Collaborations with an outstanding team of ORNL scientists from Chemical Sciences Division (including Wayne Griest (retired), Kevin Hart, Marcus Wise, Mike Burnett, Cyril Thompson, Rob Smith, Ralph Ilgner, Roosevelt Merriweather, Roger Jenkins (retired), Stacey Barshick (no longer at ORNL), and Stephen Lammert (now with Stillwater Scientific Instruments)), Life Sciences Division (Arpad Vass), Computational Sciences and Engineering Division (including Irene Robbins, Cindy Terry, and Dick Reid), and Engineering Science and Technology Division (including Dwight Clayton, Andy Andrews, and David McMillan). Organizations external to ORNL involved in the project include MSP Corporation, Colorado School of Mines and Hamilton Sundstrand (industrial partner manufacturing the systems for the US Army). Other collaborators include Dugway Proving Ground, White Sands Missile Range, and the Armed Forces Institute of Pathology.|
This project received an R & D 100 Award, an award presented annually by R&D Magazine in recognition of
the year's most significant technological innovations. The Block II Chemical Biological Mass Spectrometer (CBMS)
is the first integrated instrument capable of detecting and identifying both chemical and biological warfare agents in the field.
(See ORNL Review, Volume 33, Number 2, 2000, "Mass Spectrometer Can Detect Weapons of Mass Destruction.") After 9/11 there has been national recognition of ORNL's capabilities (ABC national news) in general and the CBMS project (WVLT local news) in particular.
Early in the CBMS program while instrument concept and design were evolving we researched analytical methods that would optimize the ability to differentiate among microorganisms using pyrolysis-ion trap mass spectrometry. Multivariate statistical techniques were invaluable for determining relative merits of electron ionization (EI) and chemical ionization (CI) techniques and then to identify the preferred chemical reagent for optimal bacterial discrimination. The most obvious benefit of chemical
ionization is the reduced fragmentation of higher mass compounds. Analyses of fatty acid standards showed very simple, unique fragmentation
patterns characterizing important biomarkers. This work appears in the open literature in S. A. Barshick, D. A. Wolf, and A. A. Vass (1999).
"Differentiation of Microorganisms Based on Pyrolysis-Ion Trap Mass Spectrometry using Chemical Ionization," Analytical Chemistry, 71,
No. 3, 633-641.
It is well-documented (e.g. D. F. Welch (1991). "Application of Cellular Fatty Acid Analysis," Clinical Microbiology Reviews, 4,
No. 4, 422-438 that cellular fatty acid composition is stable genetic trait that can be used to characterize bacteria and fungi. In fact there is a commercial system MlDI, Inc. that takes advantage of that. They have developed a highly multivariate data set consisting of more than 150 features, results from high resolution gas chromatography (GC) analysis of organism fatty acids. The system was designed for laboratory use where the growth conditions were strictly controled. The challenge for the CBMS system will be able to detect and identify bioagents grown by the enemy or by terrorists and disseminated in some way. Weather conditions, environmental interferents, and battlefield conditions are just a few of the factors affecting what the instrument will be presented for analysis in a real world scenario.
To accomplish all that needed to be done throughout the project there were other non-statistical, but important tasks that I was involved with.
I developed a database for tracking the inventory of agents, a system designed to meet the CDC accountability requirements and Project
needs. A web interface was developed for easy access to the inventory by CBMS team members.
I also developed several web applications to aid project management and data collection. And I operated a mass spectrometer.
Collaborations with George Ostrouchov (Computer Science and Mathematics Division), Les Beard and Bill Doll (now at Battelle - Oak Ridge Operations)
and Max Morris (Iowa State University)|
This project was funded through the Strategic Environmental Research and Development Program (SERDP), a Department of Defense (DoD)
corporate environmental research and development (R&D) program, planned and executed in full partnership with the Department of Energy (DOE)
and the Environmental Protection Agency (EPA), with participation by numerous other Federal and non-Federal organizations.
UXO (unexploded ordnance) surveys encompass large areas, and the cost of surveying these areas can be high. Enactment of earlier protocols
for sampling UXO sites (SiteStats/GridStats) have shown the shortcomings of these procedures and led to a call for development of
scientifically defensible statistical procedures for survey design and analysis.
