United States
Environmental Protection
Agency
Office of Pesticides
and Toxic Substances
Washington, DC 20460
EPA 560/5-86-013 //
May 1986
Toxic Substances
Underground Motor Fuel
Storage Tanks:
A National Survey
VOL. I. TECHNICAL REPORT
-------
UNDERGROUND MOTOR
FUEL STORAGE TANKS:
A NATIONAL SURVEY
Prepared by:
Westat, Inc.
1650 Research Boulevard
Rockville, MD 20850
Battelle Columbus Division
Washington Operations
2030 M Street, N.W.
Washington, D.C. 20036
Midwest Research Institute
425 Volker Boulevard
Kansas City, MO 64110
Washington Consulting Group
1625 Eye Street, N.W.
Suite 214
Washington, D.C. 20006
for the:
Exposure Evaluation Division
Office of Toxic Substances
Office of Pesticides and Toxic Substances
U.S. Environmental Protection Agency
Washington, D.C. 20460
U.S. Environmental Protection Agency
Region V, Library
230 South Dearborn Street
Chicago, Illinois 60604
May 1, 1986
-------
UtS. Environments! Protection Agency,
DISCLAIMER
This report was prepared under contract to an agency of the
United States Government. Neither the United States Government
nor any of its employees, contractors, subcontractors, or their
employees makes any warranty, expressed or implied, or assumes
any legal liability or responsibility for any third party's use
of or the results of such use of any information, apparatus,
product, or process disclosed in this report, or represents that
its use by such third party would not infringe on privately owned
rights.
-------
AUTHORS AND CONTRIBUTORS
The research on underground storage tanks contained in this
report represents the joint efforts of several organizations and
many individuals. The project team met separately to refine the
study design, analyses, and approach to data interpretation. The
names of the principal authors and the contributions of the
various organizations are summarized below.
Westat. Inc. — Designed and selected the national sample of
establishments with tanks; conducted on-site data collection
from establishment managers; designed and conducted fuel
inventory survey; conducted statistical analysis of all
tightness test and questionnaire data; provided national
estimates of leaking tanks and leak rate; produced
tabulations, multi-variate models and interpretive analysis;
wrote sections of the final report and; provided overall
coordination and editing. Key Westat staff included:
Stephen K. Dietz LuAnn G. Ruther
Ralph DiGaetano Judith F. Strenio
Patricia A. Leydig Carmen J. Vincent
Midwest Research Institute — Designed and conducted an
^ evaluation of available tank system test methods; selected
the field tightness test equipment and developed a test
protocol; directed tank tightness tests; selected on-site
tank and establishment data; provided data reduction of tank
leak data; conducted test/retests for quality control;
contributed to statistical and interpretive data analysis;
and wrote sections of the final report on tightness testing.
Key Midwest Research Institute staff included:
Jairus D. Flora, Jr. Sharon K. Perkins
Marilyn J. Gabriel H. Kendall Wilcox
Clarence L. Haile
Battellef Columbus Division - Washington Operations — As
prime contractor for the survey design, quality assurance
and data analysis, Battelle provided coordination of Westat
and Washington Consulting Group contracts and wrote the
quality control section of the final report. Battelle's
principal contributor was:
Jean Chesson
-------
Washington Consulting Group — Provided design and analysis
of inventory reconciliation analysis; analyzed inventory
data; provided statistical assessment of reliability and
consistency of various inventory analysis methods to
determine leak status; and wrote the section on inventory
analysis of the final report. Washington Consulting Group's
senior analyst and author was:
David Cox
O.K. Materials; Double Check Company. Inc.; and Protanic,
Inc. — Provided test crews and equipment to conduct the
field testing. The principal personnel representing each of
the firms were:
Phil Farrell - Double Check Company, Inc.
William Purpora - Protanic, Inc.
Thomas Warden - O.K. Materials
Vista Research. Inc. — Assisted in the development of the
test method; analyzed initial field data to determine
performance characteristics; developed techniques for
ambient noise analysis; and assisted in the preparation of
technical documents. The key Vista Research contributor
was:
Joseph W. Maresca, Jr.
EPA. OTS. Exposure Evaluation Division — EPA staff played
an active role in guiding this study, from the conceptual
design through the test and analysis methods and quality
control checks on data accuracy. Principal EPA contributors
included:
EPA Task Managers: Carol Bass
Robert G. Heath
Michael R. Kalinoski
EPA Project Officers: Joseph J. Breen
Joseph S. Carra
Cindy Stroup
-------
ACKNOWLEDGEMENTS
Many individuals contributed their time and effort to this
very complex survey effort. Especially helpful and supportive at
every stage of the survey was Martin P. Halper, Director of EPA's
Exposure Evaluation Division. Valuable advice and input was
provided by Steven E. Way and David O'Brien of EPA's Office of
Solid Waste and Emergency Response. Support in obtaining
cooperation from survey respondents was provided by M. Elizabeth
Cox of EPA's Office of Enforcement and Compliance Monitoring.
Senior statistical guidance on leak status estimation was
provided by Morris H. Hansen and Benjamin Tepping of Westat.
John Michael, also of Westat, provided guidance and review of
regression and logistic modeling efforts. We wish to recognize
the work of Annett Nold of EPA's Exposure Evaluation Division
who, together with Debra Sarvela and Jon Chen of General Software
Corporation, are responsible for Appendix H on soil data. Our
thanks to Joan Blake of EPA's Exposure Evaluation Division for
her work in summarizing the many peer reviewer comments and the
final implementation of reviewer suggestions. There were many
others in EPA's Exposure Evaluation Division and Office of Solid
Waste and Emergency Response who provided valuable support and
assistance at various stages in the survey. Finally, we would
like to thank Sandy Gallagher and Joyce Jones for their long
hours of word processing in preparing the final report.
111
-------
TABLE OF CONTENTS
Section Page
LIST OF TABLES viii
LIST OF FIGURES xii
EXECUTIVE SUMMARY xiii
1 INTRODUCTION 1-1
I. Definition of Key Terms 1-2
II. Types of Tanks Covered by the Survey 1-3
III. Limitations of the Data Presented
in This Report 1-4
IV. Overview of the Report 1-8
2 CONCLUSIONS 2-1
I. Accuracy of Estimates 2-1
II. National Estimates 2-4
III. Leak Rates Under Operating Conditions .... 2-7
IV. Establishment Characteristics 2-9
V. Tank Characteristics 2-10
VI. Tank Characteristics Associated
with Leaks 2-11
VII. Operational Findings 2-12
3 QUALITY ASSURANCE APPROACH 3-1
I. Development of Sampling Frame and
Sample Selection 3-2
II. Preparation of Questionnaires and
Interviews 3-3
III. Interviewing 3-3
IV. Data Tracking 3-5
V. Selection of Establishments for
Tightness Testing 3-6
VI. Tightness Testing 3-6
VII. Data Handling and Management 3-10
VIII. Data Analysis 3-11
4 SAMPLE DESIGN, ESTIMATION OF SAMPLE WEIGHTS
AND VARIANCES 4-1
I. Scope of the Survey 4-2
II. Sample Design and Site Selection 4-4
III. Establishment Frame Construction
and Sample 4-7
IV. Subsample of Eligible Establishments for
Physical Tank Tightness Tests 4-13
V. Calculation of Final Sample Weights
and Variance Estimation 4-14
-------
Section
FIELD PROCEDURES — QUESTIONNAIRE AND
INVENTORY 5-1
I. Westat Screening Procedures 5-1
II. Data Collection Procedures 5-2
III. Field Interview Data Collection
Statistics 5-4
IV. Followup Procedures 5-7
TANK TESTING FIELD PROCEDURES 6-1
I. Pre-test Preparations 6-1
II. Site Description Information 6-3
III. Tank System Tightness Testing 6-3
IV. Environmental Data Collection 6-13
V. Tightness Testing Field Experience 6-14
TANK TESTING DATA REDUCTION AND QUALITY
ASSURANCE (RETEST) RESULTS 7-1
I. Data Collection and Reduction 7-1
II. Retest Results 7-6
STATISTICAL ANALYSIS OF LEAK DATA AND LEAK
STATUS DETERMINATION 8-1
I. Total Measurement Error 8-2
II. Adjusting Measured Leak Rates to
Account for Test Pressure 8-3
III. Determination of Leak Status 8-6
IV. Assessing the Utility of Testing for
Leaking by Filling Tanks to Capacity ... 8-12
V. Assessing the Contribution of
Distribution Line Leaks to Measured
Tank System Leaks 8-14
VI. Percent of Tank Systems Leaking Under
Operating Conditions 8-16
STATISTICAL ANALYSIS 9-1
I. National Estimates of the Numbers of
Underground Motor Fuel Storage Tanks
and Establishments with Underground
Motor Fuel Storage Tanks 9-1
II. Characteristics of Establishments With
Underground Motor Fuel Storage Tanks ... 9-7
III. Characteristics of Underground Motor
Fuel Storage Tanks 9-16
IV. Leak Status of Underground Motor Fuel
Storage Tank Systems 9-38
V. Leak Rates of Underground Motor Fuel
Storage Tank Systems 9-56
VI. Statistical Associations of Leak Status
and Leak Rate with Other Variables 9-70
VI
-------
Section Page
10 INVENTORY RECONCILATION TECHNIQUES 10-1
I. Introduction 10-1
II. Methods and Data 10-3
III. Comparison of Methods 10-6
IV. Quality Control Samples 10-13
V. Conclusions 10-16
Appendices (Under separate cover)
A SAMPLE DESIGN AND ESTIMATION OF WEIGHTS AND
VARIANCES A-l
B SURVEY PROCEDURES AND ELIGIBILITY AND
RESPONSE RATES B-l
C DEVELOPMENT OF A TANK TEST METHOD C-l
D TANK TESTING DATA REDUCTION AND STATISTICAL
ANALYSIS LEADING TO LEAK STATUS DETERMINATION .. D-l
E INVENTORY RECONCILIATION METHODS E-l
F DATA COLLECTION FORMS AND MATERIALS Fla
G NATIONAL UNDERGROUND STORAGE TANK SURVEY
NATIONAL SAMPLE OF FARMS G-l
H ENVIRONMENTAL DATA COVERAGE H-l
I MULTIVARIATE ANALYSIS 1-1
VI1
-------
LIST OF TABLES
Table page
2-1 Estimates, by type of establishment, of the number
of underground motor fuel storage tanks and the
number of establishments with underground motor
fuel storage tanks in the continental U.S 2-2
2-2 Estimated number and percent of tank systems
judged to be leaking under test conditions by
establishment type 2-5
2-3 Estimates by establishment type of mean and
median leak rates among tank systems judged to
be leaking under test conditions 2-6
2-4 Field interviewing and inventory response rates .. 2-14
2-5 Tank testing completion statistics 2-16
4-1 Selected SIC codes for fuel tank establishment
frame 4-3
4-2 Six regions for the National Survey of
Underground Motor Fuel Storage Tanks 4-5
4-3 Initial sample sizes for fuel establishment,
large establishment, and farm samples by
survey region 4-11
4-4 Number of eligible cases for fuel establish-
ments, large establishments, and farm samples
by survey region 4-12
4-5 Summary of business and government establishment
subsample for tank tightness testing by region ... 4-15
5-1 Field interviewing and inventory status
statistics 5-5
5-2 Problems encountered in inventory recordkeeping .. 5-8
6-1 Technical problems summary 6-16
7-1 Retest results (volume change rates measured
under test conditions, not adjusted to operating
pressures) 7-7
8-1 Estimated percentage of underground motor fuel
storage tank systems judged to be leaking under
test conditions in the U.S., business and
government sectors, using statistical tests 8-9
Vlll
-------
LIST OF TABLES (Continued)
Table
8-2 Estimated percentage of underground motor fuel
storage tank systems judged to be leaking under
test conditions in the U.S., business and
government sectors, based on NFPA .05 gallons
per hour criterion 8-11
9-1 Estimates of the number of underground motor
fuel storage tanks and the number of establish-
ments with underground motor fuel storge tanks
in the continental U. S 9-2
9-2 Estimates, by survey region, of the number of
underground motor fuel storage tanks and the
number of establishments with underground
motor fuel storage tanks in the continental U.S. .. 9-4
9-3 Estimates, by type of establishment, of the
number of underground motor fuel storage tanks
and the number of establishments with under-
ground motor fuel storage tanks in the
continental U. S 9-5
9-4 Estimates of the number and percent of estab-
lishments with underground motor fuel storage
tanks across establishment types for the six
survey regions 9-8
9-5 Estimates, by survey region, of the percent of
establishments with underground motor fuel
storage tanks with selected characteristics 9-10
9-6 Estimates, by establishment type, of the
percent of establishments with underground
motor fuel storage tanks with selected
characteristics 9-11
9-7 Estimates of the number and percent of
underground motor fuel storage tanks across
establishment types for the six survey regions ... 9-17
9-8 Estimates, by survey region, of the mean and
median age (in years) of underground motor
fuel storage tanks 9-18
9-9 Estimates, by establishment type, of the mean
and median age (in years) of underground motor
fuel storage tanks 9-19
9-10 Estimates, by survey region, of the mean and
median capacity (in gallons) of underground
motor fuel storage tanks 9-20
IX
-------
LIST OF TABLES (Continued)
Table Page
9-11 Estimates, by establishment type, of the mean
and median capacity (in gallons) of underground
motor fuel storage tanks 9-21
9-12 Estimates, by survey region, of percent of
underground motor fuel storage tanks that
stored specified fuel types within the prior
year 9-22
9-13 Estimates, by establishment type, of percent of
underground motor fuel storage tanks that
stored specified fuel types within the prior
year 9-23
9-14 Estimates, by survey region, of the percent of
underground motor fuel storage tank systems
with selected installation characteristics 9-24
9-15 Estimates, by establishment type, of the
percent of underground motor fuel storage tank
systems with selected installation character-
istics 9-26
9-16 Estimated number and percent of tank systems
judged to be leaking under test conditions
within each survey region 9-40
9-17 Estimated number and percent of tank systems
judged to be leaking under test conditions
within establishment types 9-41
9-18 Estimates by survey region and establishment
type of percent of underground motor fuel
storage tank systems judged to be leaking
under test conditions 9-42
9-19 Estimated number and percent of tank systems
judged to be leaking under test conditions
within tank age categories 9-45
9-19A Estimated number and percent of tank systems
judged to be leaking under test conditions
by age and material of tank construction 9-46
9-20 Estimated number and percent of tank systems
judged to be leaking under test conditions
within tank capacity category 9-47
x
-------
LIST OF TABLES (Continued)
Table Page
9-21 Estimates by fuel types stored during the prior
year of the number and percent of tank systems
judged to be leaking under test conditions 9-48
9-22 Estimated number and percent of tank systems
judged to be leaking under test conditions
for single tanks and tank systems in manifolded
systems 9-49
9-23 Estimated number and percent of tank systems
judged to be leaking under test conditions for
tank systems with selected installation
characteristics 9-50
9-24 Estimates by survey region of the mean and
median leak rates among tank systems judged to
be leaking under test conditions 9-59
9-25 Estimates by establishment type of the mean
and median leak rates among tank systems judged
to be leaking under test conditions 9-60
9-26 Estimated mean and median leak rates among tank
systems judged to be leaking under test
conditions within tank age categories 9-62
9-27 Estimated mean and median leak rates among tank
systems judged to be leaking under test
conditions within tank size categories 9-63
9-28 Estimates by fuel types stored during the prior
year of the mean and median leak rates among
tank systems judged to be leaking under test
conditions 9-64
9-29 Estimated mean and median leak rates among tank
systems judged to be leaking under test
conditions for tanks in manifolded systems
and single tanks 9-65
9-30 Estimated mean and median leak rates among tank
systems judged to be leaking under test
conditions for tanks with selected installation
characteristics 9-66
9-31 Simple correlation of leak status and leak rate
with explanatory variables 9-71
10-1 Sample sizes for pairwise comparisons between
methods 10-4
XI
-------
LIST OF TABLES (Continued)
Table
10-2
Paqe
10-3
10-4
10-5
10-6
10-7
Number and percent of tank systems judged to
be leaking, judged not to be leaking, and
providing inconclusive results for inventory
methods and tightness testing
Comparison of EPA inventory reconcilation
method with Warren Rogers Associates inventory
reconciliation method
Comparison of EPA inventory reconciliation
method with tightness testing ,
Comparison of Warren Rogers Associates
inventory reconciliation method with tightness
testing
Simulated quality control inventory data
Inventory analysis of quality control samples
10-7
10-10
10-11
10-12
10-14
10-15
LIST OF FIGURES
Figure
6-1 Petro-Tite test equipment
6-2 Petro-Tite line test equipment
6-6
6-11
XII
-------
EXECUTIVE SUMMARY
Background and Objectives - The U.S. Environmental
Protection Agency's Office of Toxic Substances (OTS) conducted a
national survey of underground motor fuel storage tank systems.
This study was conducted in support of the EPA's Office of
Underground Storage Tanks, which has responsibility for
implementing the requirements of the 1984 Amendments to the Solid
Waste Disposal Act. The results of the survey are presented in
this report.
The primary objectives of the national survey were to
provide estimates of: (1) the total number of underground motor
fuel storage tanks; (2) the number of establishments with
underground motor fuel storage tanks; (3) the number of tanks
that leak; and (4) characteristics of tanks and tank
establishments. These tank and establishment characteristics
were analyzed in a search for possible correlations with leak
status (i.e., whether or not a tank system leaks) and leak rate.
In addition, OTS conducted an evaluation of the use of inventory
reconciliation analysis as an indirect method of detecting and
measuring leaks.
Scope - The target universe included the following kinds of
establishments if they had motor fuel stored in underground
tanks: (l) fuel-related establishments, including business,
government and military (establishments which store fuel for
retail sales or transportation services including gas stations,
trucking companies, auto dealers, marinas and other industry
groups using or dispensing motor gasolines, diesel fuel, aviation
gasoline, and jet fuel); (2) large establishments in non-fuel-
related industries (establishments which store fuel for purposes
such as company vehicles and private fleets); and (3) farms. The
Xlll
-------
survey excluded: tanks storing heating oil, used motor oil,
chemicals, hazardous wastes and sewage; tanks which are above or
partially above ground, abandoned or nonfunctional tanks; private
non-business tanks; bulk storage tanks; and tanks which do not
dispense fuel to end users.
Data Collected - Survey data collection was conducted with a
national probability sample of 890 establishments, with a total
of 2,445 tanks. A subsample of 218 establishments was selected
for physical tank testing, and at those sites there were 433 tank
system tests that yielded conclusive results.
Three different primary data collection efforts were used in
this survey. In-person interviews were conducted with tank
establishment operators in order to collect a variety of
information such as the type of business, type of fuel stored,
number of tanks, and tank characteristics (such as capacity, age,
material of construction). The second type of data collection
involved fuel inventory data which were provided by establishment
operators and analyzed to evaluate inventory reconciliation
techniques as an indirect method of detecting and measuring
leaks. The third data collection effort involved physical tank
system tightness tests at a representative subsample of
establishments.
Response Rate - A rigorous quality assurance program was
implemented at every stage of the survey. Response rate for the
interview with tank owner/operators was 99 percent, which is very
high. The tank testing phase achieved an excellent cooperation
rate of 95 percent; even after allowing for untestable tanks, the
tank testing response rate remained at a high level of 85
percent. For the inventory data collection, 78 percent of those
contacted provided complete or partially complete data. However,
only 41 percent produced data that were sufficiently complete and
XIV
-------
accurate to be used in the inventory reconciliation analysis of
this study.
Terminology - Several terms are important to the
understanding of the survey's results and are thus defined here.
The estimates apply to the tank system which includes the
underground vessel together with all connecting distribution
lines, vent and fill pipes and connections. Manifolded tank
systems consist of two or more tank systems which are joined
together. Whether or not a tank system passed the tightness test
is determined by a statistical decision rule applied to the
physical measurement data. This decision rule involves the null
hypothesis that the tank system is tight. This test has a 5
percent risk of falsely declaring a test failure, and a 5 percent
risk of not detecting a failure at 0.10 gallons per hour. The
actual test failures could be due to product loss anywhere in the
tank system — in the vessel, lines, pipes, or bungs.
Major Findings - Following are the major findings of this
study. The estimates given are subject to sampling error and
nonsampling error. The ranges in parentheses following the
estimates represent 95 percent confidence limits for the
estimates due to sampling error. These national estimates are
for the contiguous United States.
1. There are an estimated 796,000 (503,000-1,090,000)
individual motor fuel storage tanks in the United
States.
o 158,000 (35-453,000) of these are on farms.
xv
-------
2. The above tanks are located at an estimated 326,000
(296,000-356,000) establishments.
o 79,000 (58,000-100,000) of these are farms.
o The estimated mean of number of tanks per
establishment is 2.4 (1.6-3.2), overall, which
varies from 1.9 for large establishments that are
not in a fuel-related business to 3.5 for gasoline
stations.
3. Under test conditions, an estimated 35 percent (30-40%)
of the non-farm underground motor fuel storage tank
systems, including manifolded systems, did not pass the
tightness test. This represents an estimated 189,000
(153,000-226,000) tank systems. Using a different test
criterion (i.e., the commonly used NFPA 0.05 gallon per
hour cutoff) rather than the statistical significance
test used above leads to a very similar estimate (33%)
of tanks not passing the tightness test with an
estimated 44 percent total classified as
"uncertifiable" (i.e., 44% = 33% test failures plus 11%
untestable tanks and inconclusive test results which
are also counted as uncertifiable in most commercial
tests). Of the physical tank system tests attempted, 5
percent were untestable with the method used because of
unusual system configurations or large interferences.
4. The percentage of fiberglass and steel tank systems
that did not pass the tank tightness test were about
the same. Steel tanks, which comprise an estimated 89
percent of all underground motor fuel storage tanks,
show little increase in the percentage of tank systems
not passing the test as they age except for the oldest
tanks (over 20 years), for which the percent increases
substantially. There is a much smaller sample of
fiberglass tanks, so no comparison by material was
possible for the tanks aged 20 or more years, but
fiberglass and steel tank systems have no significant
difference in percent not passing the test at
comparable ages. These findings should not necessarily
be interpreted as causal effects of age and tank
material. Such statistical associations could be
caused by other associated variables.
xvi
-------
5. This report presents many descriptive statistics on the
characteristics of underground tanks and the facilities
or establishments at which they are located. For
example, based on the national sample estimates:
o Thirty-one percent of establishments with
underground storage of motor fuels also store
waste oil underground;
o Fourteen percent of establishments have one or
more abandoned tanks on site;
o Seventy-eight percent of establishments used clean
sand, pearock, or peagravel to backfill around
tanks;
o Twenty-nine percent of establishments are required
to have tank operating licenses;
o Sixty-nine percent of establishments believe they
are insured for non-catastrophic leaks;
o Eleven percent of underground motor fuel tanks are
fiberglass;
o Twenty-one percent of tanks are installed
partially or completely below the water table;
o Twenty-three percent of tanks are in manifolded
systems; and
o The mean age of tanks is 12 years.
6. The statistical analysis did not identify any single
explanatory variable (such as age of tank, type of
material, or fuel type) that is strongly correlated
with tightness test results. Additional multiple
regression and logistic models were developed which
suggested the possible influence of a few variables,
but their ability to predict the test outcome was weak,
as described in the appendices of this report. Soil
characteristics were not among the variables analyzed
because they were not available in the data base during
this study. Soil data more recently developed by EPA
and General Software Corporation are described in the
appendices of this report.
xvn
-------
7. There is only limited agreement between inventory
analysis methods and tightness test outcomes on a tank-
by-tank basis. It is possible that a longer period
than 28 days of inventory data might improve the level
of agreement. While each of three inventory methods
provided roughly similar overall estimates for the
percent of tank systems that might leak, there were
substantial disagreements among the inventory methods
as to which tank systems leak.
o It is very difficult to obtain accurate and usable
inventory data. Owners and operators had trouble
following even simple inventory data collection
procedures. The 78 percent response rate was
achieved only after extensive followup efforts.
