ABSTRACT:Asurrrogate‐prarmeterapproach to modeling groundwater basins is presented, which has the following advantages over current simulation‐type methods: (i) conducivness to modeling nonhomogeneous and nonisotropic basins; (ii) there is no need to guess boundary conditions if accurate information is not available; (iii) the model is amenable to systematic calibration or identification through the use of optimization techniques; and (iv) compatibility with systematic algorithms for analyzing a wide range of management strategies. Since the parameter identification problem is large‐scale and nonconvex, it is decomposed through application of generalized duality theory, into several subproblems of smaller size which are solved independently a number of times in order to achieve an overall solution. Results are presented for a hypothetical system of four interacting