Automatic model calibration

Models used for prognostic purposes should, as far as possible, be physically based, as empirical modelslose their validity if boundary conditions are changed too much. Yet, a physical description of hydrological processes may be possible at the scale of single pores, but not in models for entire catchments. In the lattercase, in general small-scale physics developed for homogeneous systems is used under the assumption that the corresponding equations can be applied at larger spatial scales with the same parameters. An example is Darcy’s law describing theflow in porous media. Originally, it was developed from experiments with small soilcores. In numerical groundwater models, however, it is applied at scales of up to kilometres. At these scales,parameters like the hydraulic conductivity are no longer measurable and therefore are considerably uncertain.This pertains to a wide range of models, not only to numerical groundwater models such as MODFLOW(McDonald and Harbaugh, 1989), but also to analytical models that simulate processes at the catchment scalesuch as SHE (Abbottet al., 1986) or SWAT (Arnoldet al., 1998).