SCREENING STOCHASTIC AQUIFER REALIZATIONS USING ADVECTIVE PARTICLE TRACKING METRICS

Contaminant movement in groundwater can be simulated using advective transport or a combination of advective and dispersive transport. The latter approach can be significantly more computationally intensive, particularly for large or complex models. The goal of this investigation is to assess the ability of advective transport modeling to rank order ensembles of stochastic aquifer realizations with respect to their advective-dispersive transport model behavior. This approach can provide a means for screening ensembles to identify individual realizations with extreme transport behavior without the computational expense of advective-dispersive modeling.

Stochastic ensembles for this investigation consist of Gaussian hydraulic conductivity (K) distributions ranging in σ² ln(K) from 0.29 to 1.0. Advective and advective-dispersive transport of a conservative tracer in an injection-extraction well pair were simulated using MODPATH and MT3DMS, respectively. MODPATH particle arrival times were transformed into synthetic breakthrough curves for comparison with MT3D output. Ensembles of 100 realizations were sorted using breakthrough curve characteristics including first arrival time, peak concentration, and first and second temporal moments. The Spearman rank correlation coefficient (R²) was used to compare ordered lists of advective and advective-dispersive breakthrough curve metrics.

Initial results suggest that advective particle tracking is a good surrogate for screening advective-dispersive transport behavior for some breakthrough curve characteristics but not for others. Individual realizations with minimum or maximum first arrival time, peak concentration, and second temporal moment values were successfully identified using advective transport results in some instances. Spearman R² values exceeded 0.79 for first arrival times, center of mass, and second temporal moments indicating a positive correlation between advective and advective-dispersive simulations at a 99.9% confidence level. In contrast, R² values were less than 0.31 for peak concentrations, peak arrival times, and first temporal moments. Sensitivities to the number of particles released and the MT3D solution method (central finite difference v. total variation diminishing) are also under investigation.