Upscaling transport with mass transfer models: Mean behavior and propagation of uncertainty

The choice of an adequate large-scale conceptual transport model constitutes a major challenge associated with the upscaling of solute transport. Among the different alternatives to the classical advection-dispersion model, the (multirate) mass transfer model has been proposed as a valuable and convenient alternative to model the large-scale behavior of solute transport. This paper evaluates the use of mass transfer models as a constitutive equation for upscaling solute transport. To achieve this, we compare Monte Carlo simulations of solute transport at two different support scales. Transport simulations performed at the smallest scale represent a set of reference transport solutions described at a high resolution, which are contrasted against transport simulations obtained using an upscaled model (low resolution). Several formulations of the multirate mass transfer model, which differ in the type of memory function (single rate, double rate, and truncated power law), are used as a constitutive transport equation. The large-scale scenario represents a simplified model obtained by partially homogenizing the reference solution. Results show that the double-rate and the truncated power law mass transfer models are capable of properly describing the ensemble average behavior of the main features associated with the integrated breakthrough curves. However, the uncertainty associated with the upscaled mass transfer models was substantially smaller than that attributed to the reference solution. Importantly, the cumulative distribution function of concentrations associated with the upscaled model follows a distribution similar to the reference solution but with smaller statistical dispersion. The reason is that while appropriate memory functions can be used to preserve the residence time distribution of mass particles during upscaling, the lack of memory in space prevents the model from reproducing mass fluxes in all directions. Specifically, the reproduction of mass fluxes taking place at the interface between two homogenized blocks of the upscaled model is not satisfied, thus providing a poor description of the spatial distribution of mass particles in a given realization.