Can We Predict Subsurface Mass Transport?

Predictions of contaminant transport produced by computer models are now widely used to evaluate health risks, determine remediation strategies, and guide environmental regulatory decisions. But model prediction errors are not generally well-known or quantified. In this study, the results of an extensive tracer test at the Columbus Air Force Base (CAFB), MS, test site are used to quantify error in computer model predictions of groundwater contaminant movement in the subsurface. Our modeling assumes no knowledge of the observed solute transport, uses widely applied modeling methods to generate computer predictions, and uses permeability and water levels as input data. Predicted and observed mass transport differ significantly, despite the relative abundance of aquifer data. It is not possible to accurately predict mass transport at CAFB using our approach because permeability cannot be characterized at the necessary resolution. Prediction errors at Superfund sites and other contaminated groundwater sites may be even higher because data is less abundant. This suggests that widespread use of predictive mass transport models that depend strictly on measured permeability and head values to predict transport should receive greater scrutiny and that methods of model parametrization based on early transport behavior should be more widely applied.