A hybrid Marquardt-Simulated Annealing method for solving the groundwater inverse problem

One of the limitations to the use of inverse problem algorithms is the relatively high frequency of convergence problems. Continuum optimization methods, such as Marquardt, are quite efficient, but not very robust. That is, they normally converge in a small number of iterations but often fail to converge. On the other hand, discrete optimization methods, such as Simulated Annealing, are very robust, but unacceptably expensive. We present a hybrid method aimed at coupling the efficiency of Marquardt's method and the robustness of Simulated Annealing. We explore its possibilities and adjust some of its parameters by applying it to a large number of cases.