Simulating unsaturated flow fields based on saturation measurements

Abstract

Large amounts of de-icing chemicals are applied at the airport of Oslo, Norway. These chemicals pose a potential hazard to the groundwater because the airport is located on a delta deposit over an unconfined aquifer. Under normal flow conditions, most of the chemicals degrade in the vadose 7.one, but during periods of intensive infiltration, the residence time of contaminants in the unsaturated zone may be too short for sufficient degradation. To assess the potential for groundwater contamination and to design remedial actions, it is essential to quantify flow velocities in the vadose zone. The main purpose of this study is to evaluate theoretical possibilities of using measurements of liquid saturation in combination with inverse modeling for the estimation of unsaturated flow velocities. The main stratigraphic units and their geometry were identified from ground penetrating radar (GPR) measurements and borehole logs. These observations are included as a priori information in the inverse modeling. The liquid saturation measurements reveal the smaller-scale heterogeneities within each stratigraphic unit. The relatively low sensitivity of flow velocities to the observable saturation limits the direct inference of hydraulic parameters. However, even an approximate estimate of flow velocities is valuable as long as the estimate is qualified by an uncertainty measure. A method referred to as simulation by Empirical Orthogonal Functions (EOF) was adapted for uncertainty propagation analyses. The EOF method is conditional in the sense that statistical moments are reproduced independent of second-order stationarity. This implies that unlikely parameter combinations are discarded from the uncertainty propagation analysis. Simple forward simulations performed with the most likely parameter set are qualitatively consistent with the apparent fast flow of contaminants from an accidental spill. A field tracer test performed close to the airport will be used as an independent dataset to confirm the inverse modeling results.