9526645 Yeh The reliability of groundwater modeling is often questionable, primarily because the problem of model calibration has not been completely resolved. In principle, the calibration of a distributed parameter model should include the following two aspects: the determination of model structure and the estimation of the unknown model parameters. To date, however, the determination of model structure has received limited attention because of its inherent difficulty. In practice, a complex aquifer is often represented by a simplified conceptual model based on limited geological information and subjective guess. As a result, model structure error is unavoidable. In most cases, model failure is caused by significant model structure error. Literature review indicates that there exist no systematic methods to estimate and to reduce such type of error. If this challenging problem can be solved or partly solved, groundwater modeling will be significantly improved and the reliability of model applications significantly increased. Major objectives of the proposed research include: the development of new concepts for model structure identification, the development of efficient algorithms for model structure error estimation, and the development of data collection strategy for model structure discrimination. The overriding objective is the development of a systematical methodology for model structure determination. The model structure error of using model A to replace model B is defined as a weighted sum of their distances in both the observation and prediction spaces. The calculation of model structure error is then formulated into a maximum-minimum problem. When the real structure of an aquifer system is unknown, a stepwise regression procedure will be used to estimate the model structure error and to determine an appropriate level of model structure complexity. A heuristic procedure that combines with the variational sensitivity analysis will be used to solve the proposed maxim um-minimum optimization problems. Field data collected at a major groundwater basin in Southern California will be used to validate the proposed methodology.