Uncertainty Quantification of Tracer Dispersion with the PMVP Model under Realistic Conditions

Abstract:

The polar Markovian velocity process (PVMP) model provides a computationally efficient method to propagate input uncertainty stemming from unknown permeability fields to output flow and transport statistics [Meyer and Tchelepi, WRR, 2010; Meyer, Jenny, and Tchelepi, WRR, 2010; Meyer et al., WRR, 2013]. Compared with classical Monte Carlo (MC) sampling, the PMVP model provides predictions of tracer concentration statistics at computing times that are three orders of magnitude smaller. Consequently, the PMVP model is as well significantly faster than accelerated sampling techniques such as multi-level MC or polynomial chaos expansions. In this work, we further evaluate the PMVP model performance by applying the model for tracer dispersion predictions in a setup derived from the well-known MADE field experiment [Boggs et al., WRR, 1992]. We perform detailed model validations against reference MC simulations and conclude that the model provides overall accurate dispersion predictions under realistic conditions.