P. Lynch / Journal of Computational Physics 227 (2008) 3431–3444                              3439

dynamical formulation. Many early coupled models needed a flux adjustment (additional artificial heat and moisture fluxes at the ocean surface) to produce good simulations. The higher ocean resolution of HadCM3 was a major factor in removing this requirement. The atmospheric component of HadCM3 has 19 levels and a latitude/longitude resolution of 2.5 · 3.75, with grid of 96 · 73 points covering the globe. The resolution is about 417 · 278 km at the Equator. The physical parameterization package of the model is very sophisticated. The oceanic component of HadCM3 has 20 levels with a horizontal resolution of 1.25 · 1.25 permitting important details in the oceanic current structure to be represented. HadCM3 and HADGEM have been used for a wide range of climate studies which provided crucial inputs to the Fourth Assessment Report (AR4) of the Intergovernmental Panel on Climate Change (IPCC), published in 2007. The development of comprehensive models of the atmosphere is undoubtedly one of the finest achievements of meteorology in the 20th century. Advanced models are under continuing refinement and extension, and are increasing in sophistication and comprehensiveness. They simulate not only the atmosphere and oceans but also a wide range of geophysical, chemical and biological processes and feedbacks. The models, now called Earth System Models, are applied to the eminently practical problem of weather prediction and also to the study of climate variability and mankind’s impact on it.

3. Numerical weather prediction today
It is no exaggeration to describe the advances made over the past half century as revolutionary. Thanks to this work, meteorology is now firmly established as a quantitative science, and its value and validity are demonstrated daily by the acid test of any science, its ability to predict the future. Operational forecasting today uses guidance from a wide range of models. In most centres a combination of global and local models is used. By way of illustration, we will consider the global model of the European Centre for medium-range weather forecasts.

3.1. The European centre for medium-range weather forecasts
Perhaps the most important event in European meteorology over the last half-century was the establishment of the European Centre for medium-range weather forecasts (ECMWF). The mission of ECMWF is to deliver weather forecasts of increasingly high quality and scope from a few days to a few seasons ahead. The Centre has been spectacularly successful in fulfilling its mission, and continues to develop forecasts and other products of steadily increasing accuracy and value, maintaining its position as a world leader. The first operational forecasts were made on 1 August, 1979. The Centre is currently undergoing enlargement. A new Convention has been agreed and is in the process of ratification. The ECMWF model is a spectral primitive equation model with a semi-lagrangian, semi-implicit time scheme and a comprehensive treatment of physical processes. It is coupled interactively to an ocean wave model. Its spatial resolution is 25 km and there are 91 vertical levels. Initial data for the forecasts are prepared using a four-dimensional variational assimilation scheme, which uses a large range of conventional and satellite observations over a 12-hour time window. A sustained and consolidated research effort has been devoted to exploiting quantitative data from satellites, and now these observations are crucial to forecast quality. ECMWF produces a wide range of global atmospheric and marine forecasts and disseminates them on a regular schedule to its Member States. The primary products are listed here (explanations of technical terms
will follow).
Forecasts for the atmosphere out to 10 days ahead, based on a T799 (25 km) 91-level (L91) deterministic model are disseminated twice per day.
Forecasts from the ensemble prediction system (EPS) using a T399 (50 km) L62 version of the model and an ensemble of 51 members are computed and disseminated twice per day.
Forecasts out to one month ahead, based on ensembles using a resolution of T255 (78 km) and 62 levels are distributed once per week.
Seasonal Forecasts out to six months ahead, based on ensembles with a T159 (125 km) L40 model are disseminated
once per month.