11:05 AM - 11:20 AM
[AGE27-02] Improvement of contaminant plume estimation by a geostatistical method considering groundwater flow and non-negativity
Keywords:Underground contamination, Geostatistical analysis, Groundwater flow, Gibbs sampling
For underground contamination by such as radioactive nuclides and chemicals, the contaminant plume distribution needs to be clarified accurately for effective remedy. Existing methods for plume estimation can be divided in two categories. The first one is simulation of groundwater flow and contaminant transport. However, if the history of pollution is unknown, the calculation condition cannot be set properly. The second one is spatial interpolation and extrapolation of measurement data, such as Kriging. However, once a pollutant reaches an aquifer, the transport will be affected by groundwater (advection, dispersion, adsorption and retardation). In such case, spatial modeling of measurement data without considering of physical/chemical law may be unable to estimate the contaminant plume accurately.
In this study, we considered the estimation method integrated transport information into geostatistical analysis. In this method, release history from a known source is estimated inversely using the measurement concentration. Then, the entire plume distribution is calculated by the release history. However previous method does not impose any limitations on the results; therefore, the negative value can be estimated. To gain physically feasible solution, we considered the non-negativity constraint by Gibbs-sampling, which is based on Bayes’ rule and statistically rigorous. The applicability of the method for underground contamination was confirmed for both hypothetical model and actual contamination case (Gloucester landfill, Canada). The groundwater flow and solute transport for porous media was calculated using 3D-SEEP code.
As the hypothetical model, the radioactive nuclide transport (3H) in 2-dimensional and homogeneous media (300×60 m2) was considered. The pollutant was assumed to be leaked for 300 days with 2 peaks. We used the concentration data taken after 330 days at 18 points placed at equal intervals. By spatial interpolation (Kriging with a Trend), the two peaks of plume could not be reproduced. However, the plume was reproduced well by the geostatistical method considering groundwater flow with the mean average error (MAE) of 1.0×10-4. By considering non-negativity constraint, the MAE was reduced to 2.7×10-9.
As the case study, we considered the pollution at the Gloucester landfill by 1,4-dioxane, which is highly soluble in water and resistant to biodegradation. The release history of 1,4-dioxane is almost unknown (beginning in 1969 and until 1980), but it is known that large spill is occurred sometime in 1978. We estimated the three-dimensional plume in the area of 300×300×40 m3 using analytical solution for uniform flow, whose parameters are based the investigation at the site. We used concentration data at 69 points measured in 1982. By the geostatistical method considering groundwater flow and nonnegativity constraint, the large spill in 1978 could be reproduced well compared to previous research by other methods (minimum relative entropy approach and a geostatistical approach with Metropolis-Hasting). The plume was also evaluated well with the MAE of 2.8×10-2.
This study includes the results of contract work funded by the Secretariat of the Nuclear Regulation Authority, Japan.
In this study, we considered the estimation method integrated transport information into geostatistical analysis. In this method, release history from a known source is estimated inversely using the measurement concentration. Then, the entire plume distribution is calculated by the release history. However previous method does not impose any limitations on the results; therefore, the negative value can be estimated. To gain physically feasible solution, we considered the non-negativity constraint by Gibbs-sampling, which is based on Bayes’ rule and statistically rigorous. The applicability of the method for underground contamination was confirmed for both hypothetical model and actual contamination case (Gloucester landfill, Canada). The groundwater flow and solute transport for porous media was calculated using 3D-SEEP code.
As the hypothetical model, the radioactive nuclide transport (3H) in 2-dimensional and homogeneous media (300×60 m2) was considered. The pollutant was assumed to be leaked for 300 days with 2 peaks. We used the concentration data taken after 330 days at 18 points placed at equal intervals. By spatial interpolation (Kriging with a Trend), the two peaks of plume could not be reproduced. However, the plume was reproduced well by the geostatistical method considering groundwater flow with the mean average error (MAE) of 1.0×10-4. By considering non-negativity constraint, the MAE was reduced to 2.7×10-9.
As the case study, we considered the pollution at the Gloucester landfill by 1,4-dioxane, which is highly soluble in water and resistant to biodegradation. The release history of 1,4-dioxane is almost unknown (beginning in 1969 and until 1980), but it is known that large spill is occurred sometime in 1978. We estimated the three-dimensional plume in the area of 300×300×40 m3 using analytical solution for uniform flow, whose parameters are based the investigation at the site. We used concentration data at 69 points measured in 1982. By the geostatistical method considering groundwater flow and nonnegativity constraint, the large spill in 1978 could be reproduced well compared to previous research by other methods (minimum relative entropy approach and a geostatistical approach with Metropolis-Hasting). The plume was also evaluated well with the MAE of 2.8×10-2.
This study includes the results of contract work funded by the Secretariat of the Nuclear Regulation Authority, Japan.