The development of numerical models for weather forecasts focuses in three areas:
use of new observation types,
new data assimilation methods, and
improved model physics.
Furthermore, the researchers create methods for post-processing of weather forecast data that are specifically designed for applications such as renewable energy, road weather, air traffic, water resource management, etc.
The unit has been successful with the development and production of European reanalysis, ie. a consistent historical archive of European weather.
A lot of the work is done in cooperation with European research consortia, i.e. HIRLAM/ALADIN and cooperation with other national meteorological institutes in Nordic and European countries.
More detailed forecasts, a challenge
When the resolution of the numerical models increases, we must face some new problems. With the present resolution of 2.5 km we are starting to resolve atmospheric convection in our model. This means that different formulations for the dynamics and physics are needed than for larger weather phenomena.
The flow on these scales requires a non-hydrostatic formulation which means that it is more unbalanced, and the forcing from the physical processes is more dominant than on larger scales. The parameterization of e.g. convection is questionable, since this process is partly resolved.
This implies that even these small scales have to be specified in the initial state. It is also necessary with increased computer power, and effective coding of the model.
Very short forecasts, nowcasting
Many decisions in the society are strongly dependent on the weather evolution, on a very short timescale. Examples are activities at airports and road maintenance, where the largest interest is focused on the first few hours, or even less.
To make short detailed forecasts is known as nowcasting. We are working with models for nowcasting, e.g. to improve fhe local forecast of the weather at an airport and estimations of road temperature and ice/water on the roads.
Another development area is a model to forecast precipitation, on a short timescale, by combining model data with radar information.
Since the atmosphere is chaotic, it is not feasible to regard the weather evolution as beeing deterministic.This means that the forecast can behave very differently, if started from slightly different (but equally probable) initial conditions, or if the physical formulation in the model, like a parameter, is slightly changed (also within reasonable values).
By running many forecasts in parallel, the so called ensemble technique, it is possible to say something about the probability distribution of different weather developments.
We also use statistical interpretation methods, like Kalman-filtering, to link desired prognostic information to the output of the numerical model system.