One specific goal of the project was to process data from magnetometer signal transects collected with an airborne platform.
My main contribution to that project was to develop a multivariate statistical approach to processing magnetometer signals,
identifying a number of signal characteristics that can be used for discrimination between ordnance-related and nonordnance-related items.
This approach had never been tried heretofore and has been used with success to process aerial survellance data from the
Badlands Bombing Range, Aberdeen Proving Grounds, and sites in New Mexico. Follow on funding is being sought to improve the technique.
The main emphasis of the work was spatial statistical models and optimal survey design for characterization of sites contaminated with UXO's
The problem is a very complicated one statistically, with a need to develop new approaches in survey design and data analysis.
It is critical that a new protocol for design and analysis of UXO surveys be able to address: 1) the idiosyncrasies of different
sites in terms of their geology, ordnance types, topography, vegetation, and extent of background knowledge; 2) the uncertainties
in performance of different types of instrumentation and instrument platforms that are available, or becoming available for UXO surveys;
3) the distinctions between UXO contamination, which occurs at discrete points, and chemical contamination, which has a more continuous distribution;
4) the opportunities to interrogate the site through sequential surveys; and 5) the regulatory and public-involvement environment in which
these surveys are typically performed. The protocol must allow for changes in these factors that may result from technological advances.
A protocol that neglects some or all of these issues may never be suitable for routine use.
Statistically based methodologies are being used, and should be used, to efficiently determine the extent of UXO contamination by optimizing
locations for surveys, defining how to conduct excavation, and developing procedures to incorporate survey data into decision making.
These methods may also be used to prioritize areas, compare different clearance approaches, and to estimate costs for different land uses
(i.e. for different specified levels of clearance).
Statistical methods are used to reduce the extent of data acquisition required to characterize large areas. Our protocol has led to
development of statistically valid tools that can be used to support management decisions by providing maps of estimated contamination
and associated probabilities and uncertainties at different stages of the characterization process. Such tools enable decisions that
lead to appropriate remediation.
Collaborations with Arpad Vass (Life Sciences Division)|
The Time Since Death Project involved Life Sciences and Chemical Sciences Divisions. The objective of the project was
to develop measurement technologies capable of determining time since death (TSD). A method is sought that does not depend (or depends
little) on environmental conditions and is based on chemical markers. The work is to understand human decomposition chemistries and
relate them to TSD. The result of this work is to be an analytical method and model for TSD based in technologies currently available
to most forensic and pathology laboratories.
There are several stages in the decomponsition process.
Flies begin to lay eggs in every available crevices during the initial, fresh stage.
Bloating occures in the next stage as the result of gaseous byproducts of bacteria growth.
In the third stage, called active decay, the body's soft tissue liquefies.
At the fourth stage, the body is little more than bones.
Several technologies were evaluated, Aroma Scan of gasses emitted from decaying bodies and gas chromatographic analysis of amino acid
composition of various tissues from decaying bodies.
The goal of the Aroma Scan study was to evaluate odor of decay as an indicator of time since death (TSD).
To accomplish this goal eleven human bodies were placed about
the Anthropological Research Facility at the University of Tennessee
and left to decay.
The aroma of each decaying body was measured
at daily intervals over a period of approximately one month
by an electronic nose composed of 32 sensors.
Previous research had identified accumulated degree days (ADD),
a composite measure of temperature and time,
as a useful correlate of other time since death measures.
This study investigated ADD and aroma measurement relationships.
In another study, conducted at the University of Tennessee's Anthropological Research Facility (ARF),
a database was created consisting of chemicals that are liberated during the decompositional process
from buried human remains. The work supports establishment of the chemical basis for canine’s scenting
ability when detecting human remains.
The initial database spans the first year and a half of burial, providing identification, chemical trends
and semi-quantitation of chemicals liberated below, above and at the surface of 1.5 - 2.5 ft graves for three
individuals (two males and one female). A fourth male individual (buried in 1990) was also sampled to provide
possible 'endpoint' information. In-ground, in-corpse thermocouples provided temperature information which
can be used to correlate accumulated degree days (ADDs) to surface decompositional events.
Clear, sealed, hollow pipes were also placed in the grave vault providing viewing ports by which the burial
decompositional process could be monitored.
