Among the copious challenges inversion of large scale ill-posed problems introduce lies quantification of uncertainty. Here the inherent uncertainty of the problem is tackled in terms of prior sampling and our focus is in attempting to realize how distinct are the model samples considering their dynamic fingerprint. More specifically, we consider the history matching inversion problem in the context of flow in porous medium. Production data is collected from a sparse set of wells, and the subsurface parameters are inferred through computationally intensive inversion. Despite conscientious efforts to minimize the variability of the solution space, the distribution of the posterior often remains rather intrusive. In particular, geo-statisticians often propose large sets of model prior samples that regardless of their apparent geological distinction are almost entirely flow equivalent. As an antidote, in this study a reduced space hierarchical clustering of flow-relevant indicators is proposed for aggregation of these samples. Harnessing the dynamics as dictated by the governing physics and the controls allow identifying model characteristics that affect the dynamics. The effectiveness of the method is demonstrated both with synthetic and real field data.