Heterogeneity and Prior Uncertainty Investigation Using a Joint Heat and Solute Tracer Experiment in Alluvial Sediments

In heterogeneous aquifers, imaging preferential flow paths, and non-Gaussian effects is critical to reduce uncertainties in transport predictions. Common deterministic approaches relying on a single model for transport prediction show limitations in capturing these processes and tend to smooth parameter distributions. Monte-Carlo simulations give one possible way to explore the uncertainty range of parameter value distributions needed for realistic predictions. Joint heat and solute tracer tests provide an innovative option for transport characterization using complementary tracer behaviors. Heat tracing adds the effect of heat advection-conduction to solute advection-dispersion. In this contribution, a joint interpretation of heat and solute tracer data sets is proposed for the alluvial aquifer of the Meuse River at the Hermalle-sous-Argenteau test site (Belgium). First, a density-viscosity dependent flow-transport model is developed and induce, due to the water viscosity changes, up to 25% change in simulated heat tracer peak times. Second, stochastic simulations with hydraulic conductivity (K) random fields are used for a global sensitivity analysis. The latter highlights the influence of spatial parameter uncertainty on the resulting breakthrough curves, stressing the need for a more realistic uncertainty quantification. This global sensitivity analysis in conjunction with principal component analysis assists to investigate the link between the prior distribution of parameters and the complexity of the measured data set. It allows to detect approximations done by using classical inversion approaches and the need to consider realistic K-distributions. Furthermore, heat tracer transport is shown as significantly less sensitive to porosity compared to solute transport. Most proposed models are, nevertheless, not able to simultaneously simulate the complementary heat-solute tracers. Therefore, constraining the model using different observed tracer behaviors necessarily comes with the requirement to use more-advanced parameterization and more realistic spatial distribution of hydrogeological parameters. The added value of data from both tracer signals is highlighted, and their complementary behavior in conjunction with advanced model/prediction approaches shows a strong uncertainty reduction potential