UVM Theses and Dissertations
Format:
Print
Author:
Li, Zhiqiang
Dept./Program:
Civil and Environmental Engineering
Year:
2007
Degree:
MS
Abstract:
We present a dynamic, real-time Extended Kalman Filter (EKF) used to update contaminant transport model predictions based upon available data. A novel technique that combines an Extended Kalman Filter with a flow and transport model has been developed to back out the location of the contaminant source using concentration data from a large-scale (10' X 14' X 6') physical tank experiment. A point-source tracer test, using ammonium chloride solution, was performed in the tank that is comprised of five sand and silt layers. The EKF was wrapped around the model (MODFLOW 2000 and MT3DMS 4.0), while MT3DMS was modified to compute the Jacobian matrix, state transfer, and covariance matrices for time evolution of the concentration data.
While we know the location of the injection well and the total volume injected during the fifteenday experiment, we cannot be sure of the three-dimensional configuration of the source around the injection well. A number of MODFLOW grid cells were used to represent our best guess of the initial source location; and the EKF dynamically adjusted the model predictions and initial conditions to make the model predictions more closely match the observed data when new observations were available. The results are promising. This is a Bayesian method that gives the best estimate using all the available data, which is more appropriate for the uncertainty of the source location often encountered in the real sites.
While we know the location of the injection well and the total volume injected during the fifteenday experiment, we cannot be sure of the three-dimensional configuration of the source around the injection well. A number of MODFLOW grid cells were used to represent our best guess of the initial source location; and the EKF dynamically adjusted the model predictions and initial conditions to make the model predictions more closely match the observed data when new observations were available. The results are promising. This is a Bayesian method that gives the best estimate using all the available data, which is more appropriate for the uncertainty of the source location often encountered in the real sites.