Identifying sources of groundwater pollution: An optimization approach

Abstract

Least squares regression and linear programing for least absolute error estimation are each combined with groundwater solute transport simulation to identify the locations and magnitudes of aquifer pollutant sources. Pollutant sources are identified by matching simulated and measured nonreacting solute concentration data. We have assumed known hydraulic parameters but considered concentration data errors explicitly. The identification models are demonstrated and compared using two hypothetical aquifer systems, one for the steady state case and the other for the transient case. Steady state models identified unknown pipe leak locations and leak magnitudes based upon sparse and spatially distributed chloride and tritium data. The number of likely leak locations was restricted in the models by employing mixed integer programing and stepwise multiple regression. Transient models identified several annual disposal fluxes in the aquifer based upon concentration histories collected at observation wells. In this case, conservative solute concentration data were abundant and contained substantial errors. Minimizing either least absolute or least squared errors was successful in identifying pollutant sources. Furthermore, we demonstrate error analysis for the results given by either method.