It is not that inventory control does not work, it
is just that the successful execution of it is
difficult to achieve.
o EPA feels that the failure of the inventory
analyses as part of the survey was a result of
human error and inconsistency and we do not view
it as a basic failing of inventory methods.
8. For the tank system tightness tests, EPA initially
conducted an extensive evaluation program to test
existing methods, then selected one of these, modified
it to improve its accuracy, and characterized its field
performance. The method used by EPA has stated
procedures to identify and correct for potential
interference problems which commonly occur in the field
and which can otherwise invalidate the test results.
These interferences include tank end deflection,
temperature effects, water table and vapor pockets.
With the modified method, EPA was able to detect a 0.10
gallon/hour leak with 95 percent probability while
correctly identifying a tight system with 95 percent
probability.
None of the existing test methods evaluated by EPA
could consistently and reliably achieve detection of
the 0.05 gallon/hour leak rate specified by the NFPA
329 "Recommended Practices for Underground Leakage of
Flammable and Combustible Liquids, 1983." This
conclusion is based entirely on the data collected
during EPA's evaluation program since supporting data
which had been requested from the test companies to
document their performance claims were not received.
xvi 11
-------
While some methods can provide reliable results under
some specific conditions, most of them do not take
definitive steps to deal with the commonly occurring
interference problems previously mentioned. EPA
believes that in general the field performance of
existing test methods could be improved by:
training field crews to identify interference
problems;
- developing stated procedures to deal with
interference problems; and
- increasing frequency and duration of data
collection.
In any case, it is important that those who must rely
on the results from these methods be informed about
their performance characteristics. If valid
performance data on a method do not exist, they should
be generated. If they do exist, they should be made
available to those who are potential users.
Simply put, EPA believes that there are problems with
existing tank system tightness tests. EPA believes
these problems are correctable and for this survey EPA
chose a method, modified it to deal with these problems
and was able to improve the accuracy over existing
methods.
xix
-------
SECTION 1
INTRODUCTION
The National Underground Storage Tank Survey was designed to
provide estimates of the number of underground motor fuel storage
tanks and the number of establishments with such tanks, the
number and percent of tank systems that leak, and characteristics
of tanks and tank establishments. Tank and establishment
characteristics were analyzed in a search for possible
correlations with leak status (whether a tank is leaking) and
leak rate. The survey sample was a national probability sample
of establishments in the U.S. (except Hawaii and Alaska) that had
underground motor fuel storage tanks (not abandoned).
The survey consisted of a series of information-gathering
procedures which included an in-person interview, inventory data
collection, tank tightness testing, and secondary data
abstracting. This report presents national estimates for
statistics based on data collected in the interview phase of the
survey, analysis of the inventory data collection, and results of
the tightness testing phase.
The tank and establishment characteristics data presented in
this report were collected through in-person interviews conducted
by Westat field interviewers using the "Underground Storage Tank
Survey Establishment Operator's Questionnaire" (see Appendix F
for a facsimile of this questionnaire). The results reported
here are based on interviews conducted during visits to 890
establishments. As a part of the tank tightness test fieldwork
(at a subsample of establishments), certain tank and
establishment characteristics were also collected by the Midwest
Research Institute tank tightness test field crews. This
1-1
-------
information was checked against interview data and discrepancies
resolved by telephone recontact and checking the hard data from
both sources. The tank system leak data in this report were
collected by on-site tank system and manifolded tank system tests
conducted by Midwest Research Institute field crews using a
modified PetroTite procedure.
I. DEFINITION OF KEY TERMS
For the purposes of this survey, an establishment is defined
as any site or location where underground storage tanks are being
used to store and dispense motor fuel for business, commercial,
government, and, in a few instances, farm purposes. The term
"tank system" refers to an individual underground storage tank
vessel plus the lines and equipment that are connected to that
vessel. At some establishments, two or more tanks are linked
together by piping in "manifolded tank systems." Manifolded tank
systems often present special data collection problems. For
inventory reconciliation and sometimes for tank tightness
testing, it was necessary to collect data at the "manifolded tank
system" level rather than for the individual tank systems. For
example, when two manifolded tanks have one meter, inventory data
must be collected for the manifolded system as a whole in order
to compare meter data to stick data. In physical tank tightness
tests, tanks in manifolded systems were not isolated for testing
when, for example, they were joined by inaccessible lines. In
the interview procedure it was possible to collect data (such as
age, size, construction material) on an individual tank basis.
Our analytical approach has been to report results at the
smallest unit of analysis, whenever possible. Thus, interview-
collected characteristics and national estimates of the numbers
and types of tank establishments and tanks (in Section 9) are
1-2
-------
reported for individual tank systems rather than for "manifolded
tank systems." Similarly, leak status and leak rate analyses
will be based on the smallest available unit tested. Thus, for
tanks in manifolded systems that were separated and tested
individually, leak rate and status will be reported for
individual tank systems rather than for the tanks combined into a
manifolded tank system leak rate and status. However, for those
manifolded tank systems that were not separated for testing, the
leak status and rate reported are the manifolded tank system test
status and rate. The text for each table defines the unit of
analysis used in the table.
II. TYPES OF TANKS COVERED BY THE UNDERGROUND STORAGE TANK
SURVEY
The Underground Storage Tank Survey was limited, for
practical and regulatory reasons, to underground tanks that store
and dispense motor fuel prior to end use by business, commercial
and government establishments. This limitation excludes tanks
used to store materials other than motor fuels such as
chemicals, waste-water, hazardous waste, heating oil, and used or
waste oil. Also excluded by definition are motor fuel storage
tanks that are at private residences, above-ground or partially
buried tanks, and all motor fuel tanks at bulk storage facilities
that do not dispense fuel to end users. Tanks that are abandoned
or empty were also excluded from consideration. Included within
the scope of the survey are tanks that are owned and operated by
private businesses, public and government institutions, military
facilities, and farms. The initial step of the data collection
effort was to determine, for a random sample of establishments,
whether they in fact had an active underground motor fuel storage
tank as defined above. If so, the establishment (and its tanks)
1-3
-------
were considered eligible for the survey, and the main
questionnaire was administered.
As a result, the sample includes such establishments as
gasoline stations, airports, marinas, rental car agencies, fleets
of trucks or company cars, bus companies, fire stations, parks,
police stations, and many other types of establishment.
III. LIMITATIONS OF THE DATA PRESENTED IN THIS REPORT
As in any research report of this type, there are
limitations in the study's scope and methods which should be
understood by all who interpret and use the results of the study.
The major limitations are summarized below as caveats which must
be kept in mind by the reader.
A. Sample Frame Limitations
Because of practical and economic considerations, the sample
was drawn from those establishments most likely to have the types
of underground tanks described above (Subsection II). All
establishment types and industries were covered except small
(less than 20 employees) businesses in non-fuel-related
industries. As a result, the study would not have counted any
underground motor fuel tanks in small businesses, private homes,
and less relevant industry sectors. In other words, it is
possible that the number of underground motor fuel tanks in the
nation is somewhat greater than our estimate. However, we would
expect roughly similar leakage experience in uncovered business
establishments, based on the relatively constant percentage of
tanks leaking across the different sectors studied.
1-4
-------
B. Owner/Operator Responses
The tank and establishment characteristics data presented in
this report were collected by Westat interviewers during an in-
person interview with establishment owners and operators. The
accuracy of these data is limited by the knowledge of the
responding owner/operator.
A substantial number of owner/operators responded "don't
know" to interview questions about certain tank characteristics
such as the age of the tank or the material of construction.
Because the information may prove useful in regulatory
development, we have included information on the percentage of
"don't know" responses when this was substantial.
C. Inventory Data
Reconciliation of inventory records received from
respondents was evaluated as a secondary, more economical method
of detecting tank system leaks and estimating tank system leak
rates. Inventory data were analyzed by Warren Rogers Associates,
Inc. (WRA), using proprietary inventory reconciliation analysis
software. Alternative methods were also explored. Some
limitations to the usefulness of the inventory data are related
to the ability of the owner/operator to accurately collect it.
Because many tank managers do not normally maintain such
inventory records they often produced error-prone data, which
could not be analyzed. This occurred frequently in
establishments which had fairly inactive tanks (fuel was
dispensed only once or twice a week) and when the volume of fuel
in a tank was very low. Very accurate measurements were needed
for the WRA analysis. A less demanding analytical protocol might
1-5
-------
have been able to use a higher proportion of the inventory data
received.
D. Line, Vessel, and Equipment Leaks
Based on tank test methods used in this study, it was
generally not possible to distinguish between leaks occurring in
the tank vessel and leaks occurring only in lines or equipment
such as fill pipes, manways, vent pipes, distribution lines,
joints and bungs. Leak tests of distribution lines in isolation
from their tanks were possible for about one-third of the tested
tanks found to be leaking. (A distribution line test was always
attempted but could not be completed in many cases.) The
distribution line leak data are analyzed in Section 8. Elsewhere
in this report, no distinction is made between tank vessel leaks
and distribution line leaks or other non-vessel leaks. A leak
anywhere in the system is reported as a tank system leak. Also,
for manifolded systems of more than one tank where the tanks were
not separated for testing, the entire system was tested and a
reported leak could be in any of the tanks or in any associated
line, pipe, fitting, joint, or other equipment.
E. Test Conditions Versus Operational Conditions
The tank tightness test conditions include some
circumstances which are not always present during normal tank
operations. Specifically, during the test tanks were overfilled
(i.e., tanks were filled as were the associated fill pipe and
additional testing apparatus to permit measurements) such that
the net pressure at the tank bottom was 4 psig. The test
procedure compensates for hydrostatic pressure when the water
table is above the bottom of the tank by increasing the height of
1-6
-------
overfill. As a result of the overfilling, tanks that leak only
near or at the top of the tank or in lines, pipes, or joints not
normally filled with fuel might not always (or might never) be
leaking in normal operations if the tank is kept less than full.
The impact of increased leak rate because of test pressure is
less of a problem than leaks at non-operational locations because
the test pressure is small, and test leak rates have been
adjusted downward to correct to typical operating pressures.
(Section 8 discusses the typical fill levels reported in the
interview phase and their effects on leak status statistics, and
also describes the leak rate adjustment procedures.)
F. Interpretation and Adjustment of Tank Tightness Test
Data
Many factors affect the reliability of tank tightness test
data. The most important factor is temperature effect. Because
the volume of fuel in a tank varies with temperature change, it
was necessary to measure temperature changes directly and adjust
results using a correction and smoothing process. These
adjustment procedures required careful engineering and
statistical review and data editing using engineering judgment to
rule out suspect data. Some introduction of error is possible in
such engineering judgments, but careful discarding of suspect
data increased the overall validity of the findings. (Section 7
describes the data reduction procedures applied to the raw data.)
G. Untestable Tanks and Unreliable Test Data
The primary purpose of the tightness testing phase of the
survey was to estimate the number and proportion of leaking tanks
and to estimate the leak rates of those tanks. The degree to
1-7
-------
which the test data fulfill the objectives of the survey is
limited by the number of tanks for which responses were obtained.
For approximately 10 percent of the tanks selected for testing,
the tests were unsuccessful because the tank was untestable or
the resulting test data were unreliable. Reasons for
untestability included plumbing and piping problems and other
installation factors, such as physical constraints on the placing
of the test equipment in the tank. Reasons for unreliable tests
included trapped vapor pockets in the tank vessel or lines, and
unexplained temperature variations. For an additional five
percent of the selected tanks, the leak rate could not be
measured due to the great size and speed of the leak (although
leak status was determined). Generally, leaks at a rate of three
or more gallons per hour under test pressure could not be
quantified by the test procedure.
IV. Overview of the Report
Chapter 2 presents the major findings from the interview and
tank tightness test data as well as operational findings of
interest in developing regulations. Chapter 3 describes the
quality assurance program, with results given as appropriate in
the subsequent chapters. Chapter 4 describes the sample
selection and estimation procedures, Chapter 5 gives the field
procedures for the questionnaire and inventory data collection,
and Chapter 6 describes the tank tightness test data collection.
Chapter 7 gives the tank testing data reduction process and
quality assurance results, and Chapter 8 describes further
statistical analyses applied which resulted in the final
determination of whether a tested tank was leaking. The actual
data are presented in tabular form in Chapter 9, together with
the findings of some analyses designed to search for possible
correlations between tank and establishment characteristics and
1-8
-------
leak status of tanks. Chapter 10 presents the findings on
inventory reconciliation techniques.
The Appendices provide further details supporting these
discussions.
1-9
-------
SECTION 2
CONCLUSIONS
I. ACCURACY OF ESTIMATES
A. Accuracy and Sampling Error
The major findings of this survey are the national estimates
made from the data. These are presented in tabular form, along
with statements based on the data. The numbers are statistically
unbiased estimates based on a national probability sample, and
represent a sample estimate of the result that would be obtained
from a census of the target universe in which standard
questionnaire data collection and physical tank testing was
conducted for all tanks and establishments in the target
universe. The size of the difference between sample results and
results from such a hypothetical census are measured by sample
variances estimated from the survey data. Thus, the accuracy of
the figures can be objectively assessed as far as sampling error
is concerned. Non-sampling error is discussed below.
Estimates are given together with 95 percent confidence
limits in parentheses. These confidence limits are based on the
sampling variances estimated from the survey data. (The
estimation procedures are discussed in Section 4 and Appendix A.)
The limits can be expressed as the following statements. For the
first entry in Table 2-1 for the total of all establishments,
which is 326 (296-356) thousand, one would say, "It is estimated
with 95 percent confidence that the number of establishments with
underground motor fuel storage tanks is between 296,000 and
356,000 establishments, with a point estimate of 326,000
2-1
-------
Table 2-1.
Estimates, by type of establishment, of the number of
underground motor fuel storage tanks and the number of
establishments with underground motor fuel storage tanks
in the continental United States (95% confidence bounds
in parentheses)
Number of tanks
Type of
establishment
Government
and military
Gas stations
owned by major
petroleum
companies
Gas stations
owned by other
companies
Other fuel-
related estab-
lishments
Large non fuel-
related estab-
ments (with
> 20 employees)
Total for busi-
ness and gov-
ernment estab-
listments
Total for farms
TOTAL
Number of
establishments
with tanks
(1,000's)
45
(29-62)
33
(26-41)
58
(50-67)
36
(30-43)
74
(55-93)
247
(220-275)
79
(58-100)
326
(296-356)
Number
of tanks
(1,000's)
98
(69-128)
118
(87-148)
204
(174-233)
77
(64-90)
142
(97-187)
638
(584-692)
158
(<453)
796
(503-1,090)
per establishment
Mean
2.2
(1.8-2.5)
3.6
(3.3-3.8)
3.5
(3.2-3.8)
2.1
(1.8-2.4)
1.9
(1.6-2.2)
2.6
(2.4-2.8)
2.0
(<5.0)
2.4
(1.6-3.2)
Median
2
—
3
^
3
2
"
2
3
•—
1
~
3
-
2-2
-------
establishments." This means that there is only a 5 percent
chance that the actual value falls outside of this range. When
an upper limit is given, as in the estimate of number of eligible
tanks on farms in Table 2-1, where the entry is 158 (less than
453) thousand, this indicates that the lower bound of the
confidence interval is a small number. The statement for this
estimate would be, "It is estimated with 95 percent confidence
that the number of underground motor fuel storage tanks on farms
is less than 453,000, with a point estimate of 158,000." This
means that there is only a 5 percent chance that the actual value
is greater than the upper limit.
B. Non-Sampling Error
As in any data collection effort, non-sampling error is also
present in these data. This type of error is not quantified in
the confidence intervals but has been explored and reported on in
several places in the report. Potential non-sampling errors
include deficiencies in the sampling frame, respondent errors,
physical test errors, and inventory recording and analysis
errors. In this survey, one potential non-sampling error
investigated in depth comes from the physical tank testing.
Several parts of the report discuss the test method. Section 1,
in particular, has reviewed the limitations in the interpretation
of the results which stem from the testing method chosen. First,
a leak detected by the test may represent a hole anywhere in the
system of tank vessel and associated lines, pipes and fittings,
or indeed a loose connection within this system. Second, it
cannot be definitively determined from the data where the
detected hole or loose connection is or when or whether a leak
occurs under operating conditions. Section 8 offers two relevant
pieces of information: Tanks do tend to be nearly or completely
filled at delivery (so that holes or loose fittings at or near
2-3
-------
the top of the tank would have occasion to leak in practice); and
when distribution line tests were possible, the measured rate of
the line leak accounted for very little of the measured tank
system leak (so that detected leaks do not appear to be in the
distribution lines). As is described in Sections 6 and 7,
factors which could lead to difficulties with the physical
testing such as uneven product temperature, change in
temperature, erratic measurements, vapor pockets, and tank end
deformation due to test pressure have been carefully accounted
for in the the test procedure and subsequent data reduction
process.
II. NATIONAL ESTIMATES
The major findings are given in Tables 2-1 through 2-3.
Number of Establishments with Tanks — Table 2-1
presents survey estimates of the number of underground
motor fuel storage tanks and the number of
establishments with such tanks, as well as the mean and
median number of tanks per establishment, by type of
establishment. The national estimate for the number of
tanks is 796,000 with 95 percent confidence bounds of
503,000 to 1,090,000. This total includes farms.
Since so few farms surveyed actually had underground
motor fuel storage tanks (20 out of a sample of 600),
further national estimates including farms could not be
accurately made and therefore are not presented. The
national estimate of business and government tank
establishments is 247,000 (220,000-275,000) and the
number of non-farm tanks is estimated to be 638,000
(584,000-692,000).
Percentage of Tank Systems Judged to be Leaking under
Test Conditions — Table 2-2 shows the estimated number
and percent of business and government tanks judged to
be leaking under test conditions by establishment type,
based on the physical tightness test results. Based on
tested tank systems which yielded valid test results,
an estimated 35 percent of tank systems are judged to
be leaking under test conditions, with 95 percent
confidence bounds of 30 to 40 percent.
2-4
-------
Table 2-2. Estimated number and percent of tank systems1'2 judged
to be leaking under test conditions by establishment
type (95% confidence bounds in parentheses)
Establishment type
Number of
tank systems
judged
to be leaking
(in 1,000's)
Percent of tank
systems judged
to be leaking
(of tanks with con-
clusive test results)3
Government and military
Gas stations owned by
major petroleum
companies
29
(5-54)
25
(11-38)
36
(16-55)
32
(19-45)
Gas stations owned
by other companies
Other fuel-related
establishments
Large nonfuel-related
establishments
Total
56
(40-71)
35
(25-45)
45
(19-71)
189
(153-226)
30
(22-37)
57
(43-71)
33
(18-47)
35
(30-40)
1In this table, tank test results are reported for individual tank
systems unless the tanks were tested as a part of a manifolded tank
system that was not broken apart. These manifolded systems are
included in the table.
2Does not include farm tanks.
3Excludes tank systems for which test results were inconclusive.
(Therefore the estimated number in this table, when divided by the
estimated totals in Table 2-1, will not give the percentages shown
here.)
2-5
-------
Table 2-3.
Estimates by establishment type of mean and median
leak rates among tank systems1'2 judged to be leaking
under test conditions (95% confidence bounds on the
means in parentheses)
Establishment type
Mean
adjusted
leak rate
(gph)5
Median
adjusted3
leak rate4
(gph)
Government and military
Gas stations owned by major
petroleum companies
Gas stations owned by other
companies
Other fuel-related estab-
lishments
Large nonfuel-related
establishments
0.26
(0.06-0.47)
0.42
(0.18-0.68)
0.24
(0.13-0.34)
0.45
(0.20-0.71)
0.25
(0.14-0.36)
0.27
0.29
0.28
0.32
0.14
Total
0.32
(0.24-0.39)
0.25
•'•In this table, tank test results are reported for individual
tank systems unless the tanks were tested as a part of a
manifolded tank system that was not broken apart. Results for
manifolded systems are included in the table.
2Does not include farm tanks.
3Leak rates of leaking tank systems were adjusted to operating
pressure.
Calculation of median adjusted leak rate includes tanks judged
to have unquantifiably large leaks.
Calculation of mean adjusted leak rate includes only those tank
systems judged to be leaking which had quantifiable leak rates.
2-6
-------
Percentage of Tank Systems Leaking Under Operating
Conditions — Under operating conditions, the
percentage of tank systems that are leaking might be
somewhat less. This could vary from 18 percent at a
random point in time to 29 percent at the time of
product delivery the way tanks are normally filled, and
to 35 percent at the time tanks are filled if they are
filled to capacity. (See further discussion in III
below.)
Leak Rates — Table 2-3 presents the mean and median
leak rate for tank systems judged to be leaking under
test conditions by establishment type for business and
government tanks. These leak rates have been adjusted
to typical operating conditions (see Section 8). The
mean leak rate for all business and government tanks is
0.32 gallons per hour with 95 percent confidence bounds
of 0.24 to 0.39 gallons per hour. This is based on
tank systems judged to be leaking which had
quantifiable leak rates. Some tanks showed leaks too
large to quantify so the estimated mean leak rate is
conservative.
Incidence of Underground Motor Fuel Tanks Among Various
Types of Establishments — The screening effort
revealed a low incidence of underground motor fuel
tanks for certain types of establishment. Twenty-four
percent (19-28%) of fuel-related establishments (other
than gas stations) have underground motor-fuel storage
tanks. Thirteen percent (9-16%) of large
establishments not in fuel-related establishments have
eligible tanks. Three percent (2-4%) of farms have
eligible tanks. However, as is seen in Tables 2-1 and
2-2, a substantial proportion of the tank and tank
establishment universe is found in these types of
establishments even though many such establishments do
not have underground motor fuel storage tanks.
III. LEAK STATUS UNDER OPERATING CONDITIONS
Certain features of the tank testing method are different
from typical operating conditions, especially the overfilling of
the tank during the test.
2-7
-------
It is certainly reasonable to ask whether some of the leaks
detected under test conditions might have been due to holes near
the top of the tank or in lines, pipes and fittings above normal
fill levels. Data from the survey reveal that it is common
practice to fill tanks to 100 percent capacity when product is
delivered. In fact, 100 percent was the modal value for this
variable, and the median of the reported average fill level was
83 percent of capacity. Thus, the data suggest that even holes
near the top of the tanks would be subject to leaking, at least
just after product delivery.
On the other hand, the average tank fill level just prior to
delivery had a median value of about 20 percent of tank capacity.
Therefore, as a rough approximation, a typical operating level
might be midway between the high and low point, or 52 percent of
capacity. If one were to further assume that holes were evenly
distributed between the top and bottom of the tank, then an
estimated 52 percent x 35 percent = 18 percent of the tank
systems would be leaking on the average at any point in time
under typical fill level conditions.1 Furthermore, using average
percent filled after delivery may be a conservative estimate of
operational fill levels. When asked about the maximum gallons
ever stored, most respondents reported 100 percent, and only one-
quarter were below 92 percent full.
This is a rough approximation which could be refined by
calculating highest and lowest fill levels for each tank
separately, and then computing the median and mean fill levels
as fuel is withdrawn. Fuel withdrawal rate could be assumed as
uniform over time or simulated from inventory data. Finally,
refinements could be made to account for the fact that the
assumption of uniform leak distribution over the surface of the
tank is not identical to uniform leak distribution over volume.
However, since actual leak distribution is unknown, such
refinements do not seem warranted at present.
2-8
-------
In summary, if we are willing to assume that holes are
uniformly distributed around the tank circumference (we have no
data to verify this assumption), we could calculate that:
o Approximately 35 percent of the tank systems would be
leaking if they were filled to capacity;
o If all tanks are ever filled to capacity during the
year, then an estimated 35 percent of the tank systems
in the country are leaking at one time or another
during a year;
o Approximately 29 percent (.35 x .83) of tank systems
are leaking just after the time of product delivery the
way tanks are normally filled; and
o Approximately 18 percent (.35 x .52) of the tanks are
leaking at a random point in time.
Based on a limited set of 43 leaking tank systems where it
was possible to test the leak status of distribution lines
separately, it was found that the distribution line leak rate
makes up a very small portion of total tank system leak rate.
Distribution line leaks made up a small portion of the total
system leak rate.
