Ocean forecasts are done in a similar manner as weather forecasts are done. The variables of greatest interest are
- significant wave height
- sea level
- ice extent (ice concentration, thickness, and ridges)
- sea-surface temperature
The first variable, significant wave height, is calculated at SMHI using a wave model called SWAN (Simulating Waves Nearshore). All other variables are calculated using the coupled ice-ocean model HIROMB (High-Resolution Operational Model for the Baltic).
In general, to be able to make forecasts, three things are needed:
- a good initial condition, i.e. a description of the state of the ocean right now,
- good boundary conditions, i.e. a description of what happens at open boundaries such as the sea surface or the bottom,
- a method to, by using the laws of physics, calculate how the state (of the ocean in our case) develops over time.
The first item above, about the initial condition, is fulfilled by starting with a first guess of the ocean state, which is typically a six-hour forecast valid at the starting time of the forecast. This first guess is then modified slightly using observations of e.g. sea-surface temperature, valid near the starting time of the forecast. The merging of the first guess and the observations is done in an optimal way to minimize errors, using statistics of forecast errors as well as observation errors. This is usually referred to as "data assimilation", and is a science of its own.
The reason why data assimilation is needed to make forecasts is that numerical forecast models contain errors which grow over time. This makes the ocean state drift away from the true state which sooner or later affects the forecast quality in a negative way. Assimilation of observations is usually done before each forecast, i.e. with an interval of 6-24 hours depending on the type of forecasts.
The second item can be satisfied by using information from other models, such as an atmosphere model with information about wind, ait temperature etc., and from a large-scale ocean model with information about e.g. sea levels and salinity at a possible open boundary.
To satisfy the third demand above, numerical circulation models are usually employed to make forecasts from 48 hours up to 4 weeks ahead. To make shorter, more detailed forecasts, the atmospheric model HIRLAM (High-Resolution Limited-Area Model) is currently employed, which is being run and developed at SMHI. The longer ocean forecasts are forced using the correspondingly longer forecasts from ECMWF (European Centre for Medium-Range Weather Forecasts). It is an international organization in which Sweden is a member through SMHI.
Figure 1 shows an example of a model domain used operationally at SMHI. This setup has a horizontal resolution of 3 nautical miles (about 5.5 km) and has open boundaries in the western English Channel and along a section between Scotland and Norway.
Figure 2 shows two examples of sea level forecasts using HIROMB. The left picture shows an example from the Swedish west coast, which has a peak-to-peak amplitude of about 20-30 cm. The right picture shows an example from the northern Baltic Sea, where there are no tides.
Figure 3 shows an example of an ice forecast from the Bothnian Bay in the northern Baltic Sea. The colours indicate sea surface temperature and ice thickness.