poster - International Conference of Agricultural Engineering
poster - International Conference of Agricultural Engineering
poster - International Conference of Agricultural Engineering
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Stochastic modelling <strong>of</strong> Contaminants Transport through<br />
Groundwater Using a Moving Least Squares Response Surface<br />
Method with Hermite Polynomials<br />
T. H. Bong 1 , Y. H. Son 1 *, S. K. Noh 1 , J. S. Park 1 , S. P. Kim 2 , J. Heo 2<br />
1 Department <strong>of</strong> Rural Systems <strong>Engineering</strong>, Seoul National University, Gwanak-ro, Gwanakgu,<br />
Seoul, 151-921 Korea<br />
2 Rural Research Institute Korea Rural Community Corporation, Haean-ro Sangrok-gu, Ansan,<br />
426-908 Korea<br />
*Corresponding author. E-mail: syh86@snu.ac.kr<br />
Abstract<br />
It is important to understand contaminants transport through groundwater. To obtain reliable<br />
result, uncertainties <strong>of</strong> parameter should be considered. In this study, uncertainty in Benzene<br />
concentration at various distances is analyzed. Two parameters ( K and K<br />
oc<br />
) are<br />
considered as random variables and probability density distributions <strong>of</strong> random variables are<br />
estimated using Hermite polynomials. MLS-RSM is performed to improve the efficiency <strong>of</strong><br />
uncertainty analysis and can achieve results close to accuracy <strong>of</strong> MCS with a sample size <strong>of</strong><br />
100,000. In conclusion, considering the actual probability density distributions by Hermite<br />
polynomials, more accurate results can be obtained.<br />
Key words: contaminants transport, MLS-RSM, stochastic modelling, Hermite Polynomials<br />
1. Introduction<br />
The groundwater pollution load has been increased as a result <strong>of</strong> human activities related to<br />
industrial and agricultural production on water resources. Thus understanding contaminants<br />
transport through groundwater is important for water management and remedial action plans.<br />
However, groundwater modelling is not an easy task. To get reliable results, input data<br />
should be accurate and representative <strong>of</strong> the reality <strong>of</strong> the field and uncertainties in model<br />
input data including chemical, physical and hydrogeological parameters also needs to be<br />
considered (Baalousha & Köngeter, 2006). Several recent studies have been conducted on<br />
the modelling <strong>of</strong> contaminants transport though ground water considering uncertainties<br />
(Datta & Kushwaha, 2011; Isukapalli, 1999; Jim yeh, 1992; Ranade et al., 2010).<br />
Probabilistic uncertainty propagation methods available fall into three categories: (a) Monte<br />
Carlo simulation (MCS), (b) analytical method, (c) response surface method (Phoon & Huang,<br />
2007). The MCS is a universal method regardless <strong>of</strong> the complexities underlying the physical<br />
model and/or input uncertainties. However, when the models are complex, or when there are<br />
numerous parameters, the MCS can be costly and time-consuming. Therefore, a variety <strong>of</strong><br />
variance reduction techniques have been proposed to reduce the number <strong>of</strong> runs.<br />
Among these, the response surface method (RSM) has been used in probabilistic modeling<br />
for various fields. However, the RSM requires transformation <strong>of</strong> input variable from the<br />
original space (X) to the standard normal space (U). To do this, probability density function<br />
(PDF) <strong>of</strong> input data is assumed by common PDF (e.g., normal distribution or lognormal<br />
distribution) and transformed using equivalent mean and standard deviation. It is difficult to<br />
estimate PDF from empirical data, especially when sample sizes are small. The reliability <strong>of</strong><br />
result relies on input variable, and therefore, a realistic PDF should be considered by actual<br />
measurement rather than assumed probability distributions. Phoon (2003) proved that any<br />
random variable can be expanded as a sum <strong>of</strong> Hermite polynomials. The remaining issue is<br />
how to improve the efficiency <strong>of</strong> the calculation. Compared with traditional methods, the RSM