Effective Parameter Optimization for Ground‐Water Model Calibration

Publisher Website
Google Scholar
Abstract
This paper shows that the Levenberg‐Marquardt method, modified to limit the parameter update steps to within a specified range, effectively optimizes a set of parameters in a ground‐water model. In doing so it performs better than more complicated conjugate gradient algorithms.The method has been applied to the synthetic ground‐water problem of Carrera and Neuman (1986b), which has been used previously to test calibration method performance. The descent of the error criterion and the simultaneous approach of the parameters to their optimum values are smooth, both with perfect and erroneous head observation data.It is shown that least‐squares optimization can be done effectively in a simple manner and that it can be easily integrated with practical ground‐water modeling. No changes have to be made to the model code. This ease of implementation is emphasized by the fact that all modeling presented here was done using an ordinary spreadsheet program on a PC.