IV. ESTABLISHMENT CHARACTERISTICS
Descriptive statistics for establishments include:
Thirty-one percent (27-35%) of establishments with
underground storage of motor fuels also store waste oil
underground.
Fourteen percent (11-17%) of establishments with in-use
underground motor fuel storage tanks also have one or
more abandoned underground storage tanks on site.
2-9
-------
Seventy-eight percent (73-83%) of establishments used
clean sand, pearock or peagravel to backfill around
tanks (one-quarter of owner/operators did not know the
backfill material).
Twenty-nine percent (21-37%) of establishments are
required to have tank operating licenses (whether this
was required was not known by 16% of owner/operators).
Sixty-nine percent (64-75%) of establishments believe
they are insured for non-catastrophic leaks (22% of
owner/operators did not know the answer to this
question).
V. TANK CHARACTERISTICS
Other descriptive findings include:
The mean age of eligible business and government tanks
is 12 years (11-13 years). The mean capacity is 5,405
gallons (5,026-5,783 gallons).
Forty-two percent (37-46%) of business and government
tanks store unleaded gasoline, 33 percent (30-36%)
store leaded gasoline and 21 percent (17-26%) store
diesel fuel. The remaining tanks store aviation fuel,
jet fuel, gasohol or other products used as motor fuel.
Eleven percent (7-15%) of tanks with known construction
material are fiberglass.
Twenty-one percent (17-25%) of tanks with known
positions in relation to the water table are partly or
completely below the water table (tank owner/operators
do not know this status for one-third of tanks).
Twenty-three percent (18-27%) of tanks are part of a
manifolded system.
Five percent (3-6%) of tanks for which the
owner/operators knew whether cathodic protection was
installed do have such protection (tank owner/operators
did not know the answer to this question for 13% of
tanks; it is unlikely that such a system would work
well if the operator were unaware of its existence).
2-10
-------
Twenty percent (14-26%) of tanks were installed by the
owner/operators themselves (this is among the 54% of
owner/operators who knew the identity of the
installer).
VI. TANK CHARACTERISTICS ASSOCIATED WITH LEAKS
Statistics describing tank systems judged to be leaking
under test conditions include:
Fifty-seven percent (46-67%) of tank systems storing
diesel fuel are judged to be leaking under test
conditions, while 18 percent (9-26%) of tank systems
storing leaded gasoline are judged to be leaking under
test conditions. Thirty percent (26%-4l%) of tank
systems storing unleaded gasoline are judged to be
leaking under test conditions. These differences in
percent leaking by fuel type could be due to some other
variable associated with fuel type. No conclusion
should be drawn about the effect of fuel type without
further research.
Fifty-four percent (39-68%) of tanks in manifolded
systems are judged to be leaking under test conditions,
while 31 percent (26-36%) of single tank systems are
judged to be leaking.
Thirty-one percent (15-48%) of fiberglass tank systems
(i.e., tank systems in which the tank is made of
fiberglass although lines, pipes, and fittings may not
be) are judged to be leaking under test conditions.
This figure is quite similar to the proportions of
steel tank systems judged to be leaking, whether bare
(uncoated) with 32 percent (14-49%), or coated, with 38
percent (30-46%).
Steel tanks, which comprise 89 percent of all
underground motor fuel storage tanks, show little
increase in percentage of tank systems judged to be
leaking as they age except for the oldest tanks (over
20 years of age) for which the percent judged to be
leaking increases substantially to 58 percent (29-77%).
No fiberglass tanks over 20 years old were found in our
sample, so percent judged to be leaking cannot be
compared across material type for this age category.
Fiberglass and steel tank systems show similar
2-il
-------
proportions judged to be leaking for the younger age
categories. For example, among tanks less than five
years old, 38 percent (11-65%) of steel tank systems
are judged to be leaking and 26 percent (<77%) of
fiberglass tanks are judged to be leaking. The broad
confidence bounds are due to the small sample sizes (46
and 7 tank systems tested, respectively). Furthermore,
for tanks between five and 20 years old, the percent of
tank systems judged to be leaking under test conditions
are similar for fiberglass tank systems and steel tank
systems.
The statistical analysis did not identify any single
explanatory variable (such as age of tank, type of material, or
fuel type) that is strongly correlated with either leak status or
tank system leak rate. Additional multiple regression and
logistic models were developed which suggested the possible
influence of a few variables, but their ability to predict leak
status or leak rate was weak, as described in Section 9,
Subsection VI, and Appendix I. Soil characteristics were not
among the variables analyzed because they were not available in
the data base during this study. Soil data more recently
developed by EPA and General Software Corporation are described
in Appendix H.
VII. OPERATIONAL FINDINGS
A. Tightness Test Method Development
There were a number of possible tank system testing methods
commercially available at the time of the survey. OTS modified
an existing method in order to improve the reliability for the
survey (see MRI draft report, November 7, 1985 for OTS,
"Development of a Tank Test Method for a National Survey of
Underground Storage Tanks," which is summarized in Appendix C).
2-12
-------
B. Establishment Manager Cooperation and Inventory
Participation in the Field Interview Phase
Participation in the field interview phase of the survey was
nearly 100 percent overall. As indicated in Table 2-4, 99.3
percent of all eligible respondents completed interviews. The
highest response rate among the sample segments was among the
large establishments where 100 percent of the eligible
establishments provided interview data.
It is very difficult to obtain accurate and usable inventory
data. Owners and operators had trouble following even simple
inventory data collection procedures. The 78 percent response
rate was achieved only after extensive followup efforts. About
90 percent of these respondents required technical assistance to
collect the inventory data. This contrasts with the 60 percent
of owner/operators who responded "yes" to the questionnaire item,
"Do you reconcile your stick inventory with your book inventory?"
The lowest inventory data response rate was from the farm sample,
where only 35 percent of all eligible farms provided inventory.
Of all eligible respondents, 16 percent have not yet provided any
inventory. Problems that establishment operators encountered in
keeping inventory records are described in detail in Section 5.
C. Tightness Test Cooperation
The physical tank testing response (cooperation) rate was 95
p-iicent and complete, usable results were obtained from 85
percent of the subsample of tank systems and manifolded tank
systems. Test results were judged reliable in about 89 percent
of the tank systems and manifolded tank systems where tests were
2-13
-------
in
o>
m
iH
(1)
M
G
o
ft
w
0
rH
r*l
rH
0
4J
C
(1)
c:
•H
•d
c
^1
c
"i$
Q)
•H
£>
rH
Q)
G
•H
•d
rH
(1)
•H
fa
*
T
CM
0)
rH
XI
(d
rH
fd
4J
O
EH
*
W
•d i 4->
1 0) XI C
rH 4-> (d Q)
(U (d 4-> E
3 rH W £
fa (U (U U
rH -H
rH
w
1 4-»
0X1 £
Cn fd (l)
rH -P E
(d w f,
tJ
00
o
o
00
vo
F~
0
CM
in
4J
d
(U
J
w
•H
rH
XI
4-1
M
0)
(1)
rH
XI
•H
D1
•H
rH
0)
V|H
0
rH
(U
2
*
<
o\P
oo
*
o^
en
o\°
*3*
•
cn
CTv
o\o
O
O
H
o\°
m
o\
.— .
U
(1)
•rH
0) >
rH rH
X) 0)
•H 4J
cn c
•rl -H
rH
a) -d
Q)
MH 4J
O 0)
rH
| 1 p,
C E
CD O
O O
j_i
(U O
p, r]
"•^ •$
(1) W
4-* 4-1
fd C
rH (U
•d
0) C
u o
aw
0)
U rH
0)
•
CQ
dp
r^.
•
o
o\o
vo
•
o
o\°
0
o\°
m
— —
U
0)
fl) -H
rH >
XI rH
•rH (1)
•H C!
rH -rl
(1)
•d
M-l (1)
O W
r^
4-i m
c 0
0) rH
o
rH O
ft £
U
0) 4-»
4-» C
d
4-»-H
c
0) -d
o i U
M (U
0 M
4->
d Q)
0) i-H
^ .O
d
H
•
Q
dP
^
.
rH
dP
rH
dP
ro
.
rH
dP
0
ro
I
•rl
tyd
i 1 1
rH O
0) U
0)
MH rH
O
O
4-> 4-)
C
Q) -d
o ty
rH W
o ^
ftVH
^- (U
V
4J O
fd ,C
rH S
rH W
fd 4-»
U G
3 fl)
MH *d
(D C — -
rH O >1
ft rH
>i cn o
W (U 4->
O rH C
4-1 (U
(UrH C
> XJ-H
H
•
U
dP
rH
^o
rH
dP
.
vo
rH
dP
00
.
r-H
rH
dP
O
CM
f*
•H
4J
C!
0)
D1
•rH
rH
W
•d
CO
4J
(U
t5 rH
in o
•rl 4J
rH G
XI 0)
fd £>
4J fl
W-rl
(U
m G
O-H
C
(U 4->
U 0)
rH rH
(U
•
PM
dP
^*
,
•^T
dP
CO
CO
dP
a\
t~*
dP
in
rH
O
rH
X)
•H
,£ W
O W
•rH O
rC ft
•rH
M
O (U
M-l rH
cd
U]
4-* U
C 4->
0) G
f> E
W Q)
•rl S-|
rH 3
XI (Q
fd fd
4-> Q)
in E
V
HH rH
O O
4-1
£S
0 >
0 C
M-rH
(U
CM
•
O
.p
fj
(U
j2
w
•H
rH
XI
rd
4J
(0
0)
r*<
rH
rd
4J
•H
i-H
•rl
E
•d
G
fd
.p
f3
(U
2
rH
0)
o
cn
Cn
G
•H
•d
rH
o
d
•H
*
2-14
-------
attempted. Approximately 5 percent of the tank systems and
manifolded tank systems where tests were attempted were not
testable for technical reasons which are discussed in more detail
in Section 6. Table 2-5 presents the tightness test completion
rates for the survey.
The testing methods used in the survey required that the
tank be out of service for one day and be filled to capacity at
the start of the test. Difficulties in arranging for a fuel
delivery, scheduling an acceptable test time, or physical
problems with the tank and with its associated plumbing add
significantly to both the time required and the cost of physical
testing. Severe operational difficulties requiring excavation
were encountered in about 14 percent of the tanks. More details
on these problems appear in Section 6.
2-15
-------
Table 2-5. Tank testing completion statistics
Single tank
or manifolded
tank systems
Percent for which test
results were unreliable
or inclusive
Percent with reliable,
conclusive test results
6%
89%
100%
Total number
of individual
tanks at these
systems
A. Number selected for tightness
testing 484
1. Percent of out of scope of
survey1 0.8%
2. Percent at sites refusing to
participate 5.0%
B. Number of tests attempted 456
1. Percent untestable for
technical reasons 5%
561
0.7%
4.8%
530
5%
5%
90%
100%
Response rate for estimates of
the percentage of tank systems
that are leaking2
85%
86%
-'•Became out of scope after the interview phase (for example, went out
of business).
2These response rates are the number of reliable test results out of
the eligible cases selected. From the figures presented above, they
can be calculated as (0.89 x 456)/((1.0 - 0.008) x 484) = 0.85 and
(0.90 x 530)/((1.0 - 0.007) x 561) = 0.86.
2-16
-------
SECTION 3
QUALITY ASSURANCE APPROACH
Rigorous quality assurance (QA) procedures were applied to
all stages of the survey to ensure that the data are of known
quality. These procedures are detailed in two QA Plans1; one
covering the overall survey and one dealing specifically with
tank testing. This section provides a brief overview.
In addition to the procedures outlined here and detailed in
the QA plans, the survey was audited by EPA QA personnel. This
involved site visits, productivity and response rate monitoring,
and QA audits and reviews of home office processing procedures.
Data quality is assessed by its representativeness,
completeness, accuracy, precision and comparability. The sample
of establishments and their underground storage tanks must be
representative of the population of establishments and tanks
about which inferences are to be made. A complete, or almost
complete, set of data must be collected; otherwise, its
representativeness, accuracy and precision will be compromised.
Accuracy (lack of bias) and precision must be sufficiently high
so that confidence can be placed in the numerical value of the
results. The methods employed must be well documented and allow
Quality Control Procedures. National Survey of Underground
Storage Tanks. Quality Assurance Plan. Westat, July 12, 1985;
and National Survey of Underground Storage Tanks Draft Quality
Assurance Program Plan for the Office of Toxic Substances,
Midwest Research Institute, June 7, 1985.
3-1
-------
results. The methods employed must be well documented and allow
comparison of the results with other relevant studies. The
succeeding pages outline procedures used at each stage of the
study to assess and maximize data quality. Quantitative measures
of data quality are presented in the appropriate sections
throughout the report.
I. DEVELOPMENT OF SAMPLING FRAME AND SAMPLE SELECTION
The objective of the survey is to provide national and
regional estimates of the number and proportion of underground
motor fuel storage tanks that leak, and to investigate
characteristics of leaking tanks. To achieve this objective, the
information collected must be representative of storage tanks
throughout the country. Representativeness was ensured by
careful development of the sampling frames and methods of sample
selection.
The entire contiguous United States was divided into Primary
Sampling Units (PSUs) consisting of counties or groups of
counties with a minimum number of gas stations and other fuel-
related business establishments. A probability sample of PSUs
was drawn, and sample summary statistics were compared with
summary statistics calculated from the entire frame to check that
the random selection procedure had been applied correctly.
Section 4 and Appendix A detail the construction of frames
and sample selection for the fuel establishment, large
establishment and farm samples. In each case, checks were made
between frame and sample summary statistics.
3-2
-------
II. PREPARATION OF QUESTIONNAIRES AND INTERVIEWS
Versions of the questionnaire and inventory recording forms
were pretested, both informally and in extensive formal pretests.
The questionnaire, together with an instruction booklet, was sent
to the respondent prior to the interview to allow the respondent
to look up information not immediately available and thereby
improve the quality (accuracy and completeness) of the response.
Questionnaire packages were sent by certified mail to provide a
record of receipt.
Interviewers were carefully selected and underwent a five-
day training session that included instruction at both Westat and
MRI in Kansas City. They were provided with a detailed
interviewer's manual that explained project and administrative
procedures, and provided item-by-item specifications for the
questionnaire.
III. INTERVIEWING
One or more interviewers were assigned to each PSU. They
received call records and labels for each of their assigned
cases, and were to complete a questionnaire for each eligible
case. All survey materials for completed work were sent to the
home office weekly. Call records for ineligible as well as
eligible cases had to be returned with the reason for
ineligibility clearly marked on the call record. Interviewers
were required to make weekly progress reports by telephone.
3-3
-------
In addition to questionnaire administration, interviewers
were responsible for conducting a meter accuracy test, reviewing
inventory procedures with the respondent, and collecting other
information on the location and nature of the establishment. The
measured meter errors were used to adjust the meter sales figures
to give actual quantities dispensed each day. This, in turn,
reduces the incidence of false-positive results from inventory
reconciliation analyses when quantity dispensed is compared to
stick readings. Review of inventory procedures was aimed at
improving the respondent's ability to understand and correctly
complete the inventory recording forms.
Interviews from the field were checked for completeness,
given a final interview status code, and logged into an automated
receipt control system. Refusals to comply with this mandatory
survey were reported directly to EPA's Office of Enforcement and
Compliance Monitoring. Most refusals eventually complied, and
the overall response rate was 99.3 percent. Approximately 78
percent of the eligible establishments furnished complete or
partial inventory data. Difficulties in gathering inventory data
are discussed in Section 5.
Initially, replacements for ineligible (i.e., out of scope
because no below-ground tanks, no motor fuel, etc.) fuel
establishments were selected after their ineligible status had
been determined. An unexpectedly high ineligibility rate of
about 50 percent made this procedure impractical. Subsequently,
the lists were oversampled, and the interviewers screened for
eligibility in the field. This made replacement errors less
likely.
3-4
-------
IV. DATA TRACKING
An automated receipt control system was developed to
maintain a record of the status of each data type for each
establishment. The data types and activities logged into the
system include:
o Certified mail cards - date received;
o No tank certifications - date received;
o Interview status - Complete, Partially Complete,
Ineligible, etc., with appropriate dates;
o Contact name, mailing address and telephone number;
o Prompt call status - date of call to prompt for
inventory records;
o Inventory status - Complete, Partially Complete,
Refused, etc.;
o Tightness test flag - establishments selected for
tightness testing;
o Tightness test status - Complete, Partially Complete,
Refused, etc.;
o Map status - Availability of soil and hydrogeologic
data; and
o Final status - Computed from the above status fields.
The system enabled the survey to be monitored on a
continuous basis by providing document control and producing
various lists and reports. Discrepancy reports indicating
missing data types, such as missing inventories or missing "No
3-5
-------
Tank" certification, provided measures of the completeness of the
survey.
V. SELECTION OF ESTABLISHMENTS FOR TIGHTNESS TESTING
The frame was developed from the information in the receipt
control system and checked against manual records before drawing
the sample. Weighted and unweighted summaries of the sample were
calculated to check that the sample of 198 establishments was
drawn correctly from the 876 eligible business and government
establishments. All 20 farms were scheduled for testing.
Notification packages were sent to each selected
establishment with a request for a return receipt. Four to seven
days after the mailing, each respondent was called to verify that
the package had arrived and that the testing procedure was
understood.
VI. TIGHTNESS TESTING
A. Training Tightness Teams
To assure that testing conducted by the several tank testing
crews would produce accurate and reliable results, a formal
training program was conducted prior to the beginning of the
testing program. All of the crew members participating in the
training were employees of O.K. Materials, Inc. or Double-Check,
3-6
-------
Inc. and had been trained and certified by Heath Consultants (the
manufacturer of the testing equipment) prior to the training for
the national survey. The two main objectives of the training
were to assure that all crews would follow the same test
procedures, and to provide advanced tank testing training.
The MRI field data analysts who accompanied each tank
testing crew received a four day training session that covered
procedures for collecting data, preparation of site diagrams,
computer operations, and data transmission. The analyst's
training sessions included hands-on practice experiences in data
entry, data transmission and recordkeeping.
It was necessary to add six additional test teams at
approximately halfway through the testing program. The test crews
were subcontracted through Protanic, Inc, and were selected based
on their past performance records, and years of experience. The
added crews were trained in the testing protocols by on-site
trainers or by sessions held via telephone, and the test crews
were teamed with experienced MRI data analysts who monitored the
testing procedures. All data collected by the added test teams
received extra verification review by Protanic and MRI staff
prior to entering the data in the analysis files.
B. Instrument Calibration
The Petro-Tite equipment used to collect the leak rate data
includes a thermistor in a probe that is inserted into the fuel
tank to monitor fuel temperature and a thermal sensor box which
provides an absolute temperature measurement within three degrees
3-7
-------
Fahrenheit. The electronic circuit in the thermal sensor box was
checked each time the equipment was set up (on site) to assure
that the circuit detected a simulated temperature change to
within approximately 0.003 degrees Fahrenheit, before the
equipment was used in the test.
Thermistors, glass thermometers and barometers used to
collect ancillary environmental information such as air
temperature, barometric pressure, and surface and subsurface
temperature were calibrated at MRI prior to field use, and the
calibration data were entered as a part of the project file.
C. Field Inspections
Site visits to tank test operations were conducted by the
MRI project leader, the MRI quality assurance coordinator, the
field coordinator, and EPA staff. The purposes of the site
visits were to evaluate the test protocols, and to assess the
performance of the field crews.
D. Data Management and Analysis
A computerized receipt control file was used to track the
test data. As the data were processed and reported, that stages
of the process were entered into the receipt control file.
Weekly reports were printed for project management and reporting.
3-8
-------
The tank test data were entered into a portable computer in
the field, using a thoroughly tested data entry program. These
data files were transmitted to MRI by telephone, where they were
stored on disks. The hard copy of the data on the original
Petro-Tite sheets, the field copiegofvthe disk files, and the
data collected on ambient conditions was sent to MRI on a weekly
basis. The field disk files were compared to the transmitted
disk files using a computer utility program, in order to detect
any transmission errors. All of the data elements to be used in
subsequent leak analyses were printed out and hand checked
against hard copy.
The data file for each test was hand checked for measurement
problems and outliers, which were eliminated prior to the
analysis of the file to determine test status and leak rate.
Each tank test analysis was checked individually to ensure that
the formulas for calculating the leak rate and standard error
were correct. Leak rate, test status, and environmental data
were abstracted from hard copy and computer files, and coded,
keyed, key-verified, and edited prior to being merged with the
questionnaire analysis file.
E. The Retest Program and Results
Three types of retesting were carried out to investigate
different sources of variation in the test results. Back-to-back
retests (on the same day and by the same crew) were conducted to
investigate the stability of test results over time. A leak
simulation retest was used to check the accuracy of the test in a
situation with a constant leak of known rate. "Complete retests"
3-9
-------
were independent tank tests usually conducted on different days
with different crews from the original test, and involving
rescheduling the test and refilling the tank.
A detailed analysis and discussion of the retest data is
reported in Section 7 and Appendix D of this report.
VII. DATA HANDLING AND MANAGEMENT
Questionnaires were batched in groups of ten and tracked
through the entire survey processing operation. Rules for coding
and editing were published in a Coding Manual that was updated
periodically to include new codes. Coding problems were referred
to a supervisor for resolution. Inventory records were reviewed
prior to coding so that missing or incorrect information could be
recovered or corrected by a phone call to the respondent as soon
after receipt as possible. Approximately 90 percent of
respondents received prompt or data retrieval calls in an effort
to obtain complete and accurate data on both the questionnaires
and inventory records. All coded questionnaires were verified by
the coding verifier.
All data were key-punched and then key-verified by a second
operator and transferred by telephone link to EPA's NCC computer
facility at Research Triangle Park in North Carolina. The data
were machine edited to check that each data element is in range,
that skip patterns were correctly followed, and that answers to
related questions are consistent. Errors were corrected and the
edit program rerun until no more problems were found. Frequency
distributions and various tables were generated to check on data
3-10
-------
quality. These allowed outliers to be identified and checked to
ensure that they are "true" values.
Careful coding and data entry procedures ensure that the
final data files accurately represent the information collected
by the survey.
VIII. DATA ANALYSIS
National estimates of number of establishments with tanks,
number of tanks, percentage of leaking tanks and leak rates were
obtained using the methods described in succeeding sections.
Where appropriate, the estimates are accompanied by confidence
bounds or standard errors to indicate precision. Confidence
bounds are narrowest (i.e., the estimates are most precise) for
estimates based on data from the entire survey. Estimates based
on subsets of the data (e.g., data from individual regions or for
tanks of a particular age) have broader confidence intervals.
3-11
-------
SECTION 4
SAMPLE DESIGN. ESTIMATION
OF SAMPLE WEIGHTS AND VARIANCES
The national probability sample for the Underground Storage
Tank survey was drawn in a three-stage sample design that
involved sampling establishments from establishment frame lists
within 34 survey sites which had been sampled to represent six
survey regions. Data were collected from the sampled
establishments using several data collection techniques. All
sampled establishments were first screened to determine survey
eligibility, that is, whether they had an underground motor fuel
storage tank. At eligible establishments, the owner or operator
of the tank was interviewed in person and instructed in the
completion of 30-day inventory records. A sub-sample of the
eligible establishments was selected for physical tank tightness
testing.
This section reviews the target universe of the survey and
then describes the three stages of sampling: Primary Sampling
Units (PSUs) or survey sites; establishments for questionnaire
and inventory data collection; and the sub-sample of eligible
establishments for physical tightness tests. In brief, the
sample consisted of 34 PSUs in which 2,218 establishments were
sampled. Of these, 896 establishments were eligible for the
survey, (i.e., had underground motor fuel storage tanks that were
not abandoned) and 890 cooperated with the interview phase. Two
hundred eighteen were selected for physical tank testing, which
was accomplished at 202 establishments. The section concludes by
describing the methods used to calculate the final weights used
in making national estimates from the survey and to estimate the
sampling error of those estimates. Appendix A gives more details
4-1
-------
on these subjects, and Appendix G gives a detailed account of the
farm sample.
