Untangling Diffusion from Advection in Unsaturated Porous Media: Experimental Data, Modeling, and Parameter Uncertainty

We conducted a series of experimental and modeling tests using data from the Busted Butte Unsaturated Zone Transport Test. First, we conducted a suite of reactive (e.g., Li), nonreactive (Br), and colloidal tracer experiments. These tracers were injected for 190 d from two point sources at rates of 1 and 8 mL/h, respectively. We then used a numerical simulator (FEHM), populated with laboratory‐measured hydrologic properties, to verify that our conceptual model of the tracer test yielded a good fit to the tracer breakthrough data. Additionally, we used the AMALGAM‐SO and SCEM‐UA search algorithms to find optimal parameter estimates in our conceptual model and estimate their (nonlinear) uncertainty. To this end, the FEHM model was executed more than 50,000 times using parallel computing on a distributed computer cluster. The experimental and modeling results show that (i) no breakthrough of colloids was observed, low breakthroughs of Li were found, and significant and rapid breakthrough of Br was measured, (ii) measured hydraulic parameters from rock core samples provide a relatively accurate description of flow and transport at the scale and flow rates of the Busted Butte test, and (iii) the Millington–Quirk model of diffusion as a function of volumetric water content can fit the experimental breakthrough data well; however, (iv) a constant diffusion model with a much lower effective diffusion coefficient also fits the data well, and (v) numerous different optimized parameter combinations exist that fit the observed Br data acceptably well. This implies that one should be particularly careful in assigning values of the unsaturated subsurface flow and transport parameters without recourse to examining both parameter and model formulation uncertainty.