Interactive multi-objective inverse groundwater modeling -Formulation and addressing user fatigue

This paper builds on work done on using interactive multi-objective genetic algorithms (IMOGA) to solve the groundwater inverse problem (Singh & Minsker, 2005) by searching for optimal hydraulic conductivity fields conditioned on field measurements of hydraulic heads and conductivities. The biggest challenge faced when using such interactive systems is that of user fatigue because the user is expected to evaluate many solutions during the search process. This paper discusses a two-step approach to reduce user fatigue. First the user is shown only a fraction of the total population in every generation. To ensure minimum redundancy during evaluation, the solutions are clustered using unsupervised clustering and the expert is shown unique samples from distinct clusters. Next the unranked solutions are ranked using a surrogate model that 'learns' from the user preferences. This is implemented using a supervised classification algorithm to cluster the solutions based on the 2-D images of hydraulic conductivity. We test 'content-based' and 'spectral' algorithms for the clustering and classification as these have been shown to be similar to how humans process images. The work on applying and testing these algorithms is on-going and this paper discusses some preliminary results. Complete results will be shown at the EWRI conference.