Global sensitivity analysis and adaptive stochastic sampling of a subsurface-flow model using active subspaces

摘要: Abstract. Integrated hydrological modeling of domains with complex subsurface features requires many highly uncertain parameters. Performing a global uncertainty analysis using an ensemble model runs can help bring clarity as to which these parameters really influence system behavior and for high parameter does not result in similarly predictions. However, already creating sufficiently large simulation the sensitivity be challenging, combinations lead unrealistic behavior. In this work we use method active subspaces perform analysis. While building up ensemble, already-existing members construct low-order meta-models based on first two active-subspace dimensions. The are used pre-determine whether random combination stochastic sampling is likely so that such excluded without running computationally expensive full model. An important reason choosing both activity score easily understood visualized. We test approach subsurface-flow including hydraulic parameters, boundary conditions geological structure. show detailed exist most observations interest. pre-selection by meta-model significantly reduces number full-model must rejected due essential but difficult part models approximating gradient simulated observation respect all effectively meaningfully done second-order polynomials.