Abstract: A state-of-the-art physical groundwater flow and transport model was developed. The physical model consists of a large tank (approximately 10' x 14' x 8') filled with six feet of layered sand and silt media, and a fine sand lens within one layer. A dense, three-dimensional sampling matrix consisting of 105 time domain reflectometry (TDR) probes, pressure transducers, and thermocouples was multiplexed and connected to a data acquisition system. In addition, 105 sampling tubes and 84 screened-wells for pumping were located within the tank. One of the primary objectives of this research was to characterize flow and transport within the tank, determine characteristics of media packing, and system performance. K-field determinations from slug and tracer tests, including kriged K-fields from tracer tests, were performed. A nonpoint-source tracer test was used to calibrate a physics-based flow and a transport model (MODFLOW 2000 and MT3DMS). The calibrated model can now be applied to other studies at UVM or in collaboration with interested researchers.
Abstract: Accurate detection of a contaminant source is vital to the management of groundwater restoration projects. We developed a technique for identifying the contaminant source areas that combines site data, geostatistics and artificial neural networks (ANNs). The methodology involves replicating the kriging methods with ANNs to provide estimates of spatially dependent concentration fields, and for quantifying the uncertainty associated with the estimates. Once trained the ANNs approximate the results of ordinary kriging in a more computationally efficient manner. This provides a bi-directional prediction (i.e. forecast the future plume, as well as, the original source), which is impossible for many physics-based models.
Abstract: Determining the location of the contaminant source is important for improving remediation and site management decisions at many contaminated groundwater sites. At large sites, numerical flow and transport models have been developed that use historical measurement data for calibration. A well-calibrated model is useful for predicting plume migration and other management purposes; however, it is difficult to back out the source with these forward flow and transport models. We present a novel technique utilizing Artificial Neural Networks (ANNs) to backtrack source location and earlier plume concentrations from recent plume information. For proof-of-concept, two tracer tests (a non-point-source and a point-source) were performed in a large-scale (10'×14'×6') groundwater physical model. The physics-based flow and transport model (MODFLOW 2000 and MT3DMS) was calibrated using the data from the non-point-source tracer test and validated using the point source tracer test data. ANNs (e.g. counterpropagation) were trained using the calibrated model predictions and compared to actual data. Results show this to be a promising method for determining earlier plume and source locations.