Results confirm that burial decomposition is approximately eight times slower than surface decomposition.
Movement of these chemicals through the soil column significantly reduces the chemical species which might be available to detection
devices, including canines. Triple sorbent traps (TSTs) were used to collect air samples in the field. The TSTs were thermally
desorbed in the laboratory and analyzed on a GC-MS system revealing eight major classes of chemicals containing 424 specific
volatile compounds associated with the decompositional process of burials. This research furthers our understanding of human
decomposition and has profound ramifications on cadaver dog training procedures and in the development of field portable
analytical instruments which can be used to locate buried human remains. These results will appear in the literature
(Arpad A. Vass, Rob R. Smith, Cyril V. Thompson, Michael N. Burnett, Dennis A. Wolf, Jennifer A. Synstelien, Brian A. Eckenrode,
and Nishan Dulgerian "Decomposition of Odor Analysis Database - Phase I" to appear June 2004 in Journal of Forensic Science).
The purpose of another study was to identify compounds that cadaver dogs may be sensing, determine whether human and animal bones can
be distinguished, compare bone type within an individual, and to look for time change in the compound composition. Human and animal
bones were obtained from the University of Tennessee to do bone vapor analysis.
The bones were analyzed to identify compounds that are emitted. Femur, humurus, and/or mandible bones were obtained for nine humans
dying during the period of 1973 through 2001 and femur bones from nine animals (unknown death years), bear, cow, deer, dog, gray fox,
pig, rabbit, raccoon and sheep. A controlled vapor analysis was done to determine the relative concentration of approximately
192 compounds, concentration relative to a standard of bromobenzene.
Collaborations with Mark Tardiff (now at Pacific Northwest National Laboratory),
Dick Kettelle (now at Bechtel Jacobs)
and others in Environmental Sciences Division and the various environmental monitoring and compliance organizations at X-10, Y-12 and K-25|
I provided statistical and computing expertise for more than 9 years to clients from environment-related fields. I worked primarily
on problems concerning Oak Ridge Reservation (mainly X-10) groundwater and surface water, but have worked on projects focusing on soil
and air as media of concern.
I developed procedures to evaluate data quality and trends, e.g. groundwater charge balance to assess internal consistency of analytical
results, statistical measures and graphical displays of trend in individual well groundwater constituents and surface water contaminants
to determine effects of possible contamination or season. For example to assess wells for cancer risk and to facilitate selection of
X-10 wells to retain for monitoring. I developed a type of graphical display to determine a well's components of risk and risk
trends over the well's sampling history.
I have effectively assembled, shaped and merged data from a variety of sources for analysis of
e.g. surface water and groundwater data from X-10's environmental compliance organization,
meteorological and discharge data from ESD, US Geological Service (USGS), and
National Oceanic and Atmospheric Administration (NOAA), well construction data,
plant operations data, regulatory limits and risk constants.
I am familiar with sampling and analysis issues affecting the information content of environmental data, e.g. detection and
radionuclide counting uncertainty and background determination when doing gamma scans and gross counting,
types of composite sampling,
have accompanied technicians when they sampled surface water and groundwater.
Have familiarity of the Oak Ridge Reservation (mainly X-10) geography,
hydrogeology and contaminant sources gained through consulting with the X-10's environmental compliance organization and
collaborations on environmental restoration projects.
To follow are specific projects I was involved in:
Oak Ridge Reservation Groundwater Characterization Projects at X-10, Y-12 and K-25.
The reference groundwater studies were undertaken to get a site-wide perspective on
groundwater monitoring and assessment. In the past most studies of groundwater flow
and geochemistry on which knowledge was based were narrowly focused studies of single
sites. The expanded objective was necessary because (1) groundwater flow paths are not
confined to specific waste disposal boundaries, (2) site-wide assessment is needed
to set priorities for groundwater remedial actions, (3) information about groundwater
transport mechanisms should be applied across the entire site if possible, and (4) there
is a need to establish a basis for conducting baseline risk assessments.
Representatives from the three plants met and discussed the approach to the problem,
identifying the needs for risk assessment, remedial investigations and feasibility
studies. It was agreed that statistical distributions should be estimated for groundwater
constituents for each background (reference) population. An exploratory data analysis
approach to identifying background populations was used to confirm/update the current
understanding of groundwater types.
