Dual-domain mixing cell modelling and uncertainty analysis for unsaturated bromide and chloride transport

Abstract

Land use intensification is considered the main reason for early signs of deterioration in the water quality of Lake Taupo, New Zealand. Little is understood, however about the origin, and governing flow paths of the contaminants and their respective transformation processes that affect the water quality of Lake Taupo. In this study we investigate contaminant transport and its small-scale variability in the volcanic vadose zone surrounding the Lake. Lateral and preferential solute transport is analysed to better understand the risks of diffuse groundwater pollution from contaminant sources at the land surface. As part of the investigations into this problem the Spydia experimental facility has been installed under a pastoral agriculture land use in the Lake Taupo region, New Zealand (Barkle et al. 2011). A multiple tracer experiment was conducted at the site and vadose zone drainage volumes were measured using Automated Equilibrium Tension Plate Lysimeters (Figure 1). The chemical composition of the drainage samples was analysed in the laboratory. A dual-domain mixing cell model was set up to simulate the unsaturated advective-dispersive tracer transport at selected monitoring sites for two different bromide-chloride (Br⁻, Cl⁻) tracers that were applied at the land surface at two different regions (Figure 1). Some model parameters were constrained by mixing calculations of the measured total Br⁻ and Cl⁻ load, whereas others were calibrated using the measured Br⁻ and Cl⁻ breakthrough curves and drainage volumes. Multi-objective inverse modelling using the AMALGAM evolutionary search method (Vrugt & Robinson, 2007) showed a significant trade-off between simulated transient Br⁻ and Cl⁻ breakthrough curves and corresponding drainage volumes, but also a compromise solution that fits both objective functions reasonably well. Estimates of parameter and model predictive uncertainty were subsequently derived using the differential evolution adaptive metropolis, DREAMZS adaptive Markov chain Monte Carlo algorithm (Vrugt et al., 2011) with a formal Bayesian likelihood function (Wöhling & Vrugt, 2011). Uncertainty bounds derived by this MCMC method simultaneously capture the observed Br⁻ and Cl⁻ breakthrough curves and corresponding drainage volumes. Our results demonstrate that (1) flow and transport in the vadose zone is highly variable, and (2) contaminants at the land surface can travel rapidly through the soil to larger depths and this cannot be described with the classical advection-dispersion equation.