Use of Simulation Filters in Three-Dimensional Groundwater Contaminant Transport Modeling

Abstract

Contaminants have been modeled using numerical models to predict their fate and transport. These models are simplified by introducing approximation that plagues the model with truncation and round-off errors. In order to improve the accuracy and effectiveness of contaminant concentration prediction, three simulation techniques such as particle, Kalman and extended Kalman filters were used in this research as a tool to model the groundwater contaminant using a three-dimensional (3D) subsurface advection-dispersion-reaction model. The 3D model was discretized spatially and temporally using the forward-time-central-space (FTCS) method. A total of 18 observation points were used to run the simulation filters. The dynamical models used by the simulation filters are embedded with random Gaussian noise to mimic a real-life situation. The Kalman and extended Kalman filters also have an advantage of storing only the previous estimate to make a new prediction. The particle filter however, applies sequential Monte Carlo method and probability density functions to make estimates. The filters are capable of providing a better prediction than the numerical method when sparse observation data are used. The algorithms to generate the simulation and the numerical results were run on Matlab 7.1. The effectiveness of the prediction results were assessed using the root mean square error (RMSE), mean absolute error (MAE) and maximum absolute error (