The numerical weather prediction model COSMO

MeteoSwiss operates and further developes a high-precision numerical weather prediction system, COSMO, in order to automatically generate regional and local forecasts products in complex topography. A detailed image of the future state of the atmosphere is computed, from the low stratosphere to the surface, including the evolution of the snow cover, the lake temperature and the soil characteristics. COSMO-7 products are available up to three days in advance on a domain covering central and western Europe, and more detailed COSMO-2 products are available up to 33 hours in advance on a domain including the Alpine Arch. Figure 1 shows the model suite of MeteoSwiss: The European Center for Medium Range Weather Forecast (ECMWF) operates a global model describing the synoptic scales. MeteoSwiss operates the regional scale COSMO-7 and the local scale COSMO-2 models.

The products of COSMO-7 and COSMO-2 provide a quantitative guidance to the forecasters for their daily forecasts, contribute to the security of the Swiss population by the generation of warnings, e.g. in case of high-impact weather, floods or incidents in nuclear power plants, and respond to the specific needs of many professional users. The COSMO system is permanently monitored, a prerequisite for a high forecast quality.

A numerical weather prediction model is based on the physical laws describing the atmospheric and surface processes (e.g. conservation of energy, transformation of phase, black body radiation), suitable initial and boundary conditions, and a numerical method to solve in time the resulting system of complex mathematical equations. These equations describe a variety of atmospheric processes on different temporal and spatial scales, for example the development of low pressure systems, snow fall, Föhn, summer convection. The calculations are performed on a three-dimensional grid, where the vertical spacing of the computational levels is inhomogeneous in order to better resolve the low level atmospheric conditions.

Suitable initial conditions for starting a forecast are produced by combining actual atmospheric observations, previous model guidance, and climatic information in a way compatible with the model equations; this is the so-called data assimilation process. In a typical 24 hours assimilation period, COSMO-7 ingests

1. about 120 vertical soundings,
2. about 8000 aircraft observations,
3. about 28000 surface observations, and
4. about 1000 wind profiler measurements.

In addition, COSMO-2 ingests radar-derived rainfall data to better represent convective activities.

A forecast with a numerical weather prediction model is computationally very demanding and time critical. A typical 24 hours COSMO-2 forecast requires more than 100 trillion mathematical operations and must be computed within 20 minutes. That's why the COSMO system is composed of 3 nested numerical models: one global model simulating the whole atmosphere but with medium precision, and two limited area models each with increased precision. That's also why the evolution of the models always went side by side with the development of the fastest supercomputers in the last twenty years.

COSMO is developed within an international collaboration. The six national weather services of Germany, Greece, Italy, Poland, Romania, Russia, and Switzerland closely collaborate inside the Consortium for Small-Scale Modelling (COSMO). This consortium was formed in October 1998; it's general goal is to develop, improve and maintain a non-hydrostatic limited-area atmospheric model, to be used both for operational and for research applications by the members of the consortium.