Part of this effort has been to identify groundwater regimes at the sites. We explored
a variety of options in this process which involves principal components analysis
to identify underlying characteristics of the groundwater, two forms of cluster analysis to identify
groups of wells with similar characteristics and tree-based modeling
to classify wells into groups identified by previous cluster analysis. Key choices
to be made in these analyses included which groundwater parameters to analyze, data
transformation and scaling, number of groundwater regimes (clusters), etc. It is time consuming
to implement a set choices, carry out an analysis and evalute the results. The ultimate
criterion for evaluation is whether the groundwater regimes make sense. The geologist collates the
spatial and geologic information about members of a cluster of
wells, wells whose waters are derived from the identified groundwater regime, with
the geochemical characteristics to understand the regime. This process is somewhat
circular in that the analysis process is modified until the geologist gets what he deems
sensible. However this is
an exploratory analysis. The groundwater system is complex and the derived groundwater
regimes probably do not represent exactly some fixed number of water types. Some
types may be quite distinct whereas others may be less distinct or perhaps arbitrary.
Although it is impossible to identify the "true" groundwater regimes, it may be possible
to achieve a parsimonious partitioning that appears to the geologist as valid and
Through several iterations of the analysis at least two dominant groundwater types
have been identified, one associated with shale, the other with limestone.
Investigated possible brine influence of deep groundwater on X-10 wells and seeps.
Wells and seeps were identified where the sodium and chloride constituents in the
water met certain criteria.
The analysis identified 10 groups of wells, four of which are influenced by contaminants.
The other six represent what is believed to be natural groundwater conditions.
The approach and statistical methodology were presented to representatives of DOE,
EPA and State of Tennessee Department of Environment and Conservation to elicit feedback
on acceptability for background determination.
Derived clusters separately for the K25 groundwater Fall 1994 sampling data and Spring
1995 sampling data. Four clusters may be appropriate for both data sets and membership
in two of the clusters is fairly consistent between the seasons. However, for the
two largest clusters there are wells that change membership with season. Interpretation
of the clusters is proving very difficult at K-25 because of the extent of anthropogenetic
effects, cut and filled areas, change of stream beds, tremendous amount of unaccounted for
treated water in the water distribution system, etc.
Also derived clusters separately for the Clinch River/Poplar Creek surface water data.
It has been hypothesized that there may be intimate interaction between the surface
water and the groundwater especially near Poplar Creek. There appear to be three
clusters essentially distinguished as a high total solids Poplar Creek cluster, a low
total solids Poplar Creek cluster, and an intermediate total solids cluster consisting
of only Clinch River sampling sites and some Poplar Creek sites. These clusters seem
to make sense when considering season and flows in Poplar Creek. However these clusters
do not appear to be directly comparable to the clusters obtained in the groundwater
analyses probably because of limited groundwater sampling near surface water and
sampling performed in different years.
Focused on several areas most affected by organic contamination and investigated spatial
and temporal trends in organic constituents, particularly several that decay to other
organic compounds. Attempts to identify specific sources of the organic contaminants through
this analysis was frustrated by the complexity of the site, in particular
the alterations in groundwater flow caused by building sumps (>90 at K-25 site) and
extensive underground piping.
Attempted to identify direct interactions between groundwater and surface water and
potable water leaks. This was done by comparing constituent profiles, e.g. pH, conductivity,
calcium concentration, chloride concentration, etc. Wells where sampling event profiles were most
similar to a surface water profile or a potable water profile were
identified as those most likely as having a direct link.
Investigated relationships between stage elevation in Clinch River and flow at weirs
along Poplar Creek and Mitchell Branch. It is known that there are karst features
underlying the K-25 Site, but it is not known to what extent they may be affecting
groundwater flow and discharge into surface water. There appear to gaining and losing sections
along some creeks which may be evidence of karst effects. It is a complex system
and we lacked data that fully addresses this question. However we are looked at flow
at several surface water weirs, precipitation, stage elevation and Watts Bar Dam and
sparse information on storm drain discharge to the creeks to try to understand what
X-10 Well Evaluation for Future Monitoring
Evaluated active X-10 wells wells to determine what wells to monitor in the future. Criteria considered to determine
the "usefulness" of a well included: (1) location of well, i.e. whether a well is
in a contaminated area or along a contaminant flow line, (2) usefulness in monitoring groundwater
regimes, and (3) whether the well provides reference water quality.
