Comparison of Alternative Filtering Algorithms for Estimating Background Groundwater Levels at Savan

1
Comparison of Alternative Filtering Algorithms
for Estimating Background Groundwater Levels at
Savannah River Site, SC Presented at theISEA
2001 ConferenceExposure Analysis An Integral
Part of Disease PreventionCharleston, SC
November 4-8, 2001byVikram M. Vyas, David S.
Kosson, Amit Roy,William Strawderman, and
Panos G. Georgopoulos

• Exposure Measurement Assessment Division,
  Environmental and Occupational Health Sciences
  Institute (EOHSI), 170 Frelinghuysen Road,
  Piscataway, NJ 0885
• Department of Civil and Environmental
  Engineering, Vanderbilt University, Nashville, TN
• Dept. of Statistics, Rutgers, The State
  University of New Jersey, Piscataway, NJ

2
Introduction

• Drinking water standards (DWS) require
  contaminated groundwater to be cleaned to
  minimize the risk of human exposure through
  domestic water usage.
• In certain situations, background levels of
  Constituents of Concern (COC) may exceed DWS.
• Background levels are defined as
• Naturally occurring concentrations
• Concentrations due to prior usage of site or
• Concentrations due to diffuse off-site sources.

3
Identification of Background Levels at SRS.

• US-EPA recommends that samples from identified
  uncontaminated areas be taken for determining
  background levels. This simple determination
  cannot be employed at SRS because
• There are no recognized areas or monitoring
  locations on the site that have been designated
  as completely unimpacted.
• There are large spatial gaps in the monitoring of
  the overall groundwater quality.
• There are several spatially separated sources and
  the resulting plumes have intermingled.
• There are different sources for different COC.
• There is contamination from activities prior to
  the establishment of SRS (Arsenic from
  agricultural usage).

4
Groundwater Quality Data for Determination of
Background Levels

• Groundwater monitoring data used for this study
  were obtained from the Geochemical Information
  Management System (GIMS) database maintained at
  SRS.
• Data for the water table aquifer over the period
  from October 1992 to September 1998 were used in
  this study.
• Data from other sources within and outside SRS
  were reviewed. While these data were used to
  evaluate outcomes of this study, they could not
  be used directly in the analyses because of
  inconsistencies in sampling, analysis and
  reporting of data.
• Sixteen COC were considered for analysis in the
  study aluminum, arsenic, cesium-137, cobalt,
  iron, manganese, mercury, nitrate as nitrogen,
  nitrite as nitrogen, pH, selenium, thallium, tin,
  tritium, uranium-238, and vanadium.
• About 90,000 measurements from 869 monitoring
  wells in the water table aquifer were used for
  the analysis.

5
Preliminary Analysis
6
Algorithm for Identifying Background Levels
7
Filters Applied in Method 1

• Use Intrinsic Spatial Estimation to obtain
  spatially interpolated maps of the 75th
  percentiles of concentrations in each well.
• Select potentially uncontaminated wells from the
  corresponding maps (spatial estimation results).
• Perform the Mann-Kendall and Seasonal Kendall
  tests to reject wells with temporal trends.
• Remove outliers from the time series in each well
  based on the Shewhart Cumulative Sum (CUSUM)
  test.
• Use the regression-based approach to identify the
  background wells.

8
Example of Method 1 Application of
Geostatistical and GIS Methods
9
Example of Wells Rejected by Method 1 Step 3
10
Example of Method 1. Step 5
11
Filters Applied in Method 2

• Perform the Mann-Kendall and Seasonal Kendall
  tests to reject wells with temporal trends.
• Remove outliers from the time series in each well
  based on the Shewhart Cumulative Sum (CUSUM)
  test.
• Use cluster analysis to separate monitoring wells
  into two groups impacted and unimpacted wells.

12
Example of Well Grouping by Cluster Analysis
Method
13
Comparison of Results from Two Filters Summary
Statistics
14
Comparison of Results from Two Methods Number
of Background Wells
15
Comparison of Results with Results from Other
Studies

• The 95th percentiles of tritium are about two
  times higher than levels reported by Summerour et
  al. (1996) (1600 pCi/L) on the southwest of SRS.
  The 50th percentile of background levels for
  tritium is 1480 pCi/L, which is extremely close
  to the observations from Summerour et al.
• The 95th percentile for aluminum 200.5 ?g/L is
  very close to the peak measurement of aluminum
  reported by Strom and Kabeck (1992) (193 ?g/L)
• For iron and manganese, the 95th percentiles fall
  well within the ranges reported by Strom and
  Kabeck (1992).