1. Geostatistical Modeling as a Quality Management Tool to Address Uncertainty in Decision-making for Large Scale Sediment Assessment and Remediation Projects
Judith A. Schofield1, Pierre Goovaerts, Justin Telech, Ken Miller, and Molly Middlebrook Amos
Computer Sciences Corporation
Louis Blume
U.S. EPA Great Lakes National Program Office
U.S. EPA’s 28th Annual Conference on Managing Environmental Quality Systems
May 14, 2009
1Presenter
Introduce GLNPO and us

2. Acknowledgments Diana Mally David Wethington (now with USACE)
From U.S. EPA’s Great Lakes National Program Office, we acknowledge the following project leads and contributors:
Diana Mally
David Wethington (now with USACE)
Marc Tuchman
And from the Michigan Department of Environmental Quality:
Michael Alexander 2

3. Geostatistics
Set of statistical techniques used in the analysis of georeferenced data
Increasingly popular in part due to the availability of geographic information systems (GIS) software
Powerful tool when used in combination with GIS

4. How can Geostatistics be used as a Quality Management Tool?
Pools data to get the best representation of the site
Uncertainty can be quantified
Supports generation of cost effective sampling designs
Large quantities of data can be more easily visualized
Decisions are defensible, transparent, well-documented, and reproducible
Facilitates informed cleanup decisions and effective use of remedial resources
Will illustrate these points through a case study

5. Geostatistics in Sediment Assessment and Remediation
Describe extent and nature of contamination
Identify data gaps
Generate statistical sampling designs
Calculate sediment volumes
Develop remedial design
Evaluate achievement of cleanup goals
Communicate conditions to stakeholders
The case study will illustrate many of these

6. Sediment Assessment and Remediation Projects using Geostatistics
Fox River, WI
Hudson River, NY
Minnesota Slip, Duluth Harbor, MN
East Fork Poplar Creek, TN
Great Lakes Legacy Act
Black Lagoon, MI
Hog Island, WI
Ruddiman Creek, MI
Division Street Outfall, MI
St. Louis River, WI
Ashtabula River, OH
Trenton Channel, MI
Buffalo River, NY
Lincoln Park, WI
Intro for Legacy act

7. Trenton Channel of the Detroit River
The Detroit River is one of 42 Areas of Concern (AOCs) in the Great Lakes
investigations of the Upper Trenton Channel, within the Detroit River AOC, have shown that sediments are contaminated with polychlorinated biphenyls (PCBs), mercury and total polycyclic aromatic hydrocarbons (Total PAHs) among other contaminants
EPA’s Great Lakes National Program Office (GLNPO) and the Michigan Department of Environmental Quality (MDEQ) are evaluating the extent of sediment contamination in support of a potential Great Lakes Legacy Act cleanup project

10. Trenton Channel RI/FFS
In 2006, GLNPO and MDEQ initiated a remedial investigation and focused feasibility study (RI/FFS) of the site
Sediment samples were collected and analyzed for a large variety of contaminants of concern
Initial sampling was conducted in 2006 data (Phase I of the RI/FFS)
Based on review of the Phase I data, GLNPO and MDEQ developed a series of questions that were the focus of additional sampling in 2007 (Phase II of the RI/FFS)
Phase II based on geostatistical analysis
Three study questions
Three dqos, excellent example of systematic planning!

13. Sediment Concentrations
COC – contaminant of concern
TOC – threshold of concern

14. Describe traditional approach of making decision on each sample
Can easily make wrong decision again and again
Box around sample is arbitrary
Does not tell you where to dredge and how sure you are that you are contaminated

15. Statistical and Geostatistical Analysis
Exploratory Data Analysis
Statistics
Hypothesis testing using t-tests and regression
Geostatistical Analysis
3D modeling
SGeMS (Stanford Geostatistical Modeling Software)
This talk will focus on geostatistical

16. Kriging
Evolved in mineral exploration and mining of minerals, ores, and coals
In 1963, G. Matheron named kriging after Daniel Gerdhaus Krige, a South African mining engineer, who used the technique to more accurately predict the extent of gold deposits in unsampled areas

17. Kriging Method of interpolation
Optimally predicts data values by using data taken at known locations
Creates contours or isopleths of data across an area
Other common methods of interpolation include inverse distance weighting and spline

18. Geostatistical Analysis Basics
Overlay grid
Model sediment depth and create 3D grid using ordinary kriging
Transform contaminant concentrations
Compute 3D variogram for each contaminant and fit weighted least-square regression model
Estimate contaminant concentrations for each block using kriging & surrounding observed concentrations

19. Min
Max
Note: Results are preliminary.

20. Min
Max
Note: Results are preliminary
Total PAH Concentrations in Sediment at the Trenton Channel Site, View from Southeast of the Site
(depth exaggerated 25 times)

21. Min
Max
Note: Results are preliminary
Total PAH Concentrations in Sediment at the Trenton Channel Site, View from Northwest of the Site
(depth exaggerated 25 times)

22. Min
Max
Note: Results are preliminary
Total PAH Concentrations in Sediment at the Trenton Channel Site, View from Northeast of the Site
(depth exaggerated 25 times)

23. Model Evaluation and Validation
Consistency of 3D model with core data

24. Model Evaluation and Validation
Consistency of 3D model with core data

25. Note: Results are preliminary

26. Note: Results are preliminary

27. Stochastic Simulation
Generate equiprobable models for mercury, Total PCBs and Total PAH distributions
Apply dredging scenario to each set of 3 simulations
Compute corresponding volume to be dredged
Simu #1 (Hg)
Simu #1 (PCB)
Simu #1 (TPAH) 27

28. Uncertainty about Dredging Volumes
Simu #1
Simu #3 
… … …
Simu #2
Simu #50
Best case
scenario
Worst case
scenario

29. Sediment Volume Estimates Mean Minimum Maximum 90,317 80,971 101,633
Note: Results are preliminary

30. Probability that at least one TOC is exceeded
TOC is CBSQG, One of several thresholds of concern being evaluated in feasibility study
Note: Results are preliminary
TOC – threshold of concern

31. Probability that at least one TOC is exceeded
Note: Results are preliminary
TOC – threshold of concern

32. Next steps
Depending on next steps at the site, develop sampling design that addresses areas with greatest uncertainty at threshold of concern

33. Geostatistics Pools data to get the best representation of the site
Facilitates informed cleanup decisions and effective use of remedial resources
Uncertainty can be quantified
Geostatistical analyses can support generation of cost effective sampling designs
Large quantities of data can be more easily visualized
Decisions are defensible, transparent, well-documented, and reproducible

34. Lessons Learned Follow systematic planning! Clearly define decision
Develop sampling design considering specific data analysis techniques
Communicating results is a challenge and requires investment of project lead
Collection of accurate representative sediment depth data is critical
Research is needed to ground truth and refine the tools for sediment projects