COSMO forecasting system

Forecasting systems calculate future atmospheric conditions. MeteoSwiss uses the COSMO (Consortium for Small-Scale Modelling) numerical weather forecasting model for the production of local and regional forecast products in the topographically challenging Alpine region and across Switzerland. The model also supports meteorologists in issuing storm and flood warnings.

Thanks to complex computer models, it is possible to simulate how the weather behaves. MeteoSwiss uses the COSMO numerical forecasting model for this purpose. The model calculates high-resolution weather forecasts for Europe several times a day, with a particular focus on the Alpine region.

The COSMO forecasting model calculates future atmospheric conditions in detail from ground level up to an altitude of around 20 kilometres. These calculations also cover the development of snow cover, lake temperatures and surface properties. The model is being developed in close cooperation with international partners.

COSMO: a numerical weather prediction model

A numerical weather forecasting model describes processes which take place in the atmosphere and on the Earth's surface. It is based on physical laws, including the conservation of energy, mass and momentum as well as the phase transitions of water and radiation processes.

Through the application of appropriate initial and boundary fields, future atmospheric conditions can be calculated. This allows for a variety of atmospheric processes to be described on different temporal and spatial scales (e.g. development of an area of low pressure, snowfall, Foehn wind, convection). The calculations are made on a three-dimensional grid. The vertical distances between the grid points are smaller at lower altitudes than at high altitudes to allow for a more detailed description of phenomena close to the ground.

The COSMO models

Numerical weather models that cover the entire globe are not precise enough to create regional or even local forecasts. The weather forecasts for Switzerland – with its complex topography – are therefore based on small-scale models. MeteoSwiss operates models with differing resolution for different purposes. Resolution here means the distance between the grid points for which the weather parameters are calculated.

  • COSMO-1: Model for the short term forecast for the present and the next day. The distance between the grid points is 1.1 km.
  • COSMO-E: This model computes a probabilistic forecast in terms of a probability of occurrence for certain weather events, up to five days in advance. The distance between the grid points is 2.2 km.
  • COSMO-7: The distance between the grid points is here 6.6 km. The products of COSMO-7 are created up to three days in advance and cover central and western Europe.

The European Centre for Medium-Range Weather Forecasts (ECMWF) provides a global forecast. This is used a boundary condition for the regional COSMO models.

Calculation of the COSMO models at the CSCS

All calculations are performed at the Swiss National Supercomputing Centre (CSCS) in Lugano on a massively parallel Cray computer. COSMO-1 forecasts cover up to 33 hours in the future and are initiated eight times a day (every three hours). The first calculation begins at 00:45 UTC. COSMO-E forecasts cover up to 120 hours in the future and are initiated two times a day at 00:45 and 12:45 UTC.

Within the assimilation cycle, the latest analysis is calculated with the COSMO model every three hours. These analyses provide the initial state for the forecasts. The model performs its calculations with hindsight and uses all of the observations on the ground and, in particular, in the free atmosphere in order to generate an optimal description of atmospheric conditions. For COSMO-1, the radar data is also assimilated, providing significant added value for short-term forecasts ("nowcasting").


Using nowcasting, MeteoSwiss provides high spatial and temporal resolution forecasts of weather developments for the next few minutes and up to a...

Processing of 35,000 pieces of information over 24 hours

Suitable initial and boundary fields for initiating a model are obtained by combining observation data, previous model runs and climatological information. The incorporation of this data in the current model run is referred to as assimilation. As part of the assimilation cycle, COSMO-7 processes information from around

  • 100 radio soundings
  • 3'500 aircraft observations
  • 31'000 stations observations, and
  • 700 wind profiler measurements.

COSMO-1 also processes 7,000 images from the rain radar each day, allowing for the improved representation of the processes involved in thunderstorm formation.

More than 4 quadrillion floating-point operations for a forecast

The generation of forecasts using numerical weather prediction models places high demands on computing power. For a 24-hour forecast created using the COSMO-1 model, more than 4 quadrillion mathematical operations are typically required. These have to be executed within 24 minutes. Among other factors, the continuous development of faster super computers over the past 50 years has enabled a steady increase in the precision of numerical weather simulations.

The benefit of a higher degree of resolution becomes clear in complex terrain, as shown by the following example of a wind simulation.

Successful international cooperation

To allow weather forecasts to become even more accurate in future, MeteoSwiss is developing the COSMO numerical forecasting model on an ongoing basis in collaboration with international partners. The national weather services of Germany, Greece, Italy, Poland, Romania, Russia and Switzerland are working together closely within the framework of the Consortium for Small-Scale Modelling. This consortium was founded in October 1998 with the objective of maintaining and improving a non-hydrostatic, regional atmospheric model. This model is used for operational and research purposes.


MeteoSwiss works in a consortium for small-scale modelling (COSMO) together with other national meteorological services in Europe.

Further information