Forecasts made with numerical circulation and wave models contain errors which become bigger with extending forecast periods. The reason is, amongst others, the discretisation errors of the solving methods and uncertainties of the status at the beginning of the simulations. Also the forcing data (like wind) and other phenomena (like turbulence) are not perfectly represented by the models.
To minimise the forecast error measurement data are assimilated by means of different methods using different techniques (Kalman filter, Optimal Interpolation, 3DVAR, 4DVAR, etc) in the model. The challenge is to keep the results dynamically consistent, so that still the physics is valid, e.g. conservation of mass and momentum, in the modified forecast. This is especially difficult in coastal areas, where small-scale and non-linear processes play an important role.
In our department we work on different assimilation methods. The techniques are much dependent on the considered parameters (measurements from tide gauges, current measurements with ADCP or HF radar, salinity and temperature, satellite data).