I. SCOPE OF THE SURVEY
The scope of the Underground Storage Tank survey was
limited, for practical and regulatory reasons, to underground
tanks used to store and dispense motor fuel for business,
commercial and government use. This limitation excludes
materials other than motor fuels that may be stored in
underground storage tanks, such as chemicals, waste water,
hazardous waste, heating oil, and used or waste oil. Also
excluded by definition are motor fuel storage tanks that are at
private residences, above-ground or partially buried tanks, and
all motor fuel tanks at bulk storage facilities that do not
dispense fuel to end-users. Tanks that are abandoned or empty
were also excluded from consideration. Included within the scope
of the survey are tanks that are owned and operated by private
businesses, public and government institutions, military
facilities, and farms.
As a result, our sample includes such establishments as
gasoline stations, airports, marinas, rental car agencies, fleets
of trucks or company cars, bus companies, and many other
establishments. For practical reasons for list building and
screening costs, small establishments (with fewer than 20
employees) in industries not judged to be fuel-related were not
included in the survey. Table 4-1 is a list of the industries
that were judged to be fuel-related. For these industries all
establishments were included in the listing process.
4-2
-------
Table 4-1. Selected SIC codes for fuel tank establishments frame
SIC code
4010
4110 +
4121+
4131 +
4140+
4151
4170
4210+
4231+
4469A
4511
4521+
4582A
4582B+
4583
5511 +
5521 +
5541 +
7512 +
7513 +
7519 +
7992 +
7997B+
Description
Railroads, switching and terminal companies
Local and suburban passenger transportation
companies (includes airport transportation,
ambulance and limousine services)
Taxicab companies
Intercity highway transportation services
Passenger transportation charter services
(includes bus charter, rentals and tours)
School bus companies
Passenger transportation terminal and service
facilities
Trucking companies
Motor freight terminals
Marinas
Air transportation, certificated carriers
Aircraft charter, rental and leasing —
non-certificated carriers
Airports
Aircraft maintenance services
Airport terminal services
Auto and truck dealers (new and used)
Used car dealers
Gasoline service stations
Passenger car rental and leasing agencies
Truck rental and leasing agencies
Utility and house trailer rental agencies
Public golf courses
Golf and country clubs
4-3
-------
II. SAMPLE DESIGN AND SITE SELECTION
A. Sample Design
The contiguous U.S. (forty-eight states plus the District of
Columbia) was divided into six survey regions based on broad soil
and climatic characteristics. Table 4-2 lists the states which
comprise each region. Six Primary Sampling Units (PSUs) were
drawn from each region, except for the Mountain region, where
four were drawn. The PSUs consist of counties or groups of
counties with a minimum count of gas stations and the fuel-
related establishments (see Table 4-1). They were sampled within
the survey regions on the basis of probability proportional to
this count.
Once the 34 survey sites (consisting of 76 counties) were
drawn, establishment lists for sampling were constructed for each
county. In order to construct the lists, the target universe was
divided into three sectors:
1. Fuel-related establishments — Establishments which by
the nature of their business are likely to have
underground motor fuel storage tanks. The industries
in this category are listed in Table 4-1 and include
gas stations, trucking companies, airports, marinas,
and others. Government and military establishments
with underground motor fuel storage tanks were also
part of this sector.
2. Large establishments (20 or more employees) in other
industries — Although the nature of their business
would not suggest the presence of underground motor
fuel storage tanks; by virtue of their size, these
large establishments may have such tanks.
3. Farms were listed and sampled separately — (See
Appendix G for detailed discussion of the farm sample.)
4-4
-------
Table 4-2. Six regions for the National Survey of Underground
Fuel Storage Tanks
1 — Northeast
Maine
New Hampshire
Vermont
Connecticut
Massachusetts
Rhode Island
New York
New Jersey
Pensylvania
Maryland
Delaware
Virginia
West Virginia
Washington, D. C,
2 — Southeast
Kentucky
Tennessee
Arkansas
Louisiana
Mississippi
Alabama
Georgia'
North Carolina
South Carolina
Florida
3 — Midwest
Wisconsin
Minnesota
Iowa
Missouri
Illinois
Indiana
Ohio
Michigan
4 — Central
North Dakota
South Dakota
Nebraska
Kansas
Oklahoma
Texas
5 — Mountain
Montana
Wyoming
Idaho
Nevada
Utah
Colorado
Arizona
New Mexico
6 — Pacific
Washington
Oregon
California
4-5
-------
Separate samples were drawn from the three frames thus
established, since they were expected to yield widely varying
eligibility rates.
The fuel establishment sample was drawn by region, with
equal probability of selection within each region. To reach the
target of 800 survey-eligible establishments in this sector,
1,618 establishments were sampled for screening. The large
establishment and farm samples were drawn on an equal probability
basis nationwide, six hundred establishments were drawn from
each of these frames and the survey-eligible establishments kept.
After the eligible establishments were determined, a sub-
sample was drawn for tank tightness testing. All eligible farms
were selected for this testing since so few (20) farms were
eligible. For the fuel-related and large establishments, the two
samples were combined and an equal probability sample drawn from
each region. All tanks at sub-sampled establishments were to be
tested.
B. PSU (Site) Selection
Once the six survey regions were defined (Table 4-2), a
master list of PSUs was developed. For each of the 3,111
counties in the contiguous U.S., several counts were developed.
The 1981 County Business Patterns (CBP) data base supplied
figures for the number of gas stations, other fuel-related
establishments, and establishments in other industries with 20 or
more employees. A report prepared by Versar for the EPA (Leaking
Underground Storage Tanks Containing Engine Fuels, draft, March
1984) supplied estimates for the number of gas stations on a
state-wide basis, based on figures from Petroleum Marketing News
(PMN). These counts included all retail outlets for branded
4-6
-------
gasoline, i.e., convenience stores and other outlets as well as
gas stations. The CBP county totals for gas stations were
adjusted upwards to sum to the PMN totals. These adjusted counts
were added to the CBP other fuel-related establishment totals to
get a fuel establishment count for each county. Minimum PSU
counts were established by region (so that a sampled PSU would be
sure to have enough establishments to list and sample). Counties
with fewer fuel establishments than these minima were grouped
together to form multi-county PSUs. The 3,111 counties yielded
1,362 PSUs.
Within each survey region the PSUs were sorted by urban
versus rural, then by state and finally by PSU measure of size
(count of fuel establishments). Six PSUs were selected from each
region (four in Region 5 — Mountain) with probability
proportional to measure of size. The resulting 34 sampled PSUs
are made up of 76 counties. Twenty-three PSUs are urban and
eleven rural, and together they form a probability sample
representing the entire contiguous United States.
III. ESTABLISHMENT FRAME CONSTRUCTION AND SAMPLE
A. Frame Construction
Since lists of establishments with underground motor fuel
storage tanks do not exist, it was necessary to create
establishment frame lists for each of the 34 PSUs. As described
above, the target universe of all establishments with underground
motor fuel storage tanks was divided into three segments. The
first segment consisted of establishments which, by the nature of
their business, were considered fairly likely to have such tanks.
This segment, called the "fuel-related establishments" segment,
4-7
-------
contained gas stations, trucking companies, bus services, auto
dealers, marinas, golf courses, airports and other industry
groups that use large amounts of motor fuel or dispense it to the
public. (See Table 4-2 for a list of these industries.) Also
included in this segment were government and military
establishments with underground motor fuel storage tanks.
The second sample segment, the large establishment segment,
consisted of establishments in all nonfuel-related industries
(i.e., those industries excluded from the first segment) that
have 20 or more employees. This segment was designed to provide
estimates of the number of large, nonfuel-related establishments
that have underground motor fuel storage tanks to service company
vehicles and private fleets.
The third sample segment consisted of farms. Recent census
of agriculture statistics indicate that about half of the more
than two million farms in the United States have on-farm motor
fuel storage, but no information existed on how much of this fuel
storage was in underground tanks. This segment was designed to
provide estimates of the number of farms that have underground
motor fuel storage tanks to service farm equipment.
To construct the first list of fuel-related establishments,
several sources and methods were used. A listing of all
establishments with a primary or secondary Standard Industrial
Classification code appearing on the list in Table 4-1 was
purchased from National Business Lists. By specifying firms with
a fuel-related SIC code as the secondary code, we included such
establishments as convenience stores which also sell gasoline.
This list was supplemented by adding any establishments with such
a code as their primary or secondary code appearing on the large
establishments list, purchased from another source. To complete
the fuel establishments sampling frame, lists of government
4-8
-------
(Federal, State and local) and military establishments with
eligible tanks were needed. The Department of Defense provided
lists of military tank locations in the sampled PSUs to the EPA.
The civilian government list was constructed using a telephone
contact and network approach for the government officials serving
the sampled PSUs at the local, State and Federal levels. The
frames for all 34 PSUs had about 34,000 entries.
The large establishment list was purchased from Dun and
Bradstreet. All establishments in the 76 counties with 20 or
more employees were purchased from the Dun's Market Indicators
list, a very complete business listing. As noted above, all
establishments on the purchased list with a fuel-related primary
or secondary SIC code were removed from the large establishments
frame and clerically compared with the existing fuel
establishments frame. If they were not already on that frame
they were added to it. About four percent of the final fuel
establishments frame came from the Dun and Bradstreet list. All
establishments with an agricultural SIC code were also removed
from the large establishments frame and added to the farm frame
if not already there. In this way, the particular establishments
were on the correct frame and duplication between frames was
ruled out. The final count for the large establishments frame in
the 34 PSUs was about 68,000 establishments.
The farm frame was provided to the EPA by the U.S.
Department of Agriculture (USDA). As noted, it was supplemented
by the (very few) farm establishments found on the purchased Dun
and Bradstreet list of large establishments. About 31,000 farm
owners and operators were listed on the farm frame for the 34
PSUs.
4-9
-------
B. Establishment Sample Draw
The fuel establishment sample was drawn by survey region.
The total sample size of 800 eligibles was allocated among the
six regions in proportion to their count of fuel-related
establishments in the 1981 County Business Patterns data. Based
on initial field results of 50 percent eligibility, the target
sample sizes were approximately doubled. Within each region an
equal probability sample was drawn. Table 4-3 gives the counts
of sampled cases by region.
The large establishments and farms were both sampled on an
equal probability basis nationwide. Six hundred of each were
sampled, with only the eligibles remaining in the survey.
Because so little was known regarding incidence of underground
motor fuel storage tanks in these sectors, the initial sample
size was fixed rather than the final number of eligibles being
fixed.
Table 4-4 shows the results of screening the initial sample.
Eight hundred fuel establishments and 76 large establishments
were eligible for the survey (in business and operating an
underground motor fuel storage tank). Of these, 871 provided
questionnaire data. In addition, 20 of the 600 sampled farms had
underground storage tanks. As indicated in Appendix G, about
half of all farms report motor fuel storage, but only about 10
percent have more than 1,000 gallons of storage capacity. For
these small amounts of fuel, above ground storage is often a
reasonable alternative.
4-10
-------
Table 4-3. Initial sample sizes for fuel establishment, large
establishment and farm samples by survey region
Survey
region
1
Northeast
2
Southeast
3
Midwest
4
Central
5
Mountain
6
Pacific
Total
Fuel
establishments
449
415
325
194
75
160
1,61s1
Large
establishments
158
116
142
68
29
87
600
Farms
11
88
324
142
33
2
600
•'•Subsequent fieldwork determined that six of the sampled fuel
establishments were duplicates.
4-11
-------
Table 4-4. Number of eligible cases for fuel establishments,
large establishments, and farm samples by survey
region
Survey
region
1
Northeast
2
Southeast
3
Midwest
4
Central
5
Mountain
6
Pacific
Total
Fuel
establishments
225
197
161
92
42
83
8001
Large
establishments
21
18
13
7
4
13
76
Farms
0
3
5
5
4
0
201
1Five fuel establishments and one farm refused at the interview
phase of the survey.
4-12
-------
IV. SUB-SAMPLE OF ELIGIBLE ESTABLISHMENTS FOR PHYSICAL TANK
TIGHTNESS TESTS
Since so few farms screened had underground motor fuel
storage tanks, it was decided to physically test all such tanks
at all eligible farms. At the time of sample allocation, it was
estimated that there would be at most 50 tanks at eligible farms,
so that number was set aside for farm tank tests.
This left a target number of 450 tanks or manifolded tank
systems to be tested in the business and government sector (fuel
and large establishments). The 450 were allocated to the six
survey regions in the same proportions as the original
establishment sample allocation, except that a minimum number,
40, were allocated to Region 5, the smallest region, before
allocating the remainder to the other five regions. As each
region was completed by the interviewers, a list of eligible
government, fuel-related establishments and large establishments
in the g^iestionnaire sample was constructed, with the number of
tanks or manifold tank systems for each establishment listed. At
the time of sub-sampling it was assumed that a manifolded tank
system (two or more tanks connected by various lines and pipes)
would be physically tested as one unit. Therefore the sub-sample
was drawn on that basis. During the actual testing, some such
systems were isolated, and the individual tanks (and associated
lines) were tested separately. Thus, the total number of
possible tank tests is more than the number of tanks or tank
systems reported here but less than the total number of tanks at
these establishments.
The sub-sample of tanks to be tested was drawn on an
establishment basis, with all tanks at a given establishment
tested. The establishment list for a given region was sorted by
number of tanks or tank systems, then PSU, and then fuel-related
4-13
-------
and government versus large establishment. The target number of
establishments to select was calculated from the list, which
included initial sampling weights, and the target tank sample
size, using the weighted average number of tanks per
establishment. An equal probability sub-sample of establishments
was then drawn from the list. Table 4-5 shows the target number
of tank tests and the number of establishments sub-sampled with
the number of tanks or tank systems at the sub-sampled
establishments.
V. CALCULATION OF FINAL SAMPLE WEIGHTS AND VARIANCE ESTIMATION
A. Calculation of Final Sample Weights
1. Questionnaire Weights for Business and Government
Establishments
The final questionnaire weights for establishments sampled
with fuel-related SICs other than gas stations were based on a
ratio adjustment of the initial sample weights for all such
screened establishments to 1982 County Business Patterns (CBP)
counts of these SICs followed by a nonresponse adjustment among
the eligible other fuel-related establishments to account for the
few nonrespondents. (By the time final weights were being
calculated, the 1982 data were available.) The adjustments were
made by survey region. The ratio adjustment served to calibrate
the initial sample to CBP estimates of the number of
establishments with one of the fuel-related SICs in each region.
The sum of the weights of the eligible cases is the survey
estimate of the number of such establishments with eligible
tanks, by region. The nonresponse adjustment assures that the
4-14
-------
Table 4-5. Summary of business and government establishment
subsample1 for tank tightness testing, by region
Region
1
Northeast
2
Southeast
3
Midwest
4
Central
5
Mountain
6
Pacific
Total
Target
number of
tank systems2
to subsample
for business and
government sectors
115
110
90
50
40
45
450
Number of
business and
government
establishments
subsampled
51
47
38
23
17
22
198
Number of
business and
government
tank systems1
at subsampled
establishments
112
111
86
52
43
46
450
1A11 eligible farm underground motor fuel storage tanks were assigned
for tightness testing. There were 20 eligible farms with 35 tanks.
2In allocating and drawing the subsample of establishments for
tightness testing, a manifold tank system was counted as one unit.
Some such systems were separated for physical testing.
4-15
-------
weighted results based on questionnaires received equals the
estimates based on screening results.
The gas stations were weighted in the same way. First the
initial sample was ratio-adjusted by region to CBP totals for gas
stations (SIC code 5541). The sum of the weights of eligible
cases then estimates the number of gas stations with eligible
tanks, by region. A nonresponse adjustment again assures that
the weighted results based on questionnaires received will equal
the estimates based on screening.
The sample sector of establishments with 20 or more
employees in industries not otherwise sampled (the large
establishments) was weighted the same way as the gas stations and
other fuel-related industries. The CBP totals of establishments
of this size in all but the selected fuel-related SICs were used
for a region-by-region ratio adjustment of the initial sample.
The weighted eligible large establishments then estimate the
number of such establishments with eligible tanks in the country,
by region. Since all eligible large establishments participated
in the interview phase of the survey, no nonresponse adjustment
was needed.
No national statistics are currently available to estimate
the number of individual government agencies with underground
motor fuel storage tanks, which is the universe our frame was
built to cover. Therefore no ratio adjustments can be made.
Nonresponse adjustments were made to account for the small amount
of nonresponse among government establishments.
4-16
-------
2. Physical Test Result Weights. Business and
Government Establishments
After calculating final questionnaire weights for all
responding business and government establishments as described
above, the sampling weights for establishments chosen for
physical testing were adjusted to sum to the estimated totals for
four establishment types (government, gas station, other fuel-
related, and other industry) by region. This adjustment was made
by an iterative raking procedure,in which the weights were
adjusted first to regional totals, then to establishment-type
totals, then readjusted to regional totals, and so forth, until
no further adjustment was needed. (This took five and a half
iterations to achieve.)
A final adjustment was made for tank test result weights.
The weight for the individual tank or tank system test would be
equal to the establishment physical test weight, except that some
tanks were not tested. Thus, a "tank nonresponse" adjustment was
made to the tank weights to account for the untested tanks.
3. Farm Questionnaire and Physical Test Weights
Due to the distribution of farms within the survey regions
(both overall and in our sample) and the low yield of eligible
farms from the screening, the survey regions have been
consolidated into three areas for calculating final weights for
farms. (See Appendix G for a more detailed discussion.) These
are: (1) East (combines survey regions Northeast and Southeast);
(2) Midwest; and (3) West (combines survey regions Central,
Mountain, and Pacific). Total counts of farms for these areas
were obtained from the 1982 Census of Agriculture and used to
4-17
-------
form ratio adjustments for eligible farms. Due to one refusal
among farms, a nonresponse adjustment was also made.
Since so few farm tanks were tightness tested (21 of 35 —
most not tested were smaller than 1,100 gallons), no weighted
estimates will be presented for that data, and hence no final
weights were calculated for physical test results for farm tanks,
B. Variance Estimation
National estimates from the survey are based on a sample of
cases rather than a complete census of the nation's underground
motor fuel storage tanks, so they are subject to variability
termed sampling error. This is due to the fact that drawing
several samples would result in different sets of establishments
being interviewed and different national estimates. Since the
sample was drawn on a probability basis, it is possible to use
the survey data to estimate the magnitude of this sampling error.
Due to the complex nature of the sample design, this variance is
not easily expressed as a simple mathematical formula. It has
therefore been estimated by a more empirical approach.
The method of variance estimation used in this survey is
termed the jackknife approach. Essentially, a series of sub-
samples of the survey data known as replicates are created.
Using the same series of steps given above for the full sample,
each replicate is given weights which can be used to create
national estimates based on that replicate. The variance of the
replicate estimates of the statistic from the full sample
estimate estimates the sampling error of the statistic.
In this report the sampling error is generally reported in
terms of 95 percent confidence bounds. These are interpreted as
4-18
-------
being the numeric range which one can be 95 percent confident
includes the true value of the statistic. It is centered on the
full sample estimate and its width is determined by the estimated
sampling error of the statistic.
4-19
-------
SECTION 5
FIELD PROCEDURES — QUESTIONNAIRE AND INVENTORY
Interviewing fieldwork for the Underground Storage Tank
Survey began December 2, 1984. A staff of eight Westat field
interviewers was trained to collect data from establishments
selected in 34 PSUs nationwide. The interviewing phase concluded
on June 29, 1985. This chapter includes some details about the
field procedures, inventory data collection and followup
procedures used for the survey. Interview and inventory response
rate statistics are reported. The detailed summary of the
fieldwork is found in Appendix B.
I. WESTAT SCREENING PROCEDURES
The sample of 600 farms and 600 large establishments was
pre-screened for survey eligibility by telephone. Telephone
interviewers contacted the owner or operator of the farm or
business and asked if there were any underground tanks used to
store motor fuel at the establishment. All establishments that
could not be reached by telephone were included in the
interviewer assignment lists to be located and screened in the
field.
Establishments selected from the fuel-related sample frame
(such as gas stations, government facilities, and trucking
facilities) were screened for survey eligibility by the field
interviewers.
5-1
-------
After the screening procedures were complete, survey
materials were mailed out, beginning with all sampled
establishments in Survey Region 6 (West Coast). Each
establishment was sent a survey package (Appendix F) including an
introductory letter, a copy of the survey questionnaire, an
instruction booklet, and forms for collecting tank inventory
data. The owner or operator was to review the materials to
prepare for the next phase of the survey in which a Westat field
interviewer visited the establishment and conducted an in-person
interview with the respondent. Packages were mailed according to
the schedule of the field interviewer so that the respondent
received the materials two weeks prior to the interviewer's
arrival on site. This gave the respondent time to prepare for
the in-person interview.
II. DATA COLLECTION PROCEDURES
The first phase of fieldwork in a PSU was the field
screening of the fuel-related sample of establishments.
Establishments were found to be ineligible for the survey for
various reasons; most commonly, they had no underground motor
fuel storage tanks, they were out of business, they were out of
area or scope of the survey, or the only tanks at the
establishments were abandoned (i.e., out of service permanently).
In-person interviews where scheduled with the owners or operators
of all eligible establishments within each PSU. When an owner or
operator refused to participate in the survey, the interviewer
informed the respondent that EPA would be notified, and contacted
a Westat field director immediately. An attorney with EPA's
Office of Enforcement and Compliance Monitoring was then notified
by Westat and necessary action was taken. Questionnaire and
inventory refusal rates are discussed in Subsection III.
5-2
-------
After the field interviewer arrived at the establishment
site, the data collection procedures began with the
administration of the Operator's Questionnaire (Appendix F). The
respondent received a copy of this questionnaire in the survey
package and was to have completed it prior to the interview. The
questionnaire gathered basic data about the establishment, its
operating characteristics and its tanks.
The interviewer then reviewed the inventory forms and
procedures with the respondent. The inventory record-keeping
procedure involved taking and recording dipstick readings and
meter readings (if the dispensers were metered) for 30 days. Any
deliveries made during that time period were also recorded. The
interviewer checked to see that the inventory forms were being
filled out correctly and that all tank and meter numbers on the
forms and in the questionnaire corresponded to one another. If
previously collected inventory records were used, the interviewer
made sure 30 complete readings were provided.
Once all inventory sheets were reviewed and tank and meter
numbers verified, the interviewer checked the accuracy of all
dispenser meters using a five-gallon Certified Standard Weights
and Measures Calibration Can. Five gallons of fuel was pumped
into the test can and by reading the level of fuel according to
the measuring gauge on the front of the can, the interviewer was
able to determine the calibration ratio that would correct for
any error in the meter. The inventory records were adjusted
prior to inventory reconciliation analysis to account for the
meter error.
After the meter accuracy check, the interviewer measured the
diameters of all tank fill pipes, determined whether or not there
5-3
-------
were drop tubes present inside the fill pipes and, if present,
whether the drop tube was permanent or removable. This data was
collected to aid tank testing crews in preparing for tightness
tests at selected establishments.
Before leaving the site, the interviewer located the
underground storage tanks on U.S. Geological Survey maps, which
were provided for each PSU in the survey. The interviewer also
evaluated the overall status or attitude of the on-site interview
by answering debriefing questions.
III. FIELD INTERVIEW DATA COLLECTION STATISTICS
Table 5-1 contains data collection statistics for the field
interview portion of the survey. It covers statistics on
interview and inventory response and refusal rates.
A. Interview Response Rate
The interview response rate for this mandatory survey is
nearly 100 percent overall, as well as for each sample segment.
Out of 2,800 establishments contacted, 896 had underground motor
fuel storage tanks, and were therefore eligible for the survey.
Of those, 890 or 99.3 percent completed interviews. The highest
response rate among the sample segments was among the large
establishments, where 100 percent of the eligible establishments
provided interview data.
