Field-scale tracer experiment in an aquifer with highly variable hydraulic and retention properties: Can modelling predict the outcome with reasonable confidence?

Abstract:

One of the greatest challenges in groundwater hydrology is predictive modelling that is required in many applications yet notoriously difficult to verify. Hence relatively few systematic studies are available to assess predictive capability of modelling against real data. This is particularly true in cases where hydraulic variability is large (as in the case of the MADE site), and where retention (mass transfer) processes are also important and uncertain. We investigate the predictability of a tracer experiment in a fractured aquifer at Forsmark (Sweden) on a 46m scale with three simultaneous tracers: one non-sorbing and two sorbing. The available information used for predictive modelling is hydraulic data at 0.2m resolution from the injection and detection boreholes, the pumping rate, and a set of retention parameters (diffusivity, porosity and Kd) inferred in laboratory from site-specific rock samples. The retention data are strongly variable as are the transmissivity data. Our predictive modelling takes 4 steps: 1) Infer the hydrodynamic components of transport as mean and coefficient of variation of water travel time using the pumping and transmissivity data; 2) Generate an ensemble of all possible combinations of the retention parameters to form parameter groups that control mass transfer; 3) Compute expected breakthrough curves and uncertainty envelopes using information from step 2. The results show that the outcome of tracer field experiments could be predicted within a roughly 75% confidence interval for all three tracers. The non-sorbing tracer was predicted surprisingly close with the expected breakthrough curve, whereas the strongest sorbing tracer Cs was predicted on the 75% confidence limit due to its high sensitivity to Kd that is known to be very uncertain on the field scale. We also discuss model sensitivity outlining a methodology for assessing potential significance of non-Fickian hydrodynamic transport as well as of non-Fickian diffusive mass transfer. The results are relevant for further expanding the analysis of retention effects at the MADE site. !