Comparison of Stochastic Calibration Methods for Modflow Using the Real-World Case ‘Wierden’

Abstract

A stochastic calibration method (SCM) (Gómez-Hernández, 1997) using stochastic inverse modelling techniques has been implemented for the widely used groundwater model MODFLOW. SCM combines geostatistical stochastic simulation and optimisation. It generates equally likely realisations of parameter fields conditioned on both parameter measurements and hydraulic head measurements. By this, parameters such as transmissivity, vertical resistance and porosity can be calibrated using this extended version of MODFLOW. It does not result in a single over-smooth parameter field, but an infinite set of equally likely realisations of the parameter, all of which are plausible representations of reality because they are conditioned on the available data and display the same spatial variability as observed from the field.

To compare the Monte Carlo (Stroet, 1995) method, MODFLOWP (Hill, 1992) and SCM, the real-world case ‘Wierden’-model was calibrated with each method respectively.

For the ‘Wierden’ model optimisation results of MODFLOWP hardly deviated from the earlier results using the Monte Carlo method. The optimisation process with MODFLOWP is much faster than using the Monte Carlo method. In MODFLOWP, however, some important MODFLOW-variables cannot be optimised. With the availability of MODFLOWP the Monte Carlo method for the use of calibrating MODFLOW models has been superseded.

Using the SCM parameterisation is not realised through zonation, but through generating a parameter field for each parameter. Provisional results of the re-optimisation of the ‘Wierden’ model using the SCM-extended version of MODFLOW show considerable improvements — compared to the zonation-based optimisation with MODFLOWP — in minimising the objective function and in randomising the distributing of the differences between calculated and measured groundwater heads. Using the SCM the rigid concept of zonation has become unnecessary, and it yields far more realistic parameter fields.