim). ECMWF operates a state of the art global atmospheric model with an integrated data assimilation system. For monthly and seasonal forecasting the atmospheric model is coupled to an ocean model. Resolution varies from T159 (110 km) for the seasonal system to T799 (25 km) for the 10-day deterministic forecasts. A resolution upgrade to T1279 (15 km) is currently under test and will become operational late 2009 or early 2010. Intermediate resolutions are used for monthly forecasting and for the ensemble prediction system. Given these application, ECMWF is interested in atmospheric processes like radiation, convection, clouds, precipitation, and turbulence, but it also includes all aspects of atmospheric dynamics, monsoons, troposphere-stratosphere exchange, ocean and land interactions, snow modelling, and sea-ice feedbacks. Through a recent European Project, ECMWF is getting increasingly involved in comprehensive earth system modelling. A prototype system has been developed for the modelling and assimilation of aerosols, greenhouse gases and reactive gases. The Physical Aspects Section is responsible for the maintenance and further development of the package of parametrized processes in the ECMWF model covering a wide range of resolutions, applications and time ranges. It is therefore faced with the same questions as the ones asked in the WCRP-questionnaire. The basic question is: How to implement the best possible physics into a large scale model and obtain good results? This is by no means trivial, because good physics does not necessarily give good results in a full model because of compensating errors and poorly understood interactions. This issue is often referred to as the implementation bottle neck.