CALIBRATION OF TRANSMISSIVITY IN A CONFINED AQUIFER VIA NONLINEAR LEAST-SQUARES

The calibration of mathematical models for groundwater flow is general ly performed by minimizing an objective function that measures the fit between measured and computed data. This is a non-linear least square s problem which, because of the ill-posedness of groundwater identific ation, is computationally laborious. In this regard, minimization proc edures reported in the literature usually invoke only first order deri vative information about the objective function despite the fact that exact second derivative information can be computed at relatively smal l costs. A natural question is therefore whether or not exact second d erivative information could reduce the number of minimizing iterations and thus reduce the high computational cost of aquifer calibration. N umerical experiments\reported in this paper indicate that use of secon d derivative information may not systematically reduce the number of m inimizing iterations and so may well be ignored in practice.