One criterion for assessing a well for continued monitoring is its water's cancer
risk potential. Numerical and graphical displays were created to evaluate analyte-specific
cancer risk and total cancer risk over each well's historical sampling record. Wells can be ranked by their
maximum total cancer risk. Kendall's tau, a nonparametric
measure of correlation, and a robust line-fitting procedure using the Kendall-Theil
estimate of slope were used to assess time trends. Kendall's tau is a general measure
of monotonicity. The trend model fit was simple linear model involving logarithm risk
and elapsed time.
Evaluation of Time Trends in Groundwater Constituents in X-10 Wells
Evaluated time trends in water elevation and routine groundwater parameters in wells
to be used to derive reference constituent concentrations. It is important that the
wells and sampling events included be representative of the population of interest.
Kendall's tau, a nonparametric measure of correlation, and a robust line-fitting procedure
using the Kendall-Theil estimate of slope were used to assess time trends. Kendall's
tau is a general measure of monotonicity. Conclusions from this work were that there exist trends
in a number of the routine groundwater parameters; however these trends
will probably have little effect on reference calculations. Some trends in water
elevation identified may be an artifact of the sampling schedule, because some wells
were sampled in different seasons in different years. However, some shifts in overall
water elevation appear to have occurred in some wells in the main plant area, possibly
suggesting anthroprogenic effects yet to be explained.
Extent of Groundwater Contamination in WAG 1
Assessed the extent of groundwater contamination in the WAG 1 (main plant) area. Data
were assembled from various sources, from sampling performed within WAG 1, at the
piezometer wells, at building sumps, at coreholes, at waste storage tank, at drywells surrounding the
waste storage tanks, at manholes,
and at compliance wells on the perimeter of WAG 1. Graphical and statistical analyses indicate that
the area of primary concern is a
narrow plume that extends from the area around the corehole 8, just below the X-10
cafeteria, to the western perimeter of WAG 1. The highest levels of alpha and beta contamination appear
in groundwater in the area
of corehole 8, in an area on the northwest perimeter of the main plant, and in water
collected in several building sumps. The sumps appear to have a beneficial effect
on the surrounding groundwater contaminant contaminations.
Methodology to Evaluate X-10 Wells' Groundwater Level Data
Developing methodology to evaluate historical groundwater level data from over one
thousand X-10 wells. Well data summaries are used for a number of things including
constructing groundwater level contours for the site.
Determination of 1st Creek Area Contribution to WAG 1 Total Radioactive Strontium
Investigated the 1st Creek area contribution to WAG 1 total radioactive strontium
flux. Radioactive strontium measurements obtained by Environmental Surveillance and
Protection at the 1st Creek and 7500 Bridge weirs were combined with flow measurements
collected at those locations by the USGS to determine the flux there. The 1st Creek contribution
has ranged up to 68% of the total radioactive strontium flux observed at 7500 Bridge
Developing some graphical summaries of surface water radionuclide contaminantion data
to assess the overall contributions of various conatminant sources on the discharge
into the Clinch River. Various sources include waste disposal areas and plant operations.
Modeled Sr-90 contamination in First Creek at the west end of the ORNL main plant
area. In 1991, as part of a study in WAG 1, rock core drilling was conducted to describe
subsurface geologic conditions. In corehole 8 contaminated water was discovered.
Subsequent study identified a contaminant plume that followed the contact between two
different rock types. That plume discharges (at least partially) into First Creek.
Historical total Sr data was used to model the change in the contaminant over time.
An overall decreasing trend in total activity was observed which may be consistent with a
model based upon a limited source of contamination and first-order elimination rate.
Watershed Dynamics and Contaminat Movement
Understanding watershed dynamics and contaminant movement depends upon flow measurements
and contaminant concentrations. Obtained flow information from Environmental Sciences
Divisions (ESD) for the three main weirs, Melton Branch 1 (MB1), White Oak Creek
(WOC) and White Oak Dam (WOD) and have processed USGS flow data from other weirs in
the watershed, precipitation data and groundwater level data into a form for analysis
of contaminant movement.
