ABSTRACT
Models that describe the migration of radionuclides in the ground can be computationally complex. This is why the estimation of global sensitivity measures is usually achieved after model reduction. In this work, we propose to utilize recent results that allow the direct estimation of global sensitivity measures from the given data produced through Monte Carlo simulation, without model reduction. In particular, we show that, from the given dataset, one can estimate density-based importance measures and distributional importance measures. We utilize sensitivity measures based on transformation invariant metrics, which allow to optimize computational performance in estimation. We also extract information about direction of change utilizing the estimation of the first order terms of the functional ANOVA expansion of the model output. Variance-based sensitivity measures are also extracted.
Results are compared using the model for migration of radionuclides in groundwater previously published in Ciriello et al. (2013). The combination of these sensitivity measures provides analysis with a wide range of insights concerning model behavior and key uncertainty drivers.
