HR: 0800h
AN: H21E-1075    [Abstracts]
TI: Three Model-Independent Algorithms for Optimal Selection of Monitoring Wells
AU: * Shlomi, S
EM: shaharsh@umich.edu
AF: Civil and Environmental Engineering, Technion - Israel Institute of Technology, Haifa, 32000 Israel
AU: Ostfeld, A
AF: Civil and Environmental Engineering, Technion - Israel Institute of Technology, Haifa, 32000 Israel
AU: Rubin, H
AF: Civil and Environmental Engineering, Technion - Israel Institute of Technology, Haifa, 32000 Israel
AU: Shoemaker, C A
AF: Civil and Environmental Engineering, Cornell University, Ithaca, NY 14853 United States
AB: Contamination of groundwater is quantified by sampling the aquifer at a number of wells and mathematically interpolating or physically modeling discrete values into maps of contaminant plumes. The apparent shape of the contaminant plume is extremely sensitive to the choice of wells. The objective of this work is to develop an algorithm which would choose the optimal wells for sampling. The optimization algorithms developed as part of this work can receive input (i.e. contaminant concentrations, pumping discharges, and concentration uncertainty estimates) from a variety of external models and recommend where to sample, i.e. which input the model should demand from the real world. The well utility function (WUF) simply weighs the input according to user-defined weights: given n available samples, the n wells with the highest well utilities are chosen for sampling. The cell utility function (CUF) ranks rectangular area cells by summing up the individual WUFs included in each cell. Consequently, each cell is assigned a number of samples according to its rank. Inside each cell it is much easier to allocate samples, mainly because there is a much smaller pool of wells to choose from. The major contribution of this work was the Utility Density Function. The UDF was defined as a continuous function, whose integral over an area returned that area's utility for sampling. A genetic algorithm found the best partition of the aquifer, in which the variance of the utility functions of the different polygonal cells was minimal. In the next step, one well was chosen for sampling in each cell. This ensured that no cell was allotted redundant samples (as all had the same utility). The three algorithms (WUF, CUF, UDF) were tested on several environmental settings, and in general the performance of each algorithm surpassed its precedents; i.e. sampling at the recommended sites resulted in better model predictions. This was also seen in sensitivity analyses, in which different parameters were taken to extremes.
DE: 1831 Groundwater quality
DE: 1832 Groundwater transport
DE: 1884 Water supply
SC: Hydrology [H]
MN: 2004 AGU Fall Meeting