Redundancy of Precipitation Collection Stations in the X-10 Area
Analyzed hourly precipitation data collected at 13 sites on the Oak Ridge Reservation,
mainly in Melton and Bethel Valleys. The purpose of the investigationss was to assess
spatial variability in precipitation and event duration. Redundant sites could be
closed for cost savings.
Seep C & D Area Contribution to Total Radioactive Strontium Discharge at White Oak
Two seeps in Melton Valley near WAG 5 may be contributing more than 30% of the total
radioactive strontium that moves from ORNL over WOD. Remediation efforts are underway
to capture what is being discharged from these areas. The approach to be used for
two of the seeps is an in situ zeolite interceptor and/or pump, truck, and treating of
seep discharge. Baseline monitoring of these seep areas before implementation of
the removal actions is essential to evaluate the effectiveness of the actions. Effectiveness of the
removal action for the seep areas is measured by comparing contaminant concentrations
up-stream to the seep area to concentrations down-stream. Baseline measurements taken
during baseflow conditions, during storm events and during various seasons were modeled in an attempt
to understand system dynamics.
Provided graphical summaries of the effectiveness of the removal action on Melton
Branch radioactive strontium concentrations.
Supported OECD in their task monitor the environment at the X-10 site and at certain locations outside the Oak Ridge Reservation
(ORR). Part of the work involved management of environmental data, interaction with
analytical labs and field sampling teams. Developed procedures to evaluate surface
water and groundwater data for completeness and consistency. Some checks for internal groundwater sample consistency included comparisons of measured
versus calculated conductivity and total dissolved solids, charge balance, and comparisons
of measured conductivity to ion sums and total dissolved solids.
Evaluated calibration procedures for equipment used to monitoring noble gas emissions from X-10
stacks at HFIR. Noble gases present a challenge for monitoring because of their chemically
nonreactive nature. Radioactive noble gasses are monitored in a counting chamber
by a gross counting process, that is by a non-isotopic specific method. Stack gasses
are pulled through a counting chamber where radioactive decay activity is measured.
Calibration involves introduction of levels of a reference noble gas into the counting
chamber and obtaining counts in repeated intervals. The result of the process is a
calibration equation which is used to convert instrument readings to actual activity
levels. Analyses of 1991 and 1992 calibration data yielded several recommendations
for procedure improvement.
Related Statistical Work
Developed code to do Box-Cox transformations of data to achieve approximate normality in order
to perform further analyses. The code allows for left or right censoring. Data on
analyte concentrations which may be censored due to detection limits is analyzed
to determine which transformation is appropriate. Once the decision is made any further analyses
is usually performed as if the transformation were known and not estimated.
Explored a quasi-nonparametric procedure for calculating upper tolerance regions for populations
with support on the nonnegative real numbers. The procedure is based upon the bootstrap
and appears to lend itself to application to censored data. When sample size is large enough the traditional nonparametric
tolerance bounds can be derived from the
order statistics. The quasi-nonparametric procedure can be used when the sample size
is too small for the traditional procedure. The tolerance limits derived take the
form of a multiple of a quantile estimate and allow for the possibility of exceeding the
sample maximum. The multiplicative factor is based on bootstrapped quantiles of order
statistics, though no resampling is required. This procedure is presented in "Quasi-nonparametric upper
tolerance regions based on the bootstrap" by Paul S. Horn in 1992
Communications in Statistics Theory and Methods 21
, pages 3351-3367.
One goal of some of the ER work was to determine reference concentrations for groundwater
constituents useful for characterizing extent and severity of site-related contamination,
for identifying contaminants of concern in baseline risk assessment, and for developing and evaluating the suitability
of groundwater remediation criteria. A
constituent reference concentration is defined as the 95% upper tolerance bound for
groundwater unaffected by contamination. Calculation of reference concentrations
is hampered by complications such as left-censoring (detection limits), limited number of wells,
and availability of software to handle repeat sampling at the wells and statistical
distributions other than the normal distribution. Although various approaches to
handling some of these complications have appeared in the literature, none solve all
of the problems when encountered at the same time. Have worked on a way to deal with
all of these problems.
