Prediction uncertainty for tracer migration in random heterogeneities with multifractal character

Travel-time statistics for non-reacting tracers in fractal and multifractal media are addressed through numerical simulations. The logarithm of hydraulic conductivity is modeled using fractional Brownian motion (fBm) and more recently developed multifractal model based on bounded fractional Levy motion (bfLm). These models have been shown previously to accurately reproduce statistical properties of large conductivity datasets. The ensemble-mean travel time increases nearly linearly with travel distance and the variance in the travel time increases nearly parabolically with travel distance. This is consistent with near-®eld analytical approximations developed for non-fractal media and suggests that these analytical results may have some degree of robustness to non-ideal features in the random-®eld models. The magnitudes of the travel-time moments are dependent on the system size. For fBm media, this size dependence can be explained using an e€ective variance that increases with increasing size of the ̄ow system. However, the magnitudes of the travel-time moments are also sensitive to other non-ideal e€ects such as deviations from Gaussian behavior. This sensitivity illustrates the need for careful aquifer characterization and conditional numerical simulation in practical situations requiring accurate estimates of uncertainty in the plume position. Ó 1999 Elsevier Science Ltd. All rights reserved.