Author(s): Abhishek Singh; Barbara S. Minsker; Albert Valocchi
Linked Author(s):
Keywords: Calibration; Optimization; Interactive; Multi-objective; Genetic algorithms; Machine learning; Inverse modeling
Abstract: Hydrologic models can have numerous spatio-temporal parameters. Fully automated calibration techniques that do not use the expert’s understanding can lead to parameter fields that are not consistent with knowledge about the modeled system. Many studies have shown that incorporating expert knowledge in the calibration process can lead to significant improvements in the parameter estimation process. The Interactive Multi-Objective Genetic Algorithm (IMOGA) is an interactive and adaptive framework that integrates ‘subjective’ field-knowledge of the experts with quantitative calibration objectives to find more stable and reasonable parameters for models. This paper describes the interactive multi-objective approach, addresses the issue of user fatigue in such a system, and presents results for a groundwater case study. The IMOGA considers model calibration as a multi-objective optimization process using user-preference as an additional objective along with quantitative calibration measures such as model prediction error and regularization. Given these qualitative and quantitative criteria, the IMOGA converges to a set of Pareto optimal solutions representing the best trade-off among all objectives. To make the IMOGA more feasible in practice, a two-step methodology is proposed to reduce the number of user interactions required to get reliable solutions. Finally, the methodology is applied to the Freyberg (1988) groundwater calibration case study and improvements are shown both in terms of plausibility of parameters and model prediction accuracy.
Year: 2007