The approach that appears to be most promising is a Bayesian analysis which is sampling-based
(Monte Carlo) methods. Under the Bayesian philosophy the population parameters such
as the mean and variance are not constants, but have distributions. The prior information about these parameters
in the form of statistical distributions (prior
distributions) and observed information in the form of data from a specified sampling
distribution (likelihood) are combined to form inferences about the parameters through
updated distributions (posterior distributions).
The sampling-based approach to Bayesian analysis involves computer simulation to perform
necessary integrations such as calculating the mean of a posterior distribution or
determining marginal distributions. The Gibbs sampler, a type of Markov chain Monte
Carlo (MCMC) method, is an iterative technique which replaces difficult calculations
with a sequence of easier calculations. One real advantage of this approach to the
current problem is that a complex model involving variance components (repeated sampling), censoring (detection limits),
and non-normality can be handled fairly easily. The
software to do these calculations, called BUGS, an acronym for B
ayesian analysis U
ampling, is freely-available on the network.
Used BUGS to do model fitting of Box-Cox transformed data. The transformation chosen,
determined using code refined last month, will be one that results in approximate
normality of the sampling distribution. Once chosen the transformation was treated
as known when doing further analyses. Estimates of parameters of the distribution of untransformed
data and Bayesian tolerance intervals were determined within BUGS, given the transformation
and the results obtained on the model fitted to the transformed data.
Publications and Presentations
"The Dynamics of Radionuclide Discharges from the Oak Ridge National Laboratory as
a Function of White Oak Creek Watershed Hydrology," M. F. Tardiff and D. A. Wolf,
poster at the Fourth National Technology Information Exchange (TIE) Workshop held
May 11-13, 1993 in Knoxville, Tennessee.
"Multivariate Statistical Techniques for Characterization of Groundwater Regimes,"
Dennis A. Wolf, Mark F. Tardiff, and Richard H. Ketelle, poster at the Fourth National
Technology Information Exchange (TIE) Workshop held May 11-13, 1993 in Knoxville,
Lee, R. R., Curtis, A. H., Houlberg, L. M., Purucker, S. T., Singer, M. L., Tardiff,
M. F., and Wolf, D. A. 1994. Site Characterization Plan for Groundwater in Waste Area Grouping 1 at Oak Ridge National
Laboratory, Oak Ridge, Tennessee
, ORNL/ER-248, Oak Ridge National Laboratory, Oak Ridge, Tenn.
Wolf, D. A., Methodology for Identifying and Characterizing Y-12 Groundwater Types
, presented to representatives of Department of Energy, State of Tennessee Department
of Environment and Conservation, and Environmental Protection Agency, held in Oak
Ridge on November 16, 1994 and presented to the MMES Risk Assessment Council on November 17, 1994.
Wolf, D. A., Tardiff, M. F., and Ketelle, R. H. Characterizing Groundwater at Oak Ridge National Laboratory Hazardous Waste Sites
, presented at the 5th International EnvironMetrics Conference held in Burlington,
Wolf, D. A., Tardiff, M. F., and Ketelle, R. H. Characterizing Groundwater at Oak Ridge National Laboratory Hazardous Waste Sites
, presented at the 1994 Annual Meetings of the American Statistical Association held
in Toronto, Ontario, Canada.
December 11-15, 1995, Merida, Yucatan, Mexico. SPRUCE III International Conference
on the theme Statistical Aspects of Pollution Assessment and Control. Presented paper
on December 11 "Identifying and characterizing groundwater regimes using routine
groundwater monitoring data," D. A. Wolf, M. F. Tardiff and R. H Ketelle
Collaborations with Peter Blau (Materials Science and Technology Division)|
The U.S. Department of Transportation, National Highway Traffic Safety Administration is supporting a project
to develop a standardized marking code for heavy truck foundation brake linings. This code is intended to assist
operators of truck fleets to select replacement brake linings and facilitate tracking new equipment linings.
Federal law requires all new heavy trucks sold to meet certain minimum braking requirements. However, there
are no such standards for replacement parts. One of the objectives of this statistical analysis was to develop
a simple rating system for brake linings based upon manufacturer-supplied test data.
