1.1 Background
Most of significant threats to groundwater of various contamination sources include industrial landfills, regulated hazardous waste sites, underground storage tanks, and abandoned waste sites. More recent reports indicate that the soil and groundwater once contaminated, the tillage and health of inhabitant nearby the site may be influenced. Thus, when the contaminated site is announced, the remediation is imperative to act; yet, remediation is a complex and lengthy process. The difficulties arise from the uncertainty of aquifer characteristics and the contaminant source location in the landfill. The uncertainty of aquifer characteristics affects the results of predicting the groundwater flow and contaminant transport. The uncertainty of the source location makes the contaminant zone difficult to delineate or assess; moreover, the liability of the contaminant site is not easy to determine if the responsible parties of the contaminant site are several. In addition, large degree of variation in the hydraulic conductivity field may make the shape of the plume irregular, and consequently the source location becomes more difficult to detect.
Thus, how to effectively increase the chance of source location identification is deserved further studies if the source location is unknown and the aquifer formation is
heterogeneous.
1.2 Literature Review
1.2.1 Monitoring Well Design
Among the methods of monitoring well network design in the literature, Meyer and Brill (1988) presented a method along with Monte Carlo simulation (MCS) for locating wells in a monitoring network to provide detection of contaminant plumes.
Their method utilized two linked models, a simulation and an optimization model.
The simulation model utilized an analytical solution to estimate the solute transport information. The optimization model offered the optimal placement of wells location. The Monte Carlo technique was used with the simulation model to reduce uncertainty effect of the contaminant concentration distribution. The result showed that the selected well network could detect the plume distribution effectively.
Bagtzoglou et al. (1992) proposed an approach using particle methods and MCS to estimate the source location and time history in a heterogeneous aquifer. They analyzed two heterogeneous aquifers with perfectly known conductivity fields and conditioned conductivity fields. Twenty MCS runs were used in each case with different observation numbers. The results indicate that when 60 wells are used for the designed case, source identification with conditional conductivity field performs
as well as the simulation with perfectly known conductivity field. Meyer et al. (1994) proposed a further study about monitoring network design to detect the groundwater contamination distribution. They considered the resolution of multiple conflicting objectives: minimizing the number of wells in the network, maximizing the probability of detection, and minimizing the expected area of contamination at the time of detection. The design method, an extension of Meyer and Brill (1988), can be used to predict the appropriate number and location of monitoring network.
Storck et al. (1997) presented an uncertainty analysis incorporating the Monte Carlo method and the particle-tracking model for the monitoring well network to detect the leak of the groundwater contamination. The monitoring network optimization method extended the approach of Meyer et al. (1994) and applied in three-dimensional heterogeneous aquifers. They considered three conflicting objectives and applied their method to analyze an existing landfill site. The result revealed that their method worked for detecting the groundwater contaminant leak.
Mahar and Datta (1997) applied the nonlinear optimization model in identifying unknown pollution sources and used finite differences method to simulate the physical processes of transient flow and transport in the groundwater systems. They formulated the source estimation problem as a constrained optimization form and solved the objective function by nonlinear programming. In their study, the source
information for flow in steady or transient state was successfully identified.
Recently, Chang and Yeh (2005) proposed a monitoring well design optimization method and developed a new model called SATS-GWT to identify the contaminant information. The model combines the simulated annealing, tabu search and three-dimensional groundwater flow and solute transport model MODFLOW-GWT, which was developed by the USGS. The proposed approach was employed to investigate the optimal number of sampling points and conditions for effectively estimating contaminant source information in three-dimensional homogeneous aquifers. They also suggested a guideline to allocate the sampling point location.
1.2.2 Latin Hypercube Sampling
Numerical simulation for the real world problems is often faced with a large number of simulation runs. Latin Hypercube Sampling (LHS) method can be applied to reduce the number of MCS method required. Mckay et al. (1979) first proposed LHS as an alternative method to random sampling. They compared three sampling methods, including LHS, simple random sampling, and stratified sampling, and associated estimators of the mean, variance, and the population distribution function of the model output. The results obtained from LHS appeared to be more precise than the other two types of sampling methods.
Gwo et al. (1996) constructed a hypothetical waste field and used the LHS
technique to analysis the sensitivity of subsurface stormflow parameters. In their study, the LHS could largely reduce the computational requirements in typical Monte Carlo uncertainty analysis and risk assessment. Zhang et al. (2003) applied the LHS method to compare with three random field generation algorithms: sequential Gaussian simulation available in GSLIB, the turning-bands method, and LU decomposition. The result showed that the LHS method gives a minimum deviation of the variance and it preserves the marginal distributions of the simulated variables.
By using LHS in their study, the computational effort needed in solving groundwater flow and transport models was greatly reduced.
1.3 Objectives
This study has three objectives. The first objective is to investigate the relationship between the number of sampling points and the chance of the source identification using MCS in different characteristics of groundwater sites. The second objective is to use LHS instead of MCS in source identification and compare the results with those of MCS. The third objective is to study effect of patterns of sampling location on the source identification results.