5-4
-------
o
•H
•P
-P
td
-P
w
-P
td
-P
u
(U
>
c
•H
td
iH
0)
-P
C
•H
•a
iH
0)
I
m
(U
rH
Xt
td
rH
(d
-P
o
EH
W
i -P
xi C
iH td a>
Q) -p E
d M .2
&4 W W
•H
rH
(A
1 -P
a) xt c
O> td o
M -P E
td w X!
iJ W M
•H
rH
w
E
iH
id
En
O O U3 O
rH O vo oo r-
rH rH
O VD VO \O
o
^o in
rH
CO VO O 0) -P
-H M C
EH a)
£ Q) S O
W s 0) M
-H ^ -H (1)
T) rH > ft
(U Xl M >H —
T3 -P o ft
£ E (d 6 w
3 3 j& 3 m
a ss 55 -P z PC!
• • • • t
< m u p w
0\°
m vo
•
a\
CT>
o\°
O O
O
rH
o\°
in rH
CT\
5
rH
H XI
0) -H
-P CP
C W -H
•H rH iH
rd OJ
•a w
0) d MH
-p m o
0) 0)
rH M 4->
ft £
e ^ a)
o n ft
,d Qj —
S -P
C (U
W -H 4J
-P td
C 4-1 SH
Q) O
T) rH
a n td
O 0) W
ft XI 3
W g MH
0\°
0 0
VO
0\°
in r-
w
(U
•0
d
rH ^
^ O (1) rH
r—
0\°
in vo
ro
i
C
0
0\°
r» -* o
. T^. ^.
rH rH
o\°
rH rH rH
m m
rH
H
>1
M (U
0 rH
-P X»
> ft D> C -H
)H W E -H
Q) 0) O rH
-P W O H iH C
d id i ft O
0) >H »H
!H O 13 (1)
-P C! ft
O C fd -^
rC Q)
5 > <1) <1)
C -P -P
W -H Q) (d
4-> rH iH
C MH ft
Q) O £ (U
•a o w
C M O C
O <1> O
ft x> x: a
w E -P W
> W -H
C rH rH
-H rd 0)
u
T> 3 MH
Q) MH O
C 0)
^ M -P
3 C
-P >i Q)
H
-P
C (U
W -H -P
-P rd
q 4-1 in
OJ 0
T( rH
C H (d
O 0) W
ft XJ d
W 6 MH
CD 3 0)
n z «
• •
ro ^
>t C QJ
)H O rH
O ft X)
-P W X! -H
C 0) O W
(U M -H W
> x: o
C >i 5 ft
••H rH E
O >H -H
T) -P O
in G MH 0)
O Q) H
O > W td
i
Xi O O H
5 O
!H >H -P
W 0) <1> C!
-P XI Xt
a) 3 3 c
-a z 53 -H
• •
J S
Q)
d
O
XJ
ft
0)
(U
-p
0)
x:
-P
c
__i
*M
Q)
-P
id
U
-H
l
I 1
ft
-d
Q)
XI
O
-P
•d
d
0
MH
0)
M
(U
^
U)
E
td
MH
0
.
T)
0)
ft
E
id
u
0) •
>-i U»
0) C
15 -H
£
U) 0)
E _l
0)
•*
1
•rt
U)
.
•d
0)
^_nl
r^
ft
E
id
w
a)
iH
(1)
-P
d
0)
E
w
•H
rH
X) .
td 0^
-p d
W -H
Q) d
0)
rH fl)
(1) iH
d o
MH W
oo *d
1 1
r^ r^
vo O
-•H
rH MH
CM
5-5
-------
B. Certification of No Tank Status
Since many of the sampled establishments had been selected
based on being in an industry expected to have underground motor
fuel tanks, a verification program was undertaken for
establishments which responded to the initial contact by stating
that they had no eligible tanks (i.e., non-empty, underground,
and storing motor fuel). For Region 6, the West Coast, which was
the first region fielded, the field interviewer went to each of
these establishments in person to visually confirm the statement
and to get the owner or operator's signature on the
"Certification Statement for Establishments Without Tanks" (a
copy of which appears in Appendix F). This experience showed
that those respondents initially stating they had no tanks were
correct. For the remainder of the survey, the personal visits
were not made to all such respondents, but they were all asked
(by mail) to sign and return the no tank certifications. Signed
statements were received from 80 percent of the "no tank"
ineligible respondents.
C. Inventory Response Rate
Nearly 78 percent of the eligible establishments have
furnished complete or partial complete inventory data. Even this
relatively low response rate (compared with other parts of the
survey) was achieved only after extensive edit and followup
efforts by Westat's survey staff and finally a stern warning
letter from EPA. Sixteen percent of the eligible establishments
have not yet provided inventory records. It was impossible for
4.5 percent of the eligible establishments to keep inventory
records.
5-6
-------
D. Problems Encountered in Recordkeeping
A majority of establishment operators were unable to provide
inventory data that was useable for inventory reconciliation
analysis techniques without technical assistance. Establishment
operators received detailed written instructions on inventory
recordkeeping procedures, and on-site training was provided by
the survey interviewers.
Nevertheless, almost 80 percent of the operators supplied
inventory data sets that were incomplete or incorrect, initially.
Extensive mail and telephone recontacts were made to the
operators to attempt to capture the missing information and
correct the problems. Eventually about half of the responding
establishment operators were finally able to provide inventory
records complete enough for analysis.
These experiences suggest that many establishment operators
lack skill, training and/or motivation to correctly follow
inventory recordkeeping procedures. Problems encountered with
inventory records are listed in Table 5-2.
IV. FOLLOWUP PROCEDURES
Followup procedures were implemented to complete interviews
which could not be completed during the time the interview team
was working in a PSU and to make sure as many eligible
establishments as possible returned useable inventory data.
A "clean-up" interviewer followed behind the field
interviewers to complete interviews that could not be scheduled
5-7
-------
Table 5-2. Problems encountered in inventory recordkeeping
Major problem areas
Percent of
establishments
with this problem
8.
9.
10.
11,
Meter sales and gallons used as
calculated from stick readings and
deliveries are equal (respondents
adjusted out the daily variances)
Did not provide inventory for the
total 30 days
Did not use the form correctly such
that data was unusable
Carried readings down from closing
to opening (tanks without meters)1
Did not provide enough complete
readings (some stick or meter
readings missing)
No inch-to-gallons conversions
provided
Inconsistent conversions (from
inches to gallons)
No meter readings provided
No stick readings provided
Problems with delivery records
Reported readings greater than
tank capacity
19%
15%
8%
7%
6%
4%
4%
3%
3%
2%
1%
tanks without meters, the inventory recording procedure
involves sticking the tank before and after each inactive
period. This procedure is counter-intuitive to most
respondents, who cannot understand the value of measuring and
recording what they regard as measurement error in the second
reading. Most respondents using "Tank Without Meter" forms
have, therefore, measured the level in the tank only after each
use (before each inactive period), and have (incorrectly)
carried the readings to the "after" columns.
5-8
-------
SECTION 6
TANK TESTING FIELD PROCEDURES
The field testing phase of the national survey required
collecting descriptive information from each test site, conduct-
ing tightness tests on the tank systems and associated piping,
and collecting ancillary environmental data. MRI managed the
field testing and data acquisition, developed and maintained the
test schedule, and served as the primary contact with the
establishment owner/operator (o/o). The three-person field teams
were comprised of an MRI field data technician and a two-person
tank test crew provided by a commercial tank testing firm under
subcontract to MRI. The tank test crews were provided by O.K.
Materials, Inc., Double Check Company, Inc., and Protanic, Inc.
This section describes the procedures used to accomplish the
field data collection. A more detailed description of the field
testing procedures may be found in MRI's Test and Analysis Plan.1
I. PRE-TEST PREPARATIONS
Preparations for field data collection at each establishment were
initiated soon after the site identification and survey
questionnaire results were received from Westat. The question-
naire responses and site diagrams were reviewed and a preliminary
test date was assigned for each site. Tests were scheduled to
maximize efficient use of the field teams and to complete the
1|fTest and Analysis Plan for the Tank Testing Program of the
National Survey of Underground Storage Tanks," H.K. Wilcox, J.W.
Maresca, Jr., J.D. Flora, C.L. Haile, June 10, 1985.
6-1
-------
survey as expeditiously as possible. Key scheduling considera-
tions were the geographic locations of the sites, the number of
tank systems at each establishment, and any special problems
related to tank testing anticipated from the survey questionnaire
results. Because up to three tank systems could typically be
tested each work day at a single establishment, one day was
allocated for sites with three or fewer tank systems. Similarly,
two days were allocated for sites with four or more tanks.
However, as testing three systems in a single day generally
required significant overtime, consecutive three-system days were
avoided where possible. Days were also incorporated into the
schedule for makeup tests.
As soon as possible following assignment of a preliminary
test date, the o/o was contacted by phone to arrange the test
appointment. The testing and data collection program were fully
explained and a mutually agreeable test date was established.
The o/o was also instructed how to file compensation claims to
EPA for costs incurred due to closure for testing.
The key requirements of the tightness testing were that the
system be removed from service during the test and that the tanks
be completely filled. Several gallons of additional product were
also required to top off the tank during testing. If the o/o was
unable to arrange product delivery to accommodate the test
requirements, assistance was provided in the form of contacts to
the appropriate fuel supplier.
The field crew assigned to a specific site contacted the
establishment o/o by phone or visit approximately two days prior
to the scheduled test date. This contact served to confirm the
test date, confirm that the establishment would be ready for
testing, and answer any additional questions from the o/o.
6-2
-------
II. SITE DESCRIPTION INFORMATION
A site inspection was conducted at each establishment during
the test visit prior to or during setup of the tank system test
equipment. The purpose of the inspection was to provide an accu-
rate-detailed record of the layout of the establishment, tank
system configurations, and environmental features that may be
related to system failure or leakage. This information was
recorded in the form of overall site sketches, detail sketches
for each tank, and a table of critical features. The site
sketches recorded the layout of tank systems and dispensers as
well as locations of buildings, roads and pavements, power lines,
and waterways. Color instant print photographs were taken to
supplement descriptions contained in the sketches. The following
details were recorded on a critical features data form for each
tank system:
Survey ID No.
Tank number
Product type
Number of dispensers
Tank size
Size of fill pipe
Size of gauge pipe
Size of stick pipe
Drop tube - permanent or removable
Delivery system - pressure, suction
Depth of tank from grade
Surface over tank
Presence of overhead power lines
Presence of nearby waterways
III. TANK SYSTEM TIGHTNESS TESTING
After the evaluation of a number of tank tightness test
methods (see Appendix C), it is clear that none of the standard
6-3
-------
test methods evaluated in the program can consistently and
reliably achieve detection of the 0.05 gallon per hour leak rate
specified by the NFTA 329 "Recommended Practices for Underground
Leakage of Flammable and Combustible Liquids, 1983." This
conclusion is based entirely on the data collected during the
method selection phase of the program, since supporting data
which was requested from the test companies to document their
performance claims was not received. EPA modified, for use in
the survey, one of the tank test methods to improve the accuracy
of the test results.
While some methods can provide reliable results under some
specific conditions, there are many situations which commonly
occur in the test environment (such as the presence of a water
table) which can invalidate the test results. Unless the test
crew takes specific steps to identify these conditions, the
reported results may be either misleading or incorrect. Most
test methodologies currently in use fail to take definitive steps
to identify one or more of these problem areas. The test results
obtained must therefore be suspect to the degree that these
factors are not recognized. The effects can be substantial and
cannot be generally evaluated from the test data after the test
crew has left the test site.
One of the major objectives of the program prior to the
national survey was to identify and characterize a test method
suitable for use on the program. A modification of the Petro-
Tite method was developed and characterized for this purpose.
This method was selected as the method with the most consistent
approach to identifying potential problems and taking action to
correct for them. It was judged to provide the most consistenyly
reliable data for the national survey. It has stated procedures
to identify and correct for tank end deflection, temperature
effects, water table, and vapor pockets. EPA modified the method
6-4
-------
in three ways in order to improve accuracy: (1) the test time
and sample frequency were increased; (2) the temperature
correction algorithm was improved; and (3) statistical hypothesis
testing procedures were applied to the data to determime leak
status.
A. System Test
1. Method Description
The Petro-Tite tank test measures product loss from the tank
system by monitoring the change in product level in an elevated
standpipe. Apparent volume changes are corrected for expansion
and contraction caused by product temperature changes during the
test to produce a net volume change. The net volume change over
time is equivalent to the leak rate. The key features of the
Petro-Tite method are that the test is conducted with the tank
overfilled into an elevated standpipe and that the product is
circulated during the test. The reference level in the standpipe
is set to maintain a hydraulic pressure, or fuel head pressure,
of 4 psig on the bottom of the tank in excess of any back
pressure caused by groundwater at a level above the tank bottom.
The purpose of conducting the test at an elevated pressure is to
increase the probability of detecting small leaks, to mitigate
masking of leaks by groundwater back pressure, and to stabilize
end cap deformation. The product is circulated during the test
to produce and maintain temperature homogeneity.
The Petro-Tite tank test equipment is shown in Figure 6.1.
A probe, inserted into the fill tube, consists of the circulation
pump inlet and discharge tube and a thermistor assembly. The
probe is sealed in the fill tube with an air bladder seal. The
6-5
-------
I
-p
c
0)
6-6
-------
circulation pump withdraws product from the fill pipe and
discharges it through a jet nozzle located near the bottom of the
tank. The nozzle is directed at a 45-degree angle down the
longitudinal axis of the tank to produce a swirling circulation.
The thermistor is located at the pump inlet and is connected to
an electronic thermal sensor module to provide temperature
readout. The standpipe is connected to the probe and also to a
graduated cylinder. During the test, the product level in the
standpipe is readjusted to the reference level using the
graduated cylinder. The volume of product added to or removed
from the standpipe to reach the reference level is measured from
the cylinder by difference, i.e., volumes in the cylinder are
read before and after raising the standpipe level to the
reference mark.
2. Method Operation
At the beginning of the test, the probe and thermistor units
were installed into the fill pipe and the circulation was
initiated. A small bore hole was drilled near the tank, prefer-
ably in the tank backfill, to determine if and at what level the
water table was above the bottom of the tank. The density and
temperature of the fuel product was determined with a hydrometer
and a thermometer. The product temperature and density were used
to determine the thermal expansion coefficient for the fuel from
physical properties tables prepared by the American Petroleum
Institute. The product density and depth to the water table were
used to determine the standpipe reference levels.
The standpipe and graduated cylinder were installed and
product was added to the standpipe to a "high" level reference
level to place a pressure of 5 psig on the tank bottom. Product
was periodically added to maintain this level until the rate of
6-7
-------
change indicated that end cap deformation had stabilized. Then,
the level was lowered to the "low" level to place 4 psig on the
tank bottom. This was the reference level for the leak rate
test.
The product level in the standpipe was readjusted to the
reference level at 5-minute intervals. The volume of product in
the cylinder before and after releveling were recorded on the
test data sheet. Fuel temperature readings were made and
recorded on the test data sheet at 5-minute intervals. These
data were also entered into a LOTUS 123 (tm) spreadsheet file on
a portable microcomputer. The test was conducted for 2 h with
readings at 5-minute intervals. The tank system leak rate was
calculated using the volume change, temperature data, tank
volume, and thermal expansion coefficient of the product. At the
conclusion of the system test, a line test was conducted. After
completion of the line test, the equipment was removed from the
tank.
Where possible, the entire tank system was tested as a
single unit. This included vent lines, distribution lines, and,
in the case of multiple tanks manifolded into a single system,
all tanks and syphon lines. However, in cases where vapor
pockets were found or the piping layout was not well known, tanks
were isolated and tested separately. Isolation of tanks from
associated piping generally required excavation to expose the top
of the tank.
Vapor pockets were also indicated in several single tank
systems. Vapor pockets were suspected when the standpipe level
fluctuated in an apparent haphazard manner. This was typically
caused by vapor trapped in manway or piping on the top of the
tank. In cases where vapor pockets were indicated, the top of
the tank was exposed by excavation and air bleed valves were
6-8
-------
installed on manway and bung covers. The vapor was bled from the
filled tank and normal tank tightness test was conducted.
Identification of vapor pocket problems and the need for
excavation could not be identified until the tank test was
attempted. Hence, testing tanks with vapor pocket indications
generally required at least two test days.
3. Performance Characteristics
The performance characteristics of the Petro-Tite test
method were empirically determined during the survey by examining
the variance within specific tests, and between retests on 34
pairs of data. The total variance was found to be 0.00264
ga!2/h which represents a standard error of + 0.0514 gallons per
hour. This procedure is covered in Appendix D of this report.
B. Line Tests
Tightness testing was conducted on the distribution lines
where possible. The system requirements for conducting a line
test are a suitable connection at the delivery end of the line to
install the test unit and a check valve in good working order.
The check valve, typically installed at the inlet of the
distribution line in the tank (foot valve) or in the line just
above the tank (angle check valve), prevents product in the line
from draining back into the tank. Hence, the location of the
check valve determines the portion of line subjected to the test.
6-9
-------
1. Method Description
The Petro-Tite line test system pressurizes the system with
product and measures the volume of product required to maintain
the reference pressure. Lines for pressure and suction distribu-
tion systems are tested in a similar manner, although the test
pressure differs. A diagram of the line test system is shown in
Figure 6.2. The test unit is connected to the delivery end of
the distribution line and the line is pressurized using a foot
operated pump to 15-30 psig for suction lines or 30-80 psig for
pressure lines. This pressure closes the check valve to prevent
fuel loss back to the tank. The pressure is monitored using the
gauge and the pressure restored periodically. The volume of
product required to restore the reference pressure is recorded.
2. Method Operation
The line test was conducted at the conclusion of the tank
system test, before the tank test equipment was removed. Air was
bled from the line and the test unit was connected to the
distribution line. Product was pumped into the line to achieve
the required pressure. The product level in the tank test stand-
pipe was monitored simultaneously to determine if the foot valve
was functioning properly. If product loss from the line was
observed as a volume increase in the tank test standpipe, the
check valve was considered leaking and the line test was
inconclusive. If possible, the check valve was replaced and the
test repeated.
The line set pressure was monitored and restored using the
test pump at 15-minute intervals. The product level in the
graduated reservoir was recorded before and after each pressure
restoration. The total product volume added during the 1-h test
6-10
-------
Graduated
Reservoir
Pressure
Gauge
Dispenser Line
Foot Operated
Pressure Pump
Check Valve
(Foot)
Tank
Figure 6-2. Line test equipment
6-11
-------
(a total of four 15-minute interval readings) was reported as the
line leak rate. If the volume required to restore pressure is
less than 0.025 gallons per hour for suction lines or 0.010
gallons per hour for pressure lines, the line is considered to be
tight. Volumes greater than these indicate that the line is
leaking or an invalid test.
3. Performance Characteristics
The performance characteristics of the line leak detector
have not been verified by independent measurements. However,
expert users of the device have stated its tolerance to be at
least + 0.0005 gallons per hour when used on a typical delivery
line.
The average standard error of line tests conducted on the
national survey was of the order of 0.001 gallons per hour. This
is more than an order of magnitude more sensitive than the
threshold leak rate.
Four situations can occur which can cause volume changes.
These are: leaks (or a bad check valve); changes in the liquid
temperature in the line; line expansion or stretching due to the
high pressure; and compression or shrinkage of air vapor present
in the line. All of these produce characteristics which can be
recognized by experienced personnel.
The two problems which caused the large number of line leaks
to be declared invalid were bad check valves and air pockets.
Ninety-one tests were declared to be invalid for these reasons.
In order to complete the testing on the 77 systems with bad check
valves, it would have been necessary to excavate the top of the
tank in most cases. This was beyond the scope of survey.
6-12
-------
At the conclusion of the line test, the pressure on the line
was released and the product allowed to flow back into the gradu-
ated reservoir. The bleed back volume was also measured. A
bleed back volume of greater than 0.050 gal indicated air in the
line and the test was considered inconclusive. The test was
repeated after air was bled from the line. If the repeat test
also had excessive bleed back, the test was considered inconclu-
sive.
IV. ENVIRONMENTAL DATA COLLECTION
General environmental data were also collected during the
tank system and line testing. These data included the following:
Ambient air temperature
Surface temperature above the tank
Subsurface soil temperature
Barometric pressure
General climatic conditions
Water table level.
These data were collected to provide a record of any
external temperature and pressure conditions that may have an
effect on the operation and results of the system and line
tightness tests.
All environmental data except the water table level were
recorded hourly during the test visit. The water table was
typically determined by drilling a bore hole through the tank
backfill material to the depth of the tank bottom. If water was
not encountered at that depth, it was recorded as being lower.
Anecdotal information concerning seasonal or other periodic
fluctuations were recorded as available. The ambient air tem-
6-13
-------
perature was measured using a mercury-in-glass thermometer. Sur-
face temperature was also measured with a mercury thermometer
placed on the surface at grade level over the tank. A thermistor
inserted into the bore hole previously drilled to determine water
level was used to monitor subsurface soil temperature.
Barometric pressure was measured with an aneroid barometer.
General climatic conditions, based on the observations of the
field technician, were recorded in common climatic terms such as:
light and variable winds, foggy, light rain, or sunny.
V. Tightness Testing Field Experience
A. Test Completion
A summary of the tests completed is presented here. There
were 485 manifolded tank systems from which 560 tanks were
selected for tightness testing. However, about 10 percent were
not tested because they were found to be out of scope or
untestable for technical reasons, or testing was refused by the
facility owner/operator. Out-of-scope tank systems consisted of
a closed fuel service station, small tank systems on farms (i.e.,
less than 1,100 gallons), and one system at an establishment that
had been misclassified. Technical problems included several
unused tank systems containing a residual sludge and tank systems
installed that did not permit access to install an air bleed
valve when vapor pockets were indicated. Some of the latter
cases included tank systems without bungs on the top and tank
systems installed under a building. A tank system installed
under a hospital helicopter emergency landing pad was considered
untestable due to the lack of an alternate landing location. The
final refusal rate was 3 percent.
6-14
-------
B. Technical Problems
A summary of the technical problems encountered in tank
system and distribution line testing is presented in Table 6-1.
Fuel delivery problems and vapor pockets were the most disruptive
to the test schedule. Coordination of fuel delivery scheduling
with the test schedule was a significant part of the test
preparation effort. In spite of extensive preparation, failure
of the supplier to deliver product as scheduled caused delays in
19 tests. Vapor pockets were indicated in 21 tests. These
required exposing the top of the tank by excavation and
installation of air bleed valves.
Many of the other problems involved features requiring
resolution to permit installation of the test equipment or
mitigation of vapor pockets. Permanent drop tubes, vapor
recovery systems, and pumps were removed and remote fill pipes
were excavated and disconnected to facilitate installation of the
test equipment. Manifolds were disconnected to mitigate vapor
pockets and to allow separate testing of individual tank systems
in some manifolded tank systems. Failure of foot valves was a
frequent problem encountered during distribution line testing.
Also, excessive bleed-back volumes, indicating air in the
distribution line, caused 14 line tests to be considered
unreliable.
6-15
-------
Table 6-1. Technical problems summary
Problem Number1
Fuel delivery 19
Vapor pockets 21
Permanent drop tubes 17
Vapor recovery systems 6
Pump 2
Remote fill pipe 2
Manifolds 14
Other 23
Foot valve failure 77
Excessive bleed-back volume 14
More than one problem could be encountered in a given test.
Hence the total number of problems is greater than the number of
tests with any problem.
6-16
-------
SECTION 7
TANK TESTING DATA REDUCTION
AND QUALITY ASSURANCE (RETEST) RESULTS
This section describes the statistical data reduction
process whereby the raw data generated by physical tank tightness
tests in the field were converted into estimates of volume change
rates under test conditions for tank systems (vessels plus
piping). It then gives results of the quality assurance retests
which help in judging the overall accuracy of the physical test
and data reduction process. Note that throughout this Section of
the report, quantitated volume change rates are given as measured
under test conditions and are not adjusted from test pressures to
operating pressures.
I. DATA COLLECTION AND REDUCTION
A. Data Collection and Transmission
Raw data of volume and temperature change at five-minute
intervals were collected for a two-hour period during the
physical test, as described in Section 6. These data were
collected in handwritten form on the data sheets normally used by
Petro-Tite, and were also entered onto a spreadsheet using a
mini-computer at the field site. Data were transmitted from the
spreadsheet to MRI by telephone for timely analysis. The
diskettes and hard-copy data sheets were shipped to MRI on a
weekly basis. The telephone transmission was checked against the
diskette, and the diskette against the hard copy to ensure that
the correct raw data were entered in the working spreadsheet
7-1
-------
file, which was then used to estimate volume change rates and
their (within-test) variability.
