Combining Geostatistical Analysis and Flow-and-Transport Models to Improve Groundwater Contaminant Plume Estimation.

Abstract

Groundwater is an important resource, which is often contaminated. In order to ensure a sustainable supply, groundwater has to be monitored, contaminant plumes must be estimated accurately, and remediation operations must be carried out effectively. However, groundwater monitoring networks often do not have enough monitoring wells, and those wells are not always optimally located for the purpose of plume estimation, using existing methods. Moreover, budgetary constraints limit the number of available samples. Existing methods for plume estimation rely either on the spatial correlation of plume concentrations, or on the underlying physics of groundwater flow and contaminant transport. Often, practitioners who rely on one of these approaches neglect available information which can be used in methods belonging to the other approach. For example, use of kriging, a geostatistical method relying on spatial correlation, often precludes the use of transport information, which may be readily available. Conversely, flow-and-transport models do not explicitly consider spatial correlation of contaminant concentrations. In this work, these two approaches are combined, in order to optimally use all available information, to improve the quality of plume estimation. Specifically, two geostatistical methods – Inverse/Forward Modeling and Transport-Enhanced Kriging – are developed that combine transport models with spatial or temporal correlation. These methods are versatile, can apply to a variety of situations, and can work with many kinds of available input data and transport models. A method is also developed to estimate flow and transport parameters simultaneously with the plume concentration, for cases in which this information is unknown or uncertain. Finally, as monitoring network configuration has a dramatic effect on estimation results (but is specific to the plume estimation method used), a method for choosing optimal monitoring sites is presented. All of the methods were tested in a variety of numerical experiments with synthetic homogeneous and heterogeneous data. In addition, several laboratory experiments were performed in a large sand tank, to assess the performance of the methods. Overall, the new methods yield results that are superior to those obtained by common existing methods such as kriging, with a better reproduction of the true plume shapes and lower uncertainty.