EGS XXVI General Assembly, Nice, France, March 2001

OA11. The state-of-the-art of numerical weather prediction:

Report

This Session was organized with invited speakers only to provide an overview of the state-of-the-art in some aspects of Numerical Weather Prediction. Dr F. Mesinger (NCEP Washington US; EGS-2001 Bjerknes Medal) gave a talk on "NWP: from the time of Bjerknes to the world of models", K. Emanuel (MIT, Boston US) on "Physical Processes in NWP", Dr T. Palmer (ECMWF, Reading UK) on "Forecasting uncertainty in NWP", D. Dewitt (IRI, New York US) on "Coupled Ocean-Atmosphere Models" and Dr M. Davey (UKMO, Bracknell UK) on "Future Developments in Seasonal NWP". About 500 people attended the talks. Dr F. Mesinger gave an overview of how NWP developed, and discussed the diversity of approaches used in global and limited area NWP. A list of some possible questions to be addressed in the future was presented. Will the diversity of approaches continue to increase, or will it start getting reduced? What about model configuration? Are there good reasons to expect the accuracy to continue increasing, if so how much and in what ways/at what scales; or, does this depend on progress in obtaining higher-density observations? Will the time range of operational limited area NWP continue increasing, or will it shrink, with even limited area NWP being taken over completely by very high resolution global models? Dr K. Emanuel discussed the overall philosophy underlying the representation of subgrid-scale physical processes in numerical models. One of the problems that affect the development of parameterization schemes is that they are poorly tested, or not tested at all, against real data, independently of the model framework. Systematic errors produced by the schemes are masked by tuning other model parameters; this results in models with abundant compensating (or partially compensating) errors. Progress will be accelerated by abandoning this accepted mode of development and implementation of physical parameterizations, in favor of an approach that relies heavily on classical scientific skepticism, with a more clear separation between those who test physical parameterizations and those who develop them. Dr T. Palmer gave a review of the use of ensemble integrations for predicting uncertainty in numerical weather forecasts. Techniques used to simulate uncertainties in both initial conditions and model formulation were discussed. The fact that probabilistic forecasts have more 'value' that single deterministic forecasts was stressed: sometimes not the most likely scenario is what is wanted, but the probability of some 'high-cost' (or risky) situations. Ensemble prediction system can be used to assess these probabilities. The fact that future developments will include both size and resolution increases, and pointed out that different users may benefit differently from an ensemble size or resolution increase. The application of ensemble forecasts for specific user applications was also presented. Dr D. Dewitt discussed the developments of coupled (dynamical) models from the mid-1980, when first used in the study of ENSO. Over the past decade, coupled models increasingly have become the principal modeling tools for studying and predicting seasonal and longer-term climate variability and climate change. The challenge for these models in simulating and predicting the real climate are very great, owing to the strength of coupled feedbacks that can rapidly amplify small errors, and lead to radically different climate states. Different methods developed to minimize the "climate drift" problems were presented. It was argued that one of the reasons why attention in seasonal prediction is mainly focused on the Eastern Pacific is because there is still no clear evidence that a model can beat consistency in other regions. Despite some progress has been made in controlling climate drift in fully coupled models, important problems remain. Dr M. Davey discussed atmosphere-only and ocean-atmosphere General Circulation Models currently used to produce predictions for lead times up to several months. While on the individual model-development side the focus will continue to be on the reduction of systematic errors, the future will see the implementation of large multi-model seasonal ensembles. One of the major challenges will be to convert that forecast information into diverse products suitable for a wide range of applications, both in the public domain and for commercial use. 

Roberto Buizza