Comparison of different conceptual models for uncertainty propagation on contaminant transport prediction

Abstract:

As one of the most studied heterogeneous aquifer, the MADE site offers many high quality hydrogeological datasets to test subsurface heterogeneity and contaminant transport modeling approaches. By contrast, hydrogeologists are commonly confronted to data paucity when modeling hydrogeological properties to predict contaminant transport. A way to circumvent data scarcity is to use analog data and geological concepts in the modeling process. How are the predictions of contaminant transport and uncertainty affected by these choices?

In this presentation, using the MADE-II Tritium tracer experiment as a control reference dataset, we assume a realistic scarce data situation in a sandy gravel, braided-river aquifer and investigate how the use of analog data and different conceptual models — Multi-Gaussian, Object-based, Pseudo-Genetic — impact the plume behavior predictions. The comparison of plume characteristics between the simulations and the reference show that using analog data allows reasonable contaminant transport predictions to be made. The results also underline the key role of the contaminant initial location and spreading regarding the scale of subsurface heterogeneities on the plume behavior. In addition, the numerical results emphasize the importance of considering different conceptual models and a wide parameter space in order not to underestimate the prediction uncertainty.