B. Standard Data Reduction
Several statistical analysis methods for reduction of the
tank test data were considered for use on the national survey.
The test method produced a volume change measurement at five-
minute intervals. The other measurement recorded at five-minute
intervals was a temperature measurement. The temperature was
recorded as a cumulative reading — the tank temperature -- while
the volumes were recorded as differences. In order to make the
temperature and volume data comparable, they had to be put in the
same form. Either both must be changes or both must be
cumulative.
As a result of the considerations of the types of analyses
available (see Part III of Appendix D for discussion of possible
methods) and the advantages and disadvantages of each, a standard
analysis was designed. For the standard analysis, the estimated
volume change due to temperature change and the observed total
volume change were both expressed in cumulative form, beginning
at zero for the start of the test. A straight line through the
origin was fit to the temperature-related volume change data by
least squares. The predicted values of this line were calculated
and used as a smoothed temperature correction for the observed
volume changes. The data were plotted and inspected visually for
outliers or deviations of the temperature data from linearity.
Any questionable data were checked in detail or considered for
special analysis.
If no problems with the data were found, the predicted
values from the smoothed temperature line were used as the tem-
7-2
-------
perature correction. This smoothed temperature correction was
subtracted from the observed volume data for each time point.
The resulting differences were divided by the time interval to
obtain a series of volume change rates expressed in gallons per
hour, typically based on a five-minute interval. The arithmetic
mean of these rates was calculated and used as the estimate of
the volume change rate. In calculating the variance n-1 was used
as the divisor, where n is the number of terms in the mean. The
result was divided by n to form the variance of the mean. The
square root of this is the within-test standard error reported
before adjusting for between-test variation. See Part V of
Appendix D and Part I of Section 8 for a discussion of the
between-test variance.
C. Special Analyses
A number of data set features called for a different or more
detailed analysis than that described above. These were dealt
with on an individual basis. Occasionally apparent outliers were
found. These were checked against the raw data and the test log
to see if there was any physical reason for them. A few tests
had thermistor boxes fail during the test for some reason (rain,
FM interference). These generally gave temperature data that
appeared as outliers. When outliers were found and a physical
reason identified, the aberrant data were removed from the
analysis. This generally required smoothing over the missing
data by interpolation. If errors were identified, they were cor-
rected and the analysis redone.
The typical data showed a consistently increasing
temperature, generally linear. A smaller proportion of the data
sets showed linearly decreasing temperature. Some data sets
showed evidence of temperature increase that was curvilinear. If
7-3
-------
this curvilinearity appeared or was suspected, a test for
curvilinearity was done by fitting both a linear and quadratic to
the temperature data by least squares (through the origin). If
the quadratic improved the fit significantly, the curvilinear fit
(using both linear and quadratic terms) was used for smoothing.
A few cases were found where both temperature and volume
were not only non-linear, but also not moving consistently in one
direction. Provided that they showed the same pattern, analysis
proceeded. In this event, a five point moving mean was used to
smooth the temperature data. Equal weights were used. This
resulted in the loss of four data points; two at the start and
two at the end of the test.
Some tests showed volume change rates that were initially
increasing rapidly in curvilinear fashion, while the temperature
changes were quite linear. The volumes typically increased
rapidly for the first few observations, then slowed. This was
interpreted as relaxation or tank deformation. The apparent
relaxation appeared to follow an exponential curve and to
approach the temperature change rate as an asymptote. However,
the constant of this differed by tank. The rate of relaxation
may be related to the nature of the soil in backfill and water
conditions. When this was identified, the initial points
exhibiting this relaxation of the tank deformation were deleted
before analysis.
D. Criteria for Invalid Data
A few of the data sets from the tank tests were judged
invalid based on the analysis of the data. This occurred
infrequently (in 6% of test results).
7-4
-------
There were a number of criteria for declaring a data set to
be invalid. The most common was that the data showed a volume
increase even after adjusting for temperature. Since the test
method places pressure on the tank, a volume increase cannot
occur from inflow of water. Data that showed volume increases
after temperature adjustment that exceeded levels that could be
reasonably attributed to the variability of the measurement proc-
ess were judged to be invalid tests. The reason for this is that
such an apparent volume increase with no explanation could be
eclipsing a small actual volume loss or leak. Generally any tank
that showed a volume gain rate of more than 0.1 gallons per hour
after temperature adjustment was judged to be an invalid test.
The most likely explanation for such tests is that those tanks
had trapped vapor pockets.
As described in Section 8 (Part III) and Appendix D (Part
VII), at the next stage of analysis, some additional tests were
judged to be invalid due to a measured inflow that was excessive
when compared with its estimated total standard error, even
though the inflow was not as large as 0.10 gallons per hour.
A variety of other data features led to the conclusion that
the test was invalid. A few instances were found where the
temperature as recorded fluctuated erratically during the test
while the volume measurements were relatively stable. If the
temperature data were so erratic as to preclude a temperature
adjustment, then the test was declared to be invalid.
One or two tests showed both temperature and volume
measurements that were erratic and did not appear to track
together. These tests were also judged invalid. Such behavior
may have been caused by incomplete tank deformation, followed by
relaxation, combined with mixing problems. No valid volume
change rate could be estimated.
7-5
-------
II. RETEST RESULTS
Three types of retests were conducted as part of the
national survey of underground storage tanks. One was a back-to-
back retest, conducted immediately after the original test used
to estimate the volume change rate. The second was a leak
simulation test also conducted immediately after the original
test. The third type was a complete retest conducted on a
different day and generally by a different crew. Each of these
types of second testing estimates a different source of variation
possible in the tank tests. The leak simulation and back-to-back
retests estimate variation of the overall measurement procedure
from one two-hour period to the next, with the same set-up, crew,
day of the week, and so on, while the complete retests measure
variation between tests as well. The initial test result in each
case was used as the data for the survey estimate. When the test
and retest results differed, the results were examined to
discover reasons for the differences. This led to the discovery
of the repairs that had been made in two cases. The primary
purpose of the quality assurance program was to measure the
overall performance of the test, which was accomplished. A list
of all of the retests appears in Appendix D, as does a list of
the simulated leak retests. A table summarizing the estimates of
bias (lack of accuracy) and standard deviation (precision) based
on each type of test is presented as Table 7-1. We discuss these
three types of quality assurance retests in more detail below.
A. Leak Simulations
The leak simulation tests were conducted after the original
test was concluded. Generally they were only conducted when the
original test indicated that the tank was tight or had a small
estimated volume change. The volume rate used for leak
7-6
-------
"d
Q)
-P
M
d
id
-p
O
d
V
in
d
o
•H
-P
•H
•d
d
0
o
-p
w
0)
-p
SH
w
— 0)
^_(
w O.
-p
rH O>
d d
M -r)
0) -P
SH (d
-p fl)
w 0-.
0) O
-p
tt) O
04 -P
•
rH
1
r~
0)
rH
XI
(d
EH
d
O D1 0) —
O W
•d d
SH 0
(^ .,_f ^— ^
•d -PX!
d (d &
(d -H tj>
-P t> ' — •
C/3 0)
'd
•d
0)
SH
2
. — ,
SH
^j
O
)
EH
CM
r~-
CM
o
•
0
r-
m
CNJ
0
•
o
T
r-
0
0
o
•
o
vo
VD
O
O
O
•
O
rH
rH
rH
CTl
CO
O
O
•
O
1
d
o
•H
(d
rH
^3
g
•H
W
y
-------
simulation was on the order of 0.1 gallons per hour, so a large
observed volume change would overwhelm it.
The purpose of the leak simulation tests was to document
that the testing method could detect leaks of known size in tanks
that appeared to be tight. In addition, use of the leak
simulation allows for an estimate of the accuracy of the test as
well as its precision. The accuracy refers to the ability of the
test to measure a known volume change, while the precision of the
test refers to its ability to reproduce measured rates.
Thirteen leak simulation tests were conducted. Two of these
were conducted on tanks that had estimated volume rates that
indicated that the tanks were probably leaking (as evidenced by
the observed volume changes). These tests were excluded from the
analysis because variability is known to increase for leaking
tanks.
Three rates were calculated from leak simulations. The
first was a baseline rate for the tank. This was estimated dur-
ing the regular tank test. While the leak simulation was con-
ducted, a measured rate was estimated. This is the rate observed
by the testing method during leak simulation. It is presumed to
be composed of the tank rate plus the simulated rate. The simu-
lated rate is calculated by collecting product drawn from the
tank at a constant rate, weighing it on a triple beam balance,
and converting the weight to volume at the temperature of the
product in the tank. The difference between the observed rate
during the simulation and the baseline rate provides an estimate
of the simulated rate. The difference between this and the
actual simulated rate can be used to assess the accuracy of the
test.
7-8
-------
The average difference between the measured rate and the
simulated rate was -0.00891 gallons per hour, based on the 11
leak simulations where the tank was not estimated to be leaking
(see Table 7-1). Recall that these rates are reported as
observed under test conditions and not as adjusted for test
pressure. The difference between the measured rate and the
simulated rate is interpreted as an estimate of bias. The
variance of the differences about their mean provides an estimate
of precision. This variance was estimated to be 0.00066 gallons
per hour squared. The mean squared error (MSB) is a measure
which incorporates both types of error—accuracy and precision.
It is calculated as the sum of the bias squared plus the
variance. In this case it was 0.00074 gallons per hour squared.
The bias is clearly not significant in that it does not
differ significantly from zero (t = -0.347, 10 degrees of
freedom). As a result, the variance and the mean squared error
are nearly identical. A measure of variation often used is the
standard deviation (or root mean squared error if bias is
present), which is the square root of the variance (or MSB).
This measure has the advantage that its units are the same as the
measurement, gallons per hour. The standard deviation
(estimating within-test variation)was estimated to be 0.0257
gallons per hour for these data.
B. Back-to-Back Retests
Back-to-back retests were conducted on a total of 18 tanks,
which includes the 13 tanks with leak simulations. The purpose
of the back-to-back retests was to estimate the stability of the
test method. That is, to ensure that the volume change estimate
did not differ markedly if based on the succeeding two hours
after the test.
7-9
-------
As with all of these tests, variability is expected to be
larger if the initial leak rate or volume change is larger. For
this reason, the results of the back-to-back retests are pre-
sented here for the 14 tests with volume change rates less than
0.1 gallons per hour in absolute value. Retest results for tanks
with larger volume rates were more variable but generally
consistent. (See Appendix D, Part IV for a discussion of these
retests.)
The average difference between the original and retest for
the 14 tests with small volume changes was 0.00629 gallons per
hour. The variance estimate was 0.00053 gallons per hour
squared, giving a mean squared error of 0.00057 gallons per hour
squared. The corresponding standard deviation was 0.0231 gallons
per hour and the root mean squared error estimate was 0.0239
gallons per hour (not adjusted for test pressure). The mean
difference was not significantly different from zero (t = 0.272,
13 df).
C. Complete Retests
The complete retests consist of revisits to the site on a
different day. Typically this includes a different crew and
involves rescheduling and refilling the tank. The complete
retests incorporate all of the features of a tank test and so
include all the sources of error including potential difference
from crew to crew (including differences between sets of testing
equipment) and differences due to weather conditions, nearby
traffic, day of the week, etc. In addition, there is a
possibility that the tank is different at the time of the retest.
In fact, two of the retests originally scheduled were canceled
when it was found that the tanks had been repaired between the
7-10
-------
initial test and the scheduled retest. In addition, two retests
were performed and it was then discovered that the tanks had been
repaired between the initial test and retest. These data are
also not included as they would measure an additional source of
variation which is not of interest (i.e., repair). Two other
retests were performed on tanks that were initially determined to
have large vapor pockets. These two tanks were retested later
and on retesting were again found to have large vapor pockets.
The results of the test and retest for these tanks with vapor
problems agreed qualitatively; however, the numerical agreement
was not close. The reason for this may be that the vapor pocket
trapped in the tank was of different size. There were also
different ambient conditions that would affect the vapor
differently. For these reasons, the vapor retests were not
included in the estimate of the variance from the retests.
The mean difference from the set of 34 relevant retests was
0.00297 gallons per hour. The variance of the difference was
0.00254 gallons per hour squared, giving a mean squared error of
0.00255 gallons per hour squared. The standard deviation of the
differences for these 34 retests is 0.0504 gallons per hour and
the root mean squared error is also 0.0505 gallons per hour. The
mean difference is not significantly different from zero (t =
0.059, with 33 df).
D. Results
The retest data analysis showed no evidence of bias in the
test methods. All three retest schemes had very small estimates
of bias which were not significantly different from zero. Given
the historical leak cut-off of 0.05 gallons per hour, bias of
less than 0.01, as was found in all three data sets (less than
7-11
-------
0.005 in the largest set) is not of practical concern, in
addition to not being statistically significant.
The variability, or magnitude of the measurement error of
the physical test can also be assessed using these data. Both
the back-to-back retest and the leak simulations estimated
within-test standard deviations on the order of 0.025 gallons per
hour. The complete retest data gave a standard deviation of 0.05
gallons per hour for the total variability of volume change rate
estimates. As is discussed in Section 8, Part I, the difference
is probably due to a between-test component of variation which is
measured by the retests but not by the back-to-back or leak
simulation tests.
In summary, the physical test is accurate (not significantly
biased) and has a known precision (total standard error of an
estimated volume change) of 0.05 gallons per hour (measured at
test pressure and not adjusted to operating pressure). That this
standard error coincides with the historical cut-off value for
declaring a leak is an interesting coincidence.
7-1 2
-------
SECTION 8
STATISTICAL ANALYSIS OF LEAK DATA
AND LEAK STATUS DETERMINATION
Section 6 above described the physical tightness test
procedure used in this survey, a modification of a commercially
available method. A field test for a single tank system
produced raw data which required analysis and interpretation
before a determination could be made as to whether the test
showed evidence that the tank system was leaking. Section 7
described the initial steps of data reduction which yielded a
measured volume change rate and an estimate of the measurement
variability of that rate for each tightness test. In that
section, results of quality assurance retests were also given.
This section of the report discusses further statistical
analyses required to evaluate the total measurement variability,
to estimate the actual leak rate and to determine whether a given
tank system can be judged to be tight or leaking, based on the
tightness test. It also includes three further analyses. One
speaks to the issue of whether the leaks measured by the test can
be attributed to leaks in distribution lines. The second looks
at how full tanks are kept in practice, which sheds some light on
the relevance of assessing tank system leaks by filling tanks to
capacity. The third discusses the possible impact of the typical
filling behaviors reported on the estimates of percent of tank
systems that leak in practice.
xDue to the nature of the test, a "tank system leak" means a leak
anywhere in the tank vessel, associated fill pipe, vent pipe,
distribution lines, fittings, or connections.
8-1
-------
I. TOTAL MEASUREMENT ERROR
The quality assurance retest data given in Section 7 offers
a means of estimating the size of the error in the physical test
measurement that is due to variation from one test occasion to
another. It also indicates that the magnitude of this component
of measurement error is substantially greater than the error
measured for a single test result. In the usual statistical
analysis of components of variance, the component measured for a
single test result is the within-test variance, the component due
to factors varying from test to test is the between-test
variance, and their sum is the total variance of a single test
result.
In order to estimate the total variance of a given test
result, the average between-test variance must be estimated and
added to the within-test variance estimated for that test. Two
data sets were used to estimate the between-test variance: (a)
the complete retest (34 cases) and (b) those test results from
tanks which are clearly tight (observed volume change was a flow
into the tank system of 0.0 to 0.2 gallons per hour, 133 cases).
Data set (a) allows an estimate of total variability because both
between-test and within-test components are involved in the
test/retest but the underlying tank leak rate would not vary
between test times. Data set (b) provides measures of leak which
are due to random error alone. No liquid could actually flow
into the tank during the test, since product in the tank was
under test pressure. Thus, data set (b) also includes both
components of variance. An estimate of total variance was
computed from the measured volume changes in each data set, and
the average within-test variance was estimated from the measured
within-test variances for the same data sets. The between-test
variance was then estimated by subtraction. Appendix D, Part V
8-2
-------
describes the estimation of between-test variance in more detail,
specifying the model and formulas used.
The two sets of variance components estimates agreed very
closely. In order to get a single estimate of between-test
variance to use in adjusting the stated within-test variances,
the two estimates were combined in a weighted average using the
number of cases in the data set as the weights. The final
estimate of between-test variance was 0.00199 gallons per hour
squared, which in terms of gallons per hour is about 0.04 gallons
per hour. (This figure has not been adjusted for test pressure
as described below.) The estimated between-test variance was
added to the square of each within-test standard error to get an
estimated total measurement variance for each test result. The
square root of this was then used as the total measurement
standard error in the statistical hypothesis test described
below.
II. ADJUSTING MEASURED LEAK RATES TO ACCOUNT FOR TEST PRESSURE
This subsection describes how leak rates were adjusted from
test pressure to "typical" operating pressure (i.e., a set of
standard assumed conditions) for tank systems judged to be
leaking under test conditions. In order to conduct the physical
test, increased hydrostatic pressure is placed on the tank
system. As a consequence of this, any leak or flow through an
orifice in the tank system would be increased over what would
occur under the (smaller) pressure encountered in operation.
Torricelli's form of Bernoulli's Law was used to calculate
adjustments to the measured flow rates, under certain
assumptions. It should be noted that the basis for the
adjustment is the assumption that the measured flow represents a
leak through an orifice or hole. Thus, it is not logically
8-3
-------
consistent to adjust test volume change rates for pressure unless
the system, tank, or line was judged to be leaking, even though
it would be computationally feasible to do so, and leak rate
adjustments were made after the leak status had been determined.
Torricelli's and Bernoulli's Law assumes a flow is through
an orifice with neither resistance nor turbulence. In our
situation, this is not the case. The flow rates generally will
be small enough so that the assumption of no turbulence is
reasonable. However, in most cases, leaks will be through
corroded sections and will be into soil which may present some
resistance. The effect of resistance would be to lower the flow
rate. How much the flow rate would be lowered under the
different pressures is not known. Consequently, the effect of
violation of these assumptions on the adjustment to leak rates is
not known, but it is assumed to be negligible. There are some
other implicit assumptions. These include that the orifice is
constant, that the temperature and density do not change, and
that the product is not viscous.
Since the test is conducted at elevated pressure, flow rates
through any orifices will be larger under the test conditions
than they would be under actual tank operation. The magnitude of
the difference depends on a large number of variables. In
particular, flow rates would vary by location of the hole in the
tank (distance from the bottom), amount of fuel in the tank, and
pressure of a water table part way up the tank. The adjustment
factors would also vary with diameter of the tank. Since diesel
tanks were tested at the same pressure (hence at a lower head-
distance) as gasoline tanks, the adjustment also varies with fuel
type because of the density difference. The assumed operating
conditions used in calculating the adjustment factors (in
8-4
-------
addition to the basic assumptions of Bernoulli's law) are as
follows:
The water table is assumed to be below the bottom of
the tank;
The tank is assumed to be buried to the depth of three
feet from grade to top of tank;
Three tank diameters are assumed based upon volume,
since actual diameter of tanks was not known);
The average operating level of the tank is assumed to
be half full; and
The orifice or hole is assumed to be in the bottom of
the tank.
The table below gives the adjustment factors used to adjust
the estimated tank leak rates to these assumed standard operating
conditions.
Adjustment factors for tank system leak rates
Tank diam
associated vo
48"
64"
96"
(0-1,100
(1,101-7,
(7,001-15
eter and
lume ranges
gallons)
000 gallons)
,000 gallons)
Fuel type
Gasoline Diesel
0.395 0.430
0.456 0.496
0.558 0.608
The factors were multiplied by the leak rates estimated by
the physical tests to obtain the adjusted leak rates. The
adjusted total measurement error is calculated by multiplying the
total measurement standard error by the adjustment factor. Leaks
measured by the line tests can be similarly adjusted either to
the system test pressure or to assumed operating conditions. As
8-5
-------
is discussed in Section V below, valid distribution line tests
could not be conducted on about 60 percent of the tank systems
judged to be leaking. Further, distribution line leak rates
accounted for very little of the system leak rate. Thus for the
major findings in this report, system leak rates were adjusted
directly to assumed operating conditions. See Part VII of
Appendix D for details.
III. DETERMINATION OF LEAK STATUS
The physical leak measurement technique was described in
Section 6. As a result of variability in the instrument readings
and temperature adjustment process, the physical test does not
produce an absolutely positive determination that a tank system
either is leaking or is not leaking.2 Instead, the test produces
an estimated leak rate (or flow rate) along with a measure of
uncertainty in the estimate (i.e., standard deviation of the leak
rate).3 The determination that a tank system is "leaking" is,
therefore, a statistical judgment. The approach taken to leak
status determination in this report is the statistical hypothesis
testing model. The condition of "non-leaking" is represented by
the hypothesis of a zero leak rate. The condition of leaking can
then be stated as "having a measured leak rate that is
significantly different from zero (flowing out of the tank
system)."
2In common practice, a tank is certified or not based on
comparing the observed volume change rate to the NFPA standard
of 0.05 gallons per hour.
3In some cases, the tank system test produced data that were
judged to be unreliable. These are considered inconclusive
results, by any decision rule. In other cases, unguantifiably
large leaks were encountered. These are judged to be leaks by
every decision rule.
8-6
-------
The ideal statistical test should have:
o A low probability of false alarm (i.e., for non-leaking
tank systems, have a low probability of falsely calling
the tank system leaking. This probability is also
known as the significance level); and
o Have a low probability of failing to detect a real
leak. (This probability is one minus the power of the
test.)
The probability of failing to detect a real leak depends on
three factors: the size of the total measurement error of the
tank system test, the size of the "real leak" one wishes to
detect, and the probability of false alarm for the statistical
test adopted. Recognizing the inherent conflict between
objectives (a) and (b), above, we have considered two statistical
tests with the attributes shown below:
Probability of
Probability of failing to detect
a false alarm a leak of 0.10
for a tight gallons per hour1
tank system or more
Test 1 5 percent 5 percent
Test 2 1 percent 16 percent
^•A true leak (i.e., adjusted for test pressure) of 0.10 gallons
per hour would have the stated probability of detection on
average, since the average adjusted total standard error is
0.030 gallons per hour.
We have selected Test 1 as the approach used in this study
because it provides a low probability of a false alarm and
provides an equally low chance of failing to detect a leak of
0.10 gallons per hour when one really exists. The null
8-7
-------
hypothesis was that the true leak rate is zero, and the
alternative hypothesis was that there is a leak out of the tank.
The test statistic was the observed volume change rate divided by
its standard error. The null hypothesis was tested at a 5
percent significance level by comparing the test statistic to
1.645. If the test statistic was greater than this value, which
is taken from the normal distribution, the tank was judged to be
leaking. Table 8-1 shows the resulting estimates for percentage
of tank systems leaking in the United States (35 percent using
Test 1; 32 percent using Test 2). Note that the leak status
determination is made before adjusting the leak rate to operating
conditions for tank systems judged to be leaking by the
statistical test.
Although the probability of failing to detect a leak of a
given size (one minus the power) is reported on average for the
statistical test used in the report, it is not guaranteed for
each tank system tested. One way of specifying this probability
as well as the probability of false alarm (significance level) is
to run a second test which declares a tank system not leaking
only if its measured flow rate is significantly greater than
would be consistent with a stated actual leak rate. The final
decision for a given tank system test would be inconclusive if
the two statistical tests disagreed. Applying this approach with
a five percent (or less) probability of failing to detect a leak
of 0.10 gallons per hour (on an adjusted basis) or more does not
noticeably change the results. The same tank systems are judged
to be leaking as were under Test 1 above, and a few of the tank
systems judged not leaking by Test 1 become inconclusive after
applying the second rule (six cases in the raw data). The
percent judged leaking calculated from this slightly reduced base
remains 35 percent when rounded to the nearest percent. Thus the
stated probability of failing to detect a fairly large leak does
hold for most actual tests.