An approach to rating the bake linings can be accomplished by summarizing test data for each material based
upon the relationship between retardation torque and friction force. The validity of that relationship would
allow a simplification of future material testing. Theory suggests that within each gross axle weight rating
(GAWR) and brake pressurized air chamber size (ACS) category the brake line material retardation force is a
linear function of air pressure. Data for several of the materials are plotted in the
each plot showing
the three replicates, the different brake lining samples representing each material. Line segments connect
the data points for individual replicates. Most of these materials show good repeatability and apparent
linearity, exceptions are perhaps AA30 where linear looks good but the replicates are not as consistent
and AA51 and AA23 where there may be some curvature.
Statistical analysis further suggests that a simplified linear relationship exists, namely retardation force
is directly proportional to air pressure. A material's proportionality coefficient C (in units of in-lbs/psi)
would be a natural measure for rating brake lining materials within category, the larger the constant the better
the material. Other analyses reveal that the rating system could effectively be given for each GAWR and that ACS
might be ignored. The available test data indicates that C increases with GAWR so that it would be appropriate
to have ratings that are GAWR-specific. If the results of this study are borne out in future work, a simplified
approach to brake lining testing might be warranted. That is, if the simple linear fit is appropriate, testing
at only a higher pressure would be needed for the most accurate measure of a material's C. Test resources might
be apportioned differently to increase the precision of the estimate of a material's average C or to check for model validity.
Collaborations with Michael Ryon, Glenn Cada and John Smith (Environmental Sciences Division)|
Fish may be subjected to turbulence stresses as they pass through turbines of hydropower systems; they may suffer
temporary loss of equilibrium or reduced ability to avoid predators. This laboratory study was designed to investigate
through video the short-term effects of turbulence on fish escape behavior. Qualitative and quantitative measures were
derived from the filmed responses to shock stimulus at time intervals following the stress event. Two species of fish
were used, striped shiner and fathead minnow. Each group of fish, target size 10, was exposed to one of the possible
stress levels and received a shock stimulus (rod strike on the side of the tank) at each six times since the stress.
The stress levels were 0, 10, 20, or 30 minutes of turbulence or a 2-minute dip in an anesthetic.
After each filming, the appropriate portion of the taped reaction was saved to a file and later analyzed to measure
the effects of turbulence stress. Various measures were used, some derived from the characteristic C-shape (see figure),
the behavioral response to a startling stimulus. The study was designed so that each fish could serve as its own control,
i.e. each fish was observed multiple times, twice before the stress and four times post stress. The study also included
another control, a group of fish not stressed but observed in the same fashion as the treated groups. The benefit of
repeated measurements is that we can estimate the components of response variance, components due to differences among
fish within a species and differences within a fish.
We wanted to learn from this laboratory study how best to conduct a field study at a larger scale where the investigator
will have less control over the conditions. The fish species chosen for this study were chosen out of convenience and
logistic necessity and as such are not of direct interest. Therefore we looked for a consistency of results in the two
species and not for differences between the species.
The statistical methods used to analyze the data were chosen to answer the following questions:
A. Is there a significant effect of treatment (turbulence stress and anesthetic) on fish response as compared to control?
B. Which proposed measures of response are sensitive to treatment effects?
C. Do the fish become desensitized to the shock stimulus over time during the course of repeated tests?
D. How quickly do the fish recover from stress?
E. What portion of a fish's overall response is due to differences among fish and to variation within an individual fish?
A standard predator preference test was also conducted with largemouth bass as the predators and
fathead minnows as prey. In this test design, groups of unstressed fish (controls) and stressed fish
were put in a tank with a predator. The stressed fathead minnows were exposed to turbulence or fish
anesthetic. The predator was allowed to eat half of the prey, and the data were evaluated to determine
whether predators consumed greater proportions of stressed minnows than control minnows. The
predation test indicated that exposure to the anesthetic resulted in significant predation in fathead minnows, but
exposure to turbulence did not. This pattern was the same as seen in fathead minnows using the startle
response (escape behavior) test.
For the sublethal stresses we applied, evaluation of changes in fish escape behavior yielded results
comparable to traditional predator preference tests. Because this fish behavior test is simpler and quicker
to conduct than predator preference tests, it shows promise as a useful technique for assessing indirect
mortality resulting from sublethal stresses.