Solute Tracer Tomography: Field Implementation and Parameter Estimation using the Ensemble Kalman Filter

Abstract

The performance of groundwater flow and solute transport models depends, to a large extent, on the resolution at which aquifer heterogeneity is resolved. Large datasets are needed to estimate aquifer parameters at a high resolution, but their size is usually limited in real-world applications. The evolution of modern measurement techniques including better, smaller and affordable sensors have fostered the development of field tests with a tomographic layout, where the system is stressed in different directions. Recent numerical studies show the advantages of using large datasets of different type for describing small-scale features of aquifer properties. While hydraulic tomography, where multiple pumping tests are performed, has been repeatedly applied at field scale, applications of tracer tomography, where multiple tracer tests are performed, lag behind due to technical limitations. This study pursues to narrow the gap between numerical studies and field applications, showing the potential of hydraulic and tracer tomography for high-resolution aquifer characterization. In contrast to the few reported applications of tracer tomography, in which heat was used as a tracer, the experimental setup developed in this work was designed to use fluorescein as a conservative tracer, and was applied in the shallow alluvial aquifer at the Hydrogeological Research Site Lauswiesen, Germany. The experimental results demonstrate that solute-tracer tomography can be efficiently applied at the field scale, if a nested-cell forced-gradient flow field is generated prior to tracer injection. Field data were analyzed with the Ensemble Kalman Filter, coupled to a three dimensional groundwater flow and dual-domain transport model to estimate spatially distributed flow and solute transport parameters. The efficiency of the filter allows the description of aquifer heterogeneity at a high resolution while keeping reasonable computational costs. The filter was tested with a synthetic study based on a two dimensional model resembling the hydrogeological features and well facilities at the field site. The filter settings with the best performance were applied for the estimation of aquifer parameters based on real data. Results of the synthetic study show that parameters estimated with the Ensemble Kalman Filter applied to hydraulic data already contain the main features of the reference field. However, with the inclusion of concentration data the spatial structure of the parameter fields is accentuated, their uncertainty is considerably reduced, and flow and transport model predictions are improved. While the standard update scheme of the Ensemble Kalman Filter is applicable to hydraulic head, it leads to mass balance errors during assimilation of concentration data. Therefore, a restart scheme was applied where the steady-flow model is reinitialized after each parameter update, and transport is simulated from the initial time until the next available measurement time-step. The estimation of parameters based on drawdown curves measured during the field tracer tomography shows the potential of the Ensemble Kalman Filter in adjusting model parameters to improve groundwater flow simulations. When using the concentration data, the spatial structure of hydraulic conductivity was accentuated and the associated variance reduced. Transport simulations were largely affected by numerical dispersion, and by estimating effective transport parameters such as porosity and dispersivity as uniform values, rather than considering them as spatial fields. The Monte Carlo approach of the Ensemble Kalman Filter imposes limitations in the number of parameters that can be estimated, therefore efficient methods to optimize the grid resolution and reduce matrices dimensions are required.