8-8
-------
Table 8-1. Estimated percentage of underground motor fuel
storage tank systems judged to be leaking under test
conditions in the U.S., business and government
sectors, using statistical tests
Leak
status of tank system1
Test 1
(Pi = -05
P2 = .05)2
Test 2
(PI = .01
P2 = .16)2
Percent judged to be leaking
under test conditions3 35% 32%
Percent judged to be not leaking
under test conditions 65% 68%
both tests, 5.5 percent of tanks tested had an
inconclusive result. This includes cases where the data were
judged unreliable and cases with a statistically significant
measured inflow, indicating a vapor pocket or other problem with
the test. They are not included in the base on which the
percentages are figured.
2P± = Probability of a false alarm (i.e., falsely declaring
a tight tank as leaking) on any one test.
P2 = Probability of failing to detect a leak of 0.10 gallons
per hour when one exists, using a value of 0.03 as the
standard deviation of leak rate (the average adjusted total
standard deviation)
3Includes cases with unquantifiably large leaks as well as with
statistically significant leaks.
8-9
-------
As a final comparison, we have looked at the estimate of
percentage of leaking tank systems that would have resulted if we
had used the 0.05 gallons per hour National Fire Protection
Association criterion4 (or "cut-off") as the critical value to
distinguish tight or certifiable tank systems from not
certifiable (leaking) tank systems. Although this is not a
statistical test for detecting non-zero leak rates, it is
included for comparison since many commercial tank testing
companies apply this criterion, although they use various field
test equipment and procedures. Results of applying this
criterion are shown in Table 8-2. Applying the .05 gallons per
hour cut-off to the estimated leak rates at test conditions
results in an estimate of 42 percent of tank systems leaking.
Adjusting the leak rates for these tank systems downwards to
compensate for the pressure used in the test, as described in
Section 8-II above, and applying the .05 gallons per hour cut-off
to the adjusted leak rates results in 33 percent of tank systems
leaking (because 21 percent of the leak rates initially greater
than 0.05 gallons per hour were reduced to less than 0.05 gallons
per hour as a result of the pressure adjustment).
These results are given solely as illustrative of what might
be found if a national certification program were conducted.
Each testing company conducts its tests under different pressures
and with different sensitivities. The cut-off is then applied to
the observed volume change rate without any pressure adjustment.
The results using the NFPA criterion show a large fraction of the
underground motor fuel storage tank systems in the United States
to be leaking, even though the NFPA test ignores leaks below the
4ANSI/NFPA 329, "Recommended Practices for Handling Leakage of
Flammable and Combustible Liquids," Section 4-3.10.1, 1983.
8-10
-------
Table 8-2.
Estimated percentage of underground motor fuel
storage tank systems judged to be leaking under test
conditions in the U.S., business and government
sectors, based on NFPA .05 gallons per hour criterion
Leak status of tank systems1
Percent judged to be leaking
under test conditions
Percent judged to be not leaking
under test conditions
Using NFPA .05
gallons per hour criterion
At test
pressures2
42%
58%
Adjusted to
typical
operating
conditions
33%
67%
1In both columns, 6.4 percent of tanks tested had an inconclusive
result. This includes cases where the data were judged unreliable
and cases with a measured inflow of greater than 0.05 gallons per
hour, indicating a vapor pocket or other problem with the test.
These cases are not included in the base on which the percentages
are figured.
2The test procedure resulted in a small pressure, 4 psi, at the
bottom of the tank.
3Leak rates were adjusted to typical operating conditions only for
those tank systems initially judged to be leaking.
4Includes cases with unquantifiably large leaks as well as with
measured leak rates greater than 0.05 gallons per hour.
8-11
-------
arbitrary 0.05 gallons per hour and does not examine the
estimated leak rate relative to its standard error.
In summary, this report uses a statistical test to
distinguish leaking and tight tank systems. The test has a five
percent probability of false alarm, while running on average a
five percent risk of failing to detect leaks of 0.10 gallons per
hour or larger. Furthermore, the use of other statistical test
criteria described above would not substantially alter the
overall results.
IV. ASSESSING THE UTILITY OF TESTING FOR LEAKAGE BY
FILLING TANKS TO CAPACITY
When tank systems were tested for leaks, the process
involved filling the tank to capacity and then observing the
resulting leak rates, if any. A concern arising from this
approach is the extent to which this procedure reflects the
general status of tank storage. For example, if it is the case
that tanks are seldom filled to capacity, the discovery of leaks
in the tops of tanks using the specified testing procedure does
not provide information with general application to real world
situations. On the other hand, if tanks are filled to capacity
or near-capacity routinely, the testing procedures are
appropriate.
In order to assess the extent to which tanks are utilized in
their full capacity, an analysis of the average proportion of
each tank utilized before and after delivery of a motor fuel for
storage (as reported by tank owner/operators in the
questionnaire) was undertaken for the nearly 2,300 cases on which
such data was available. Just before delivery, the median
average proportion of a tank utilized was 20 percent, with the
8-12
-------
75th percentile being 33.3 percent and the 25th percentile at 10
percent. The most frequent or modal value is 25 percent. Thus,
on average, the product stored within tanks is allowed to drop to
a rather low level before delivery of a further supply. After
delivery, the median average proportion of the tanks1 capacities
utilized jumped to 83.3 percent, with the 75th percentile at 97.5
percent and the 25th percentile at 62.7 percent. Thus, 25
percent of the tanks were filled over 97.5 percent full. The
most frequent or modal value was 100 percent.
Respondents were also asked the largest amount their tank
was ever filled. Responses to this question had a median value
of 100 percent. The 25th percentile was 92 percent full. Thus,
the majority of tanks are filled to capacity at some time, and
most tanks have filled nearly to capacity.
Such sample estimates suggest that a substantial number of
the tanks surveyed routinely utilized all or most of their
storage capacity, consequently exposing virtually all of a tank's
surface area to the possibility of being a potential source of a
leak. It is therefore reasonable to employ a testing procedure
which .involved filling a tank to capacity in an effort to detect
possible sources of leakage.
8-13
-------
V. ASSESSING THE CONTRIBUTION OF DISTRIBUTION LINE LEAKS TO
MEASURED TANK SYSTEM LEAKS
A. Comparison of Leaking Tanks with Valid Distribution
Line Tests With Those Whose Distribution Lines Were
Not Tested
In assessing the potential proportion of measured tank
system leaks5 which might be attributed to distribution line
leaks, a subset of our representative sample of tank systems was
analyzed. This subset was first restricted to single tank
systems judged to be leaking and with quantified leak rates, of
which there are 110 in the data set. Although distribution line
tests were attempted for all tested tanks, such tests often
failed due to bad check valves, bleed-back or other problems (see
Section 6). Thus, valid distribution line tests with quantified
leak rates were available for only 43 of the 110 quantified
leaking tank systems. Before analyzing the data obtained from
the completed valid distribution line leak tests, it is important
to consider the issue of the extent to which they may differ from
tanks where valid distribution line test results were not
obtained. For example, is there evidence that the ability to
complete valid distribution line tests was linked to the type of
fuel stored in the tanks, the material used in the construction
of the tanks, etc? Cross-tabulations comparing the distributions
of those leaking tank systems with valid line leak data and those
without across categories of variables with potential impact were
examined in conjunction with tests of association, such as the
chi-square test, where appropriate.
5Recall that a "tank system leak" refers to a determination of a
leak based on the physical test. The leak could actually be in
the tank vessel or any of the associated lines, pipes, or
fittings.
8-14
-------
Variables for which there was evidence of a difference in
the distribution between those leaking tank systems with
completed valid tests and those without included: type of
establishment (government sites had a higher than expected valid
completion rate), material of construction (fiberglass had a
lower than expected valid completion rate), and type of delivery
system (pressure pumps had a lower than expected valid completion
rate). Comparisons between the group with valid data and that
without on the actual measure of leak rate indicated no
significant differences between their average values, although
the leak rates for those tank systems with incomplete tests were
significantly more variable than those without.
B. Consideration of Distribution Line Leakage as a
Proportion of Total Measured Leakage
The finding of a leak during testing can be attributed to a
distribution line leak, a vessel leak, both, or some other
leak(e.g., fill pipe, vent pipe, manway fitting, etc.) For the
43 cases known to have leaks and for which valid distribution
line leak data was obtained, an examination of the proportion of
each total measured leak attributable to a distribution line leak
was undertaken. In doing this, it was necessary to adjust the
line leak rate values to the system test pressure, taking account
of differences in tank diameters and the type of pump used
(suction or pressure). (Appendix D, Part VI, described this
adjustment.)
Of the 43 cases, in only one case (2.3%) was the proportion
of the system leakage attributable to the distribution line more
than 25 percent. In fact, leakage in the distribution line
accounted for less than 10 percent of the leakage for 38 of the
8-15
-------
43 cases (88.4% of the cases) and for less than two percent of
the measured leakage for 23 of the 43 cases (53.5% of the cases).
Thus, at least for those cases in which valid leak rate
measurements could be obtained, leaks in the distribution lines
account for very little of the total measured leakage.
VI. PERCENT OF TANK SYSTEMS LEAKING UNDER OPERATING CONDITIONS
Certain features of the tank testing method are different
from typical operating conditions, especially the overfilling of
the tank during the test.
It is certainly reasonable to ask whether some of the leaks
detected under test conditions might have been due to holes near
the top of the tank above normal fill levels. Data from the
survey reveal that it is common practice to fill tanks to 100
percent capacity when product is delivered. In fact, 100 percent
was the modal value for this variable, and the median of the
reported average fill level was 83 percent of capacity. Thus,
the data suggest that even holes near the top of the tanks would
be subject to leaking, at least just after product delivery.
On the other hand, the average tank fill level just prior to
delivery had a median value of about 20 percent of tank capacity.
Therefore, as a rough approximation, a typical operating level
might be midway between the high and low point, or 52 percent of
capacity. If one were to further assume that holes were evenly
distributed between the top and bottom of the tank, then an
estimated 52 percent x 35 percent = 18 percent of the tank
systems would be leaking on the average at any point in time
8-16
-------
under typical fill level conditions.6 Furthermore, using average
percent filled after delivery may be a conservative estimate of
operational fill levels. When asked about the maximum gallons
ever stored, most respondents reported 100 percent, and only one-
quarter were below 92 percent full.
In summary, if we are willing to assume that holes are
uniformly distributed around the tank circumference (We have no
data to verify this assumption.), we could calculate that:
o Approximately 35 percent of the tank systems would be
leaking if they were filled to capacity;
o If all tanks are ever filled to capacity during the
year, then an estimated 35 percent of the tank systems
in the country are leaking at one time or another
during a year;
o Approximately 29 percent (.35 x .83) of tank systems
are leaking just after the time of product delivery the
way tanks are normally filled; and
o Approximately 18 percent (.35 x .52) of the tanks are
leaking at a random point in time.
Because of the nature of the test procedures used (i.e.,
overfilled tanks), we might be concerned about the possibility
that a large portion of the leaks could be near the top of the
tank and its associated pipe fittings, if that were the case,
test results in this study could overstate the percentage of
6This is a rough approximation which could be refined by
calculating highest and lowest fill levels for each tank
separately, and then computing the median and mean fill levels
as fuel is withdrawn. Fuel withdrawal rate could be assumed as
uniform over time or simulated from inventory data. Finally,
refinements could be made to account for the fact that the
assumption of uniform leak distribution over the surface of the
tank is not identical to uniform leak distribution over volume.
However, since actual leak distribution is unknown, such
refinements do not seem warranted at present.
8-17
-------
tanks leaking under operating conditions. While we do not have
information in our study on the location of the leaks, an
American Petroleum Institute report (API letter of March 27, 1981
from F.B. Killmar to Mr. Paul J. Sausville, P.E. Chief, Northeast
Sector Wastewater Management Bureau, New York State Department of
Environmental Conservation) provides some relevant statistics.
This API report found that among all 318 leaking tanks, only 4.4
percent had leaks limited to the upper one-third of the tank.
Even though the API study was based on volunteer reports from
leaking tank inspections rather than a representative probability
sample, these findings seem to strongly dispel any concern that
leaks are predominantly or disproportionately close to the top of
the tank. As a result, the 18 percent estimate given above for
percentage of tanks leaking at a random point in time under
operating conditions may be an underestimate.
8-18
-------
SECTION 9
STATISTICAL ANALYSIS
I. NATIONAL ESTIMATES OF THE NUMBERS OF UNDERGROUND MOTOR FUEL
STORAGE TANKS AND ESTABLISHMENTS WITH UNDERGROUND MOTOR FUEL
STORAGE TANKS
In the continental United States (i.e., excluding Alaska,
Hawaii, and the trust territories) there are an estimated 796,000
underground tanks currently being used for the storage of motor
fuel at an estimated 326,000 business, government and farm
establishments. Table 9-1 shows the estimated number of tanks
and establishments with tanks (as well as the mean and median
number of tanks per establishment) overall and within the
business/government stratum and the farm stratum. Overall there
are an average of 2.4 tanks per establishment with such tanks,
and a median of three tanks per establishment.
Among the business and government establishments (based on
analysis of the combined fuel establishment and large
establishment strata), there are an estimated 638,000 underground
motor fuel storage tanks in use at approximately 247,000
establishments. The average number of tanks per establishment
for the business and government sample is 2.6 tanks. The median
number of tanks per establishment is three. (The sample strata
are described in detail in Section 4 of this report.)
According to survey estimates, there are approximately
158,000 underground motor fuel storage tanks in use on about
79,000 farms, or an average of 2.0 tanks per farm with
underground motor fuel storage tanks. The median number of tanks
9-1
-------
Table 9-1. Estimates of the number of underground motor fuel storage
tanks and the number of establishments with underground
motor fuel storage tanks in the continental United States
(95% confidence bounds in parentheses)
Number of tanks
per establishment
Type of
e s t ab 1 i shment
Number of
establishments
with tanks
(1,000's)
Number
of tanks
(1,000's)
with tanks
Mean Median
Business & 247 638 2.6
government (220-275) (584-692) (2.4-2.8)
Farms 79 158 2.0
(58-100) (< 453) (< 5.0)
Total 326 796 2.4
(296-356) (503-1,090) (1.6-3.2)
9-2
-------
per farm is one tank. Because of the small sample size for
farms, the resultant large sampling weights, and the associated
broader confidence bounds, the farm sample was not combined with
the samples of fuel and large establishments in the succeeding
analyses. For a detailed discussion of the farm sample, see
Appendix G of this report.
A. Estimates bv Survey Region of the Number of Underground
Motor Fuel Storage Tanks and Establishments
The survey sample was stratified to provide estimates for
six predefined survey regions. Table 9-2 lists the six survey
regions and shows the estimated number of tanks and
establishments with tanks, plus the mean and median number of
tanks per establishment, within each of the regions. The
Northeast survey region has the highest estimated number of tanks
(186,000) and establishments (69,000). The survey region with
the smallest number of tanks is the Mountain region, with an
estimated 36,000 tanks at 14,000 establishments. The average
number of tanks per establishment is 2.6 tanks, and ranged from a
high of 2.7 tanks per establishment in the Northeast to a low of
2.3 tanks per establishment for the central region. The median
number of tanks per establishment for all regions combined is
three. The Northeast, Mountain, and Pacific survey regions each
have a median of two tanks per establishment. In the Southeast,
Midwest, and Central survey regions the median number of tanks
per establishment is three.
Table 9-3 displays the estimates of the number of underground
motor fuel storage tanks and establishments with motor fuel
9-3
-------
Table 9-2.
Estimates, by survey region, of the numberof underground
motor fuel storage tanks and the number of establishments
with underground motor fuel storage tanks in the continental
United States (95% confidence bounds in parentheses)
Survey
region
1
Northeast
2
Southeast
3
Midwest
4
Central
5
Mountain
6
Pacific
Total
Number of
establishments
with tanks
(1,000's)
69
(52-86)
48
(39-58)
53
(39-67)
27
(15-39)
14
(12-17)
35
(29-42)
247
(220-275)
Number
of tanks
(1,000's)
186
(175-198)
126
(106-147)
139
(108-170)
63
(43-84)
36
(23-49)
87
(58-116)
638
(584-692)
Number of tanks
per establishment
with tanks
Mean
2.7
(2.1-3.3)
2.6
(2.4-2.8)
2.6
(2.4-2.9)
2.3
(2.0-2.6)
2.5
(2.1-2.9)
2.4
(2.0-2.9)
2.6
(2.4-2.8)
Median
2
3
3
3
2
2
3
(1)ooes not include farms or tanks at farms.
9-4
-------
Table 9-3.
Estimates, by type of establishment, of the number'1) of
underground motor fuel storage tanks and the number of
establishments with underground motor fuel storage tanks
in the continental United States (95% confidence bounds
in parentheses)
Type of
establ ishment
Government
and military
Gas stations
owned by major
petroleum
companies
Gas stations
owned by other
companies
Other fuel-
related estab-
lishments
Large non fuel-
related estab-
ments (with
^20 employees)
Total
Number of
establishments
with tanks
(1,000's)
45
(29-62)
33
(26-41)
58
(50-67)
36
(30-43)
74
(55-93)
247
(220-275)
Number of tanks
per establishment
with tanks
Number
of tanks
(1,000's)
98
(69-128)
118
(87-148)
204
(174-233)
77
(64-90)
142
(97-187)
638
(584-692)
Mean
2.2
(1.8-2.5)
3.6
(3.3-3.8)
3.5
(3.2-3.8)
2.1
(1.8-2.4)
1.9
(1.6-2.2)
2.6
(2.4-2.8)
Median
2
™
3
—
3
-
2
2
3
~
(1)ooes not include farms or tanks at farms.
9-5
-------
storage tanks as they are distributed by type of establishment.
Establishment types include:
o Government and military establishments, including
state, local and federal facilities;
o Gas stations (SIC code 5541) owned and/or operated by
major petroleum companies (according to the
establishment manager);
o Gas stations (SIC code 5541) owned by other companies;
o Other fuel-related establishments (i.e., transportation
related industries); and
o Large non-fuel-related establishments (with 20 or more
employees).
Gasoline stations without regard to ownership account for about
half (321,000) of all tanks, but a little over a third (92,000)
of the total number of establishments with tanks. Of the five
establishment types, gasoline stations owned by other (i.e., non-
major) companies account for the largest number of tanks
(203,000). The other fuel-related establishments have the
smallest share of tanks (an estimated 77,000) and establishments
(an estimated 36,000). The large non-fuel-related establishments
category has the highest estimated number of establishments with
tanks (74,000 establishments), and has second highest estimated
number of tanks (142,000). The average number of tanks per
establishment, which is 2.6 for establishments of all types,
ranges from a high of 3.6 for gas stations owned by major
petroleum companies, to a low of 1.9 for the large non-fuel-
related establishments. The median number of tanks per
establishment is three for both of the categories of gas
stations, and two for the non-gas station categories.
9-6
-------
II. CHARACTERISTICS OF ESTABLISHMENTS WITH UNDERGROUND MOTOR
FUEL STORAGE TANKS
There are an estimated 247,000 government and business
establishments that use underground tanks to store motor fuel.
Table 9-4 displays the estimated number, the 95 percent
confidence bounds for the estimates, and the percentage
distribution of the type of establishment, within region. The
differences in the distribution of establishment types within the
survey regions are described in Section 9.II.A. In addition,
specific establishment characteristics were analyzed by region
and establishment type, and are displayed in Tables 9-5 and 9-6.
These tables are discussed in Section 9.II.B.
A. Distribution of Types of Establishments within Survey
Region
The distribution of establishment types is relatively
similar across the six survey regions, although there are some
differences, particularly in gas station ownership type in
certain regions. While gas stations owned by major petroleum
companies account for 13 percent of the establishments across all
regions, 23 percent of the establishments in the Pacific region
were gas stations owned by major petroleum companies. In
contrast, the lowest proportion of gas station establishments
owned by major petroleum companies (8%) occurs in the mountain
regions. Gas stations owned by other companies (i.e., not owned
by major petroleum companies) account for 24 percent of the
establishments in all survey regions, but for only 11 percent of
the establishments in the Pacific survey region. The Midwest
survey region has the highest proportion (29%) of gas stations
owned by other companies.
-------
in
in
o
Vj
o en
CO 4J
C
-> 0)
H E
-" X
eo eo
-ii -H
C i-H
CO X
4J CO
4J
o) en
60
C -H
3 UH
C
x o
4-1 O
•H
5 dP
in
en ON
4J •—
C
0)
E^
XCM
•H en
rH C
X 0
CO -H
4J t>0
en o)
0) IH
>
4-1 lj
C 3
o> co
u
IH X
0) -H
ex en
•O 0)
C X
CQ 4J
VH H
0) O
X >4H
§ en
C 0)
a
0) >-,
X 4J /•>
4J in
4J 0)
tH C 10
O 0) 0)
E X
en x 4J
O VI C
4J -H 0)
CO rH IH
E X co
•H CO O,
4-1 4J
in w c
W 0) -H
-3
ON
0)
rH
X
CO
H
rH
CO
4-)
O
H
en
•a 4->
i o) C
C 4-1 0)
0 co E
C rH JS
0) en
0) IH -H
60 i rH
IH r-l X
CO 0) cfl
_] 3 4J
<4H en
0)
•O en
0) 4-1
4-> C
CO 0)
rH E
IH 0) X
0) IH eo
X ' -H
4J rH i— 1
O 0) X
3 to
OH 4J
en
0)
M
en o>
C X
O 4-) en
•H O 0)
4J -H
CO r*i C
4J X CO
en p.
•0 E
en o) o
cO C U
O 5
O
IH
en o
C T-)
O ce g en
•H E 3 0)
4-> a> -H
<0 >M-H C
4J X O CO
10 IH Ou
•O 4J E
en o> 0) O
CO C (X 0
OS
o
en
r*N 4J
4J IH C
C CO 0)
0) 4-1 E
E -rl X
C rH 10
rH >rH *H
0) E rH
> X
O TJ CO
O C 4J
co in
a>
C
o
•H
60
0)
r-l
>,
E
w
^
o
o
dp 0
O O -
O O vo
O rH 00
«. 1
ON O
vo O
o
CM
in
V— '
^-^
o
0
o
o
O dP ON
O rH CM
- ro i
rH O
CM O
O
ro
r-l
v^
, — .
0
o
O
O
O dP vo
O VO rH
- rH 1
rH 0
rH O
O
VO
*— '
^-^
O
o
o
0
O dP O
O -* CM
- CM i
r~ o
rH O
0
ro
rH
""^
^^
O
o
o
o
O dP rH
O O rH
- rH i
r- O
O
O
ro
^— . '
/-^
O
o o
O dP O
O ON
- rH r-.
CO CM
rH
V
•»— '
en
TJ
.. o
C X
0
'oOMH
4J •• 01 C
en IH IH o
co 0) U
0) X C
X E i-i dP
4J 3 m
IH 2 dP ON
Q
Z
0
o
o
O dP 00
o o m
O O i
- rH O
00 O
•* O
ON
CO
^^
^*,
o
o
o
O 00
O dP CM
o m i
- ro O
r— o
rH O
m
**^
s-^
O
o
o
o
O dP 00
O CM i
- rH O
vo O
o
U
0) X C
X E 1-1 dp
4-i 3 in
3 Z dP ON
Q
e/3
/— V
o
o
o
O dp -
O o r~-
O O vo
- rH 1
CO O
m o
0
ON
ro
^^
O
o
0
o
O dP ON
O ro CM
" CM i
CM O
rH 0
O
sj-
^^
O
o
o
o
O dP CM
O U-l rH
- rH i
00 O
o
o
.3-
v^
^^
o
o
o
o
O dP O
O ON CM
- CM 1
vo O
rH O
O
CM
rH
^"^
*-v
o
o
o
o
O dp CM
O -3- rH
- rH i
r- O
O
o
ro
S-'
^^
O
o
o
o
O dp in
O ON rH
* rH i
O O
rH O
O
»
VO
**^
in
•O
•• 0
C X
0
bOUH
•• 0) C
»H (H O
4-10) U
W X C
a> E •H dP
> 3 m
•O Z dP ON
•rl
^-v
o
o
o
O dP -
O O ON
O O co
- rH i
Is- O
CN O
O
in
iH
^^
O
o
o o
O dP -
o ON m
- CM rH
r~
V
^~*
^^
O
o
o
o
o dp m
o m i
- rH O
-3- O
O
ro
*— *
^•N
O
o
o
o
O dP O
O in rH
- CM 1
r-- o
O
O
E — i dP
C 3 m
3 Z dP ON
O
£
^^
O
o
o
0 dP -
O O CM
O O -*
- rH 1
in o
CO O
o
ON
CM
«^'
^^
O
O
o o
O dP -
o in ON
- ro rH
CM i
rH O
O
O
r*-.
