By doing nowcasting we mean to produce very short forecasts, i.e. up to about 6-9 hours. Due to data processing time our numerical 3-dimensional models are roughly 6 hours old, at observation time, which means that our best +6h forecasts are based on a +12h model run. The challenge in nowcasting is to utilize our latest observations, together with models, to improve these forecasts. We have chosen to use the hourly produced mesoscale analyses, Mesan, as one of the basis components in nowcasting.

One of the problems with 3-dimensional models is to force recent information (observations) into the model. Due to geostrophic adjustment, and physical spin up problems it is likely that much of the new information, will show up as spurious noise.

We are here working along two different approaches to the nowcasting problem:

• 1-dimensional modelling, using 3-dimensional forcing.
• 3-dimensional modelling, using nudging and digital filters.

1-dimensional modelling
A time step in the 3-dimensional model (i.e. Hirlam) could be regarded as a sum of two components, a dynamical tendency (adiabatic tendency) and a tendency due to physical processes (radiation, turbulence, condensation etc.)

We have developed a tool, where the 3-dimensional coupling has been eliminated

The idea is as follows:

• Compute the dynamical tendencies from a 3-dimensional forecast and store them for the place of interest.
• Modify the vertical profile at these places, using late observations (Mesan)
• Run a 1-dimensional (vertical) model at each place and force it with the precomputed dynamical tendencies.

This model can be run, both at specific points or on a grid. Here follows an example of a time series output from this model at the airport of Luleå.

Reference
Gollvik, S., Olsson, E., 1995: A one-dimensional interpretation model for detailed short range forecasting. Meteorol. Appl. 2, pp. 209-216

3-dimensional modelling
Here the idea is to force the recent information into the 3-dimensional model.

The basic idea is:

• Rerun the 3-dimensional model from -6h to +3h, and include a nudging term (between -3h to +3h) where the model state is relaxed towards a state estimated from observations (at present only the humidity field).
• At the same time perform a digital filtering of this run, producing a smooth initial state at observation time.

Here follows an example of a Hirlam forecast at 980914 00UTC, +06, +09, +12, shown as artificial satellite pictures (first row), together with our nowcasting products, Hirmes (second row) and corresponding analyses Mesan (third row)