The observation data is used to assess the quality of the current parameter set p. After a direct calculation run, their evaluation usually provides the main part of the current error f(p). The following observation data is currently possible:

• Potential heads at individual nodes of the model (data type POTE)

• Leakage rates (e.g. from discharge measurements) as a sum for a specific section of a receiving water that is integrated into the model via the data type LERA (data type LKNO, inverse modelling only),

• Mass flow rates as a sum for a certain number of nodes at which fixed potential heads are prescribed in the model (data type KNOT).

All data can be available at different times for transient calculations.

Which observed variables are used is independent of the choice of model parameters to be optimised and the availability of measurement data. For example, K values can be optimised using observed potential heads and leakage rates without also optimising leakage coefficients. Conversely, leakage rates do not necessarily have to be available as observed data for the optimisation of leakage coefficients.

The number of available observation data has a significant influence on the success of inverse modelling. The more observation data is available, the greater the chance of finding an "optimal" parameter set. However, successful optimisation can only be achieved if the zoning of the parameters to be optimised is also meaningful. Defining a sensible zoning is the main task of the processor in inverse modelling.

The observation data must be defined in a separate file for an optimisation run. This file is referred to below as the observation file. The structure of the file is described in the chapter "Realization in SPRING".

Zoning of the parameters