Wind power forecasting - Wind Power Integration: Connection and System Operational Aspects - page 221

Telecommunication System (GTS), conversion of the accepted data to a format
required by the analysis scheme, processing of boundary values at the lateral
boundaries of the model domain and statistical procedures employed to make cor-
rections to first guess forecasts (usually forecasts from the previous analysis time), so
that the differences between the corrected first guess and the accepted observations
at the analysis time are minimised. All of these improvements are computer intensive
and will be facilitated by ever increasing availability of computer power.
One of the basic principles underlying the numerical modelling of the atmo-
sphere is that the atmosphere as a continuous fluid can be represented by the values
of the seven meteorological variables at a finite set of discrete points. These points
are arranged in a three-dimensional grid with fixed spatial coordinates. The spacing
of the grid points is also known as the spatial resolution or the grid length. Different
spatial resolution is generally used in the vertical compared to the horizontal, due to
the stratified structure of the atmosphere in the vertical. Increasing the spatial
resolution increases the accuracy of the atmospheric model, but at the cost of
additional computational effort. Thus, if the grid size is halved, the computational
effort increases by a factor of 2 4 ¼ 16 as the number of grid points is doubled in

each space direction and the number of time steps must also be doubled.
The choice of vertical and horizontal spatial resolutions is clearly very
important. In the atmosphere there are motions with spatial scales ranging from
many thousands of kilometres down to a millimetre. In order to capture all atmo-
spheric processes, the atmospheric model should ideally be constructed with spatial
resolutions down to a millimetre. Clearly this would impose impossibly high bur-
dens on computer resources. On the other hand, a particular spatial resolution will
exclude from consideration any physical atmospheric process whose spatial scale is
smaller than the spatial resolution. The choice of spatial resolution is therefore a
trade-off between an acceptable level of accuracy and available computer power.
For a particular grid resolution there will always be sub-grid scale physical
processes that the model cannot resolve. In order to be realistic, the model must
take account of the net effect of these processes. This is done by parameterisation
of these effects, based on statistical or empirical representations of the sub-grid size
physics. These parameterisations will also contribute to the computational effort.
NWP models can be global or limited area models. The limited area model is
nested within the global model. The high-resolution limited area model (HIRLAM)
is used widely in Europe and is a result of cooperation between the meteorological
institutes of Denmark, Finland, Iceland, Ireland, Netherlands, Norway, Spain,
Sweden and France. HIRLAM receives its lateral boundary conditions from the
European Centre for Medium Range Weather Forecasting (ECMWF) global
atmospheric model every six hours. The Danish Meteorological Institute (DMI)
HIRLAM domains are shown in Figure 6.2. The inner domains (around Denmark
and around Greenland) are of higher resolution, with 5.5 km grid length, and use
lateral boundary conditions from the outer, lower resolution domain that has a
15 km grid length. All domains have 40 layers in the vertical. The time step for the
outer domain is 300 seconds while for the inner domains it is 90 seconds. The
physical parameterisations are applied every third time step.