N-X
^-^
O
o
o
o
O dP VO
O vo i
- rH O
vo O
O
m
^^
*~**
o
o o
O dP O
O i-H -
• rH OO
•3" i
O
^^
O
O
o
o
O dp O
O rO rH
- CM i
CO O
o
o
..
VO
>*rf»
X— s
o
0
o
o
O dP ON
o in i
- rH O
m o
o
-
CM
\^
in
1
H|
o
•H
OOMH
•• 0) C
IH IH O
O 0) U
•rl X C
KH E -H dp
•H 3 m
CJ Z dP ON
co
BH
O
O
O
.
o dP m
O O r^
O O CM
- i-H i
r- o
•3- o
CM 0
o
CM
CM
^^
^^
o
o
0 0
O dP -
O O ro
- ro ON
•3- i
r- o
0
o
in
in
v^
s~*.
o
o
o
O ro
o dp dp CM
O 00 VO
•• rH 1
i/"> O
•3- 0
O
~
o\
CM
N—'
eo en
C "O
o •• c
•ri in 3
00 CO
0! OX
Vj "^
60 •
>, 0) <4H
0) •• M C
> M 0
VH 0) rH O
3 X rH
wen*
3 in
rH Z dP ON
rH
t>0
C
•H
"O
c
3
O
IH
O
4J
VH
O
•H
IH
a
• TJ
en a)
E 4J
IH CO
CO rH
HH 3
O
0) rH
T) CO
3 0
i-H
U 0)
C IH
•H CO
4-1 en
O 4J
C C
0)
in o
0) rH
O 0)
Q CM
/~s >-v
rH CM
^ Nw^ Nrf'
9-8
-------
The Central region has the highest proportion of government
and military establishments, with 24 percent compared to 18
percent for all regions. The Midwest survey region has a
slightly lower percentage of large non-fuel-related
establishments, with 23 percent compared to 30 percent for all
regions. The proportion of other fuel related establishments is
nearly identical across regions.
B. Selected Establishment Characteristics, Analyzed by
Region and by Establishment Type
Selected establishment characteristics were analyzed by
region and by establishment type for the establishments with
underground motor fuel storage tanks. These analyses were based
on a file of 876 establishments, weighted to represent 247,000
establishments. The percent of establishments exhibiting each of
the selected characteristics is displayed in Table 9-5 (by
region) and Table 9-6 (by establishment type).
1. Establishments with Underground Waste Oil Tanks
Overall, 31 percent of the establishments have underground
waste oil tanks on site. Most owner/operators knew whether or
not their establishments have waste oil tanks (over 99%). The
Central and Southeast regions have the smallest percentage of
establishments with underground waste oil tanks (24% and 25%
respectively) and the Pacific region has the largest percentage
(43%). Among establishment types, gas stations of both types
have higher percentages of underground waste oil tanks than other
establishment types. More than half (59%) of the gas stations
9-9
-------
•o
a)
o
QJ
O
10
_.
4J
*•>
3
H
_s
CO
C
u
0)
bO
2
M
0)
3
M-t
i-i
O
4J
i
•o
o
00
14
1
3
4J
tn
4J
C
1
•^ M
-3 *
rH W
j 1 ^"*
" 01
* u
•s s.
jj.5
c
a|
rt! ^
3
p. o
0) -0
•H •!-!
s" s
") M
H *
JJ °
CO
m
Ov
01
i-H
43
H
to
"W C
O 01
C 'S
01 -H
O i-H
01 CO
a, jj
CO
01
CO
JJ
O 01
E
C to
0) -H
O rH
U 43
01 CO
&4 JJ
CO
01
M
jj
<*H C
O 01
u e
C to
01 -H
I-" 43
01 CO
CL4 JJ
to
01
to
JJ
"w C
°|
El
0) 1H
O rH
(£ JJ
to
01
t[ [
o
JJ
c
01
u
01
64
01
u
£
CO
M
3
to
c
JJ
•rl
*
0)
CJ
c
W
C
•H
r!
JJ
•H
3
o
•o
01
'3
cr
01
S
•o
01
•H
3
tr
01
jj
rj
01
10
-H
rH
CO
CO
01
O
JJ
C
CJ
14
01
&4
MH
O
JJ
01
u
01
BH
p
0
j^
01
>
o
u
o
yj
t
o
o
o
^
c
CO
JJ
>
CO
4=
-*
CO
>
CO
CO
O
01
01
3
JJ
C
to
rH
43
CO
JJ
01
to
JJ
c
01
J5
CO
•1-1
rH
43
01
JJ
to
01
O to
JJ rH
O rH
a
C to
01
"O rH
•O 01
3 3
CO ttH
O to
JJ rH
13
a
c to
01
•O r-l
•O 01
3 3
to OO *P CM C*P m
in ^ CM ' rH i
Csl rH rH
S_X ^^ N^X
dP dP dP
rH >tf fn
X-N
dp dp 0
ON rH O
d* 00 dr> ON d* rH
CM i CO ' OO i
oo m r* m co m
r^ ^o r^
dP dP dP
d*> tO dr> CM tfP in
•H rH in CM tO CM
r^ oo r-.
dP dP dP
ON ON CO
g1? K^ S1?
en rH CM o en co
CM CM CM
4J 4J
M M JJ
trj a) w
QJ 0) Q)
oox
Z CA
dP
en i
tO rH
^^
dP
•xf
en i
f* CM
m
^-^
dP
d* v£>
in CM
**''
dP
en I-H
rH
dP
O
en i
m
dP
dP O
r* en
*^s
dP
CO
CM o
rH
rH
* C
a
dP
ON
eJP ON
00
^_^
dP
diP 0^
O \D
ON
^p
m
dr> m
rH
dP
CM i
\c o
<*•
dP
o
rfP ON
CM '
m
dP
m CM
in
dP
CM
CM t
CM
ON
ON
CO
r*
m
to
d*>
CO
rH <
CO VO
£,
dP
ON *
CM
dP
*P ON
in i
v-"
^^
dp
00 CO
r* <
en
r-
dP
gD rH
•3- '
rH
NaX
dP
in
rH I
en r^
CM
rH
S
O
H
VI
Cfl
1
y
C
o
c
CO
1
tQ
/-s
rH
9-10
-------
*o
Q)
4J
U
— '
C
CO
4J
0)
00
CO
o
4J
W
to
3
M-t
0
o
E
•O
B
3
O
no
•o
H
X
3
w
4J
1
n
*^
•§
3 '"-
u »
U CO
ofj
4J C
g £
-U
" n
*s *^
w E
'j
i§
r"1 U
4J °
1 «
*rl r*
4J *
VD
i
ejv
0)
rH
X)
U 01
4J CJ i V4
CM E 6 E 0 0]
O 01 CO O 4J rH
E -1 C O rH
w X 3 6 -I
B W CO rl O.
01 -H B 01 B CO
U I-H -r< > o
14 XI O 17 i-H
01 to xi cj *o 01
0* 4J 4J 33
CO T< O CO UH
01 9 4-1
to 01
4-1 O 14
IH E E O to
O 01 CO IH U i-H
4-1 _c 3 r* € *"H
E n co o a
01 -rl E O B 10
O rH -H 01
rl X) 0 T) r-l
01 CO _*"• 4-1 *T3 01
Cu 4J u 33
m %
V f
to
4-1
CM E o
o o 4J ^: no
E E E CO
4J X! T3 CO «r4 01
E CO 01 iJ 4J CO
01 -rl U CO E
O I-H 01 0
o co tr to a -H
P-t 4J 01 X! O rH
CO H
a
CO
u E
UH E 0 O
O IP 4J M iH
E E 4J CO
4-1 X! "O CO CO 4J
E tO 01 4J rH .rl
01 -rl IH rH E
O rH -H 01 O IH
rl XI 3 > " 01
01 CO CT CO « CL
CM 4-1 01 X! E
01
XrH
CO O i-H
4-1 O iH
IM E E t-i MH
O 0) co to ^f
E o) 0) o
B to U X>
01 fH IH
U I-H «l O CO
H XI rl CO
£ 4J x! E -O
CO S O 01
01 tQ CO
CO -O
4J 01
IH B E
o a) o
JT3
B to
B to CO J oo d* oo
' 00 i •* '
in ^o r^
s^. ^^ ^^
/~x /~\
dO CM 00
rfr O «*» SO 6* j
m %J CM
xs, ,— » ^s
dP dP dP
r-^ a* vo
d* < d* o> d* r-
m • r*» i r-i i
en QO co in *»o ^3"
C ^ 2.
<"~s »-~v x-v
dP dP dP
CM o^ en
M3 i CM ' 0» •
Is* O ON \O 00 -31
fv OO 00
^^
»—»»—» dP
dP dP ON
«»> vo d* P- d» CM
•H iH «T i CM '
rH i rH CM SO
V— V ,»—v y^v
dP dP dP
fH rH m P*. *» OO
rH -d" (**^
O V
X W •»-> W »C
ij M ft S no
B U -n B to -n co
OliH 4J X 3 41 4J >, 4)
Bi-H COXIVlH tOXI-rl
5 M U rH E U B
rlB I013OCO 10'OCO
> T3 toEUB* tOCB
OB a f o o a y o
Oft) UOP~U O O O
CO
H
^^
dP
O
d* CO
en i
VD
dP
•* 00
O i
ao en
r-
s^
dp
d» -9
CM i
en o
CM
^
dP
dr r^
O i
2,
^^
dP
m
d* co
SO 1
r* co
dP
d* •--
«H i
00
,*•*,
dP
vo
ON i
CM m
CM
.£>
Other fuel-
related esta
lishments
^^
dP
rH
Or 00
r- m
dp
m
CO *
oo o
oo
dp
«)*» en
CM i
dp
in
«)P r<*
Ut CM
<,-s
dP
O
dP oo
vo i
VO .a
Large non fu
related esta
lishnents
/-v
dP
m
ON i
SO
-------
owned by major petroleum companies have underground waste oil
tanks, and 46 percent of the gas stations owned by other
companies have underground waste oil tanks. Government and
military establishments and large non-fuel-related establishments
have the smallest percentage of establishments with underground
waste oil tanks, with 17 percent and 16 percent respectively.
Approximately 98 percent of the establishments that have
underground waste oil tanks have only one of these tanks.
Altogether, there are an estimated 78,000 underground waste oil
tanks at establishments that also have underground motor fuel
tanks. (This estimate does not include waste oil tanks that may
be at establishments that do not have underground motor fuel
storage tanks, since those establishments were not surveyed.)
2. Establishments with Abandoned Underground Tanks
Approximately 14 percent of the establishments that have in-
use underground motor fuel storage tanks also have one or more
tanks on site that have been abandoned (i.e., are no longer in
use, and have not been removed.) Again, most owner/operators
(95%) knew whether their establishment had such a tank on site.
Across regions the percentages of establishments with abandoned
tanks is similar, ranging from a high of 17 percent in the
Central region to a low of 9 percent in the Pacific region. A
more striking difference can be seen in the percentages of
establishments with abandoned tanks across establishment types.
Only 4 percent of the gas stations owned by major petroleum
companies have abandoned tanks, while 22 percent of the gas
stations owned by other companies have abandoned tanks.
Nearly 70 percent of the establishments with in-use motor
fuel tanks that have abandoned underground tanks have only one
9-12
-------
abandoned tank, but the number of abandoned tanks in the survey
sample ranges from one tank to ten tanks per establishment.
Overall, there are an estimated 53,000 abandoned underground
tanks at establishments that have in-use underground motor fuel
storage tanks. (Note that the estimates provided in this section
are based only on establishments with in-use underground motor
fuel storage tanks. The numbers and percentages given here do
not include abandoned tanks at establishments that do not have
in-use underground motor fuel storage tanks.)
3. Establishments with Clean Sand or Gravel as Tank
Backfill
Overall, 78 percent of the establishments claim to have used
clean sand, pearock or peagravel for backfill around their tanks.
(This estimate is based on the three-quarters [74%] of
owner/operators who knew the backfill material.) The remaining
22 percent used excavation soil, rubble, clay and combinations
including these materials to fill around their tanks. In the
Central and Pacific regions a lower percentage of establishments
used the clean sand/clean gravel backfill (63% and 65%
respectively), while in the Midwest the percentage using this
type of backfill was higher (88%). A higher percentage of both
types of gas stations used clean sand/clean gravel backfill.
Among gas stations owned by major petroleum companies, 92 percent
used clean sand/clean gravel as backfill, while among gas
stations owned by other companies 89 percent used this type of
backfill. Only 66 percent of the large establishments used clean
sand/clean gravel as backfill.
9-13
-------
4. Establishments Required to Have Tank Installation
Permits
Operators of 56 percent of the establishments with
underground motor fuel storage tanks did not know whether the
establishment was required to obtain a permit for the
installation of the underground tanks. Among those who did know,
56 percent said they were required to obtain such a permit. The
Central region has the lowest percentage of establishments
reporting a required installation permit (37%) while the Pacific
region has the highest percentage (78%). Among gas stations
owned by major petroleum companies, where this question was
answered, 87 percent said they were required to have installation
permits. However, among government and military establishments,
where this question was answered, the owner/operators said an
installation permit was required for only 33 percent of the
establishments.
5. Establishments Required to Have Tank
Operating Licenses
Whether the establishment was required to have an operating
license was known by 84 percent of responding owner/operators.
At these establishments, only 29 percent of the establishments
overall are required to have tank operating licenses, according
to the establishment operators. Differences in the proportions
of establishments required to have licenses are noticeable across
survey regions and establishment types in patterns similar to the
differences in percentages of establishments required to have
installation permits. Again the Pacific survey region has the
highest percentage of establishments requiring licenses (49%, or
two thirds more that the overall percentage), and the Central
9-14
-------
survey region has the lowest (15%, or about half of the overall
percentage.) Only 11 percent of the government and military
establishments are required to have operating licenses, while 55
percent of the gas stations owned by major petroleum companies
have such licenses.
6. Establishments with Insurance Coverage for Sudden
Motor Fuel Spills
Among all regions and establishment types, 81 percent of
establishment operators who answered this question believe they
have insurance to cover damage to people or property caused by
sudden motor fuel spills. Most (85%) did answer the
questionnaire item. The Mountain region has the highest
percentage of establishments with sudden spill coverage, with 96
percent of the establishment operators claiming to have this type
of coverage, closely followed by the Pacific region, with 92
percent of the establishment operators claiming to have this type
of coverage. Among the five establishment types, government and
military have the lowest percentage of establishments covered for
sudden spills (70%) and large non-fuel-related establishments
have the highest percentage covered, with 88 percent of the
establishment operators claiming to have this type of coverage.
7. Establishments with Insurance Coverage for Non-
sudden Motor Fuel Spills (Including Leaks)
A total of 69 percent of the establishment operators who
answered this item believe that their establishments have
insurance to cover damage to people and property caused by non-
sudden spills of motor fuel (such as tank system leaks). (Since
9-15
-------
this type of insurance coverage is not yet common, this belief
may be unfounded.) A slightly lower number (78%) answered this
item than answered the "sudden spills" question. A higher
percentage of the operators in the Mountain and Pacific regions
believe that their establishments are covered by insurance for
non-sudden motor fuel spills (94% in the Mountain region and 89%
in the Pacific region). There was very little difference in the
percentage of operators claiming non-sudden spill coverage across
establishment types.
III. CHARACTERISTICS OF UNDERGROUND MOTOR FUEL STORAGE
TANKS
There are an estimated 638,000 underground motor fuel
storage tanks at business and government establishments in the
continental United States. Table 9-7 shows the numerical and
percent distribution of these tanks by survey region and
establishment type. The differences in these distributions
across survey region are described in Section 9.III.A below.
Selected tank characteristics, including tank age, tank capacity,
products held in the tank, and various installation
characteristics, were also analyzed by survey region and by
establishment type. The results of these analyses are reported
in Tables 9-8 through 9-15, and are discussed in Section 9.III.B.
A. Distribution of Tanks by Type of Establishment within
Survey Region
As is shown in Table 9-7, tanks at government and military
establishments account for 15 percent of all of the underground
motor fuel tanks. This proportion is about the same across all
9-16
-------
4-1
CO
01
s/
B
CO
H
4J
cd
to
K/
B
CO
H
4J
CO
01
^
B
H
01
B
4-> O
CO i-l
4J
01 CO
»y> 4J
B 01
CO
EH 01
CO
00
4J
CO
01
B
CO
H
rH
CO
4->
O
H
01
*O 4->
i o> B
B 4J 41
O co B
B rH J5
0) 01
0> rJ -H
bO i rH
rJ rH J3
CO 41 CO
rH 3 4J
VH 01
4)
01
•C 4J
4) B
4J 0)
CO B
VJrH.fi
41 4) 01
Xi >J -H
4J 1 rH
0 rH J2
4) CO
3 4J
41
VJ
01 4)
E JS
O 4J 01
•H O 0)
4J i-l
(Q ^*, £
4J XI CO
01 (X
•0 B
01 4) O
CO B O
oo 5
o
V-l
o
•rn
CO E 01
B 3 4)
4) t-l
^*i rH B
X> O CO
M O-i
T) 4J B
0) 41 O
E a o
o
to
r"*i 4J
4J l-t E
E <8 0)
0) 4-1 E
B rH 0)
rJ -H -H
4) B rH
0 T3 18
bO E 4J
to 01
41
E
o
1-1
00
41
M
^
4)
W
O
O
O
»
r-
00
o
o
o
>*
-*•
o
0
0
A
r»
CM
o
o
o
•
rH
vo
O
o
0
«
ON
CM
O
o
o
in
CM
4_> ..
10 r-l
tO 4)
4) X>
* i
H Z
o
Z
0
o
o
00
dP ON
O rH
O i
rH O
O
O
m
p^
£,
^^
O
o
0
dp oo
-* vo
CM 1
0
O
O
*.
O
CM
^
O
o
o
dp in
-3- en
rH 1
O
O
O
oo
tH
^**
O
o
o
dp -J
en r--
en i
O
O
O
*.
00
s
^^
O
O
O
dp O
vo in
rH i
O
O
O
ON
"**"
^*.
O
O
o
dp ON
-* -3-
tH 1
O
O
o
CM
^^
, ,
01
•0
B
3
•• O
B X>
O
•H
bOM-l
01 E
I-l O
U
B
•H dp
in
dp ON
O
O
O
O ^
O dP -3-
O O rH
- O '
VO rH O
CM O
rH O
VO
O
tH
/— s
O
o
o
o
O dP CN
o vo in
« CM i
en o
tn o
0
•.
-3-
~
^
O
O
o
o
O dp r^
O ON rH
* 1
CM O
tH O
o
VO
^
x-s
o
o
o
0
O dp O
O CM in
• en i
O O
•* o
Q
1.
rH
en
x-x
O
O
O
o
t^ ON en
- rH 1
sr 0
CM 0
O
CM
rH
X-s
O
o
o
o
O dP in
O •• 01 B
01 r-l r-l O
CO Ol U
4-> 3 m
3 Z dp ON
0
O
O
o
»
ON
tn
i-4
O
o
o
o*
CM
O
o
o
A
vo
O
o
o
•t
CM
in
O
0
O
in
CM
O
O
o
vo
CM
M
^) Q)
01 XI
P
£
O
0
O
o"
dP r»
O rH
O i
rH O
O
o
CO
O
tH
s-^
O
o
o
dP ON
en en
CM i
O
O
O
K
CM
^
0
O
o
dP •*
m CM
rH 1
O
O
0
ON
%""'
/-N
o
0
o
dp in
ON VO
CM I
o
O
O
„
ON
tn
^s
O
o
o
dp O
rH I
O
o
o
ON
^-^
O
o
o
dp r--
ON en
rH 1
O
o
o
^
rH
. .
to
E
3
•• O
B XI
0
•H
OOU-I
4) E
r-l O
O
B
•H dp
m
dP ON
Jlf^
O
8
O dp -
O O -3-
O O CO
• rH i
«n o
vo O
O
en
•*
^^
O
O 0
O dP O
O -* -
• rH ON
ON rH
V
o
o
o
o
O dp O
O rH 1
0 O
tH O
O
-31
"— '
^^
o
Q
O dP OO
O O rH
« CM i
tn O
rH O
„
OO
. .
1
3
.. o
E X>
O
•H
bOM-i
•• 01 E
M M O
rH 41 O
CO XI B
»-i e -H dp
4-1 3 in
B Z dp ON
0)
U
/-N
O
8
O dP -
O O ON
O O -*
- rH i
vO O
en o
0
en
CM
"**
^^
O
o
o
o
O dp rH
O O rH
- CM i
t>- O
O
O
„
en
~
^*^
o
o
0 O
O dP -
O en vo
- rH 1
in o
o
o
^
"-'
^^
o
Q
o
Q dp •$
O VO CM
- en i
en o
O
CM
^
X«s
O
0 0
O dP O
0 rH •
• tH tH
•* rH
V
s~*
o
o
o
o
O dP rH
O O rH
- CM i
1^ O
O
O
en
\^
, .
0)
•o
B
3
E XI
O
•H
00 IH
•• 4) B
B l-i M 0
•H 01 O
tO XI C
4-1 B -H dP
E 3 m
3 Z dP ON
O
£
O
o
o
O dP vo
O O rH
O O rH
« rH 1
f~ O
CO O
o
00
in
^
s~*
O
o
0 0
O dP -
O en en
- en in
00 '
01 0
o
o
0
en
O
O
o
o
O dp O
O O rH
- rH i
CO O
8
f*^
v-'
s^t
o
o
o
o
o dp r~
O VO CM
- rH i
-» O
i-4 O
O
rH
^
^-v
o
o
o
o
O dP rH
O ON en
« CM 1
VO O
CM O
O
o
CM
^-s
o
o
0
o
O dP ON
O CM r-l
- rH i
O O
^ 8
CM
>^x
01
•o
E
3
.. o
B X>
O
•H
oo VH
• • 4) B
M M O
O 01 O
id fa 5 dp
•H 3 in
0 Z dp ON
co
PM
O
O
O
O dp CM
O O ON
O O vo
• rH 1
09 O
en o
vo o
-3
oo
in
^^
o
o
o
o
O dP r^
O CM 00
- CM tH
CM i
•* O
r-l O
o
r^
ON
O
O
O
O O
O dP ON
O CM i
- lH O
1^ O
f» O
^
vo
O
O
o
en
O en
O dP CM
O CM i
- en O
•a- o
0 O
CM
,^
r^
rH
o
o
o
O 00
O df 0 BO
4) 0 XI
r-l -H
bO .
>-, 4) M-l
4> •• U B
> * 0
r-l 4) rH O
3 XI rH
01 | CO dP
3 m
rH Z dP ON
rH
<
00
E
T3
B
3
O
f-l
o
4J
I-l
O
•H
^
cx
• T)
IT Q)
E JJ
V. tfl
CC rH
(4^ 3
O
4) rH
•O co
3 0
rH
U 4)
e ^
.H CO
4J 01
O 4J
C C
4>
to y
41 kl
C 41
C O,
rH CN
W v_/
9-17
-------
Table 9-8. Estimates, by survey region, of the mean and median
age (in years) of underground motor fuel storage
tanks (•*•' (95% confidence bounds about the mean
in parentheses)
Survey region
1
Northeast
2
Southeast
3
Midwest
4
Central
5
Mountain
6
Pacific
Total
Mean
tank age
(in years)
12
(11-13)
12
(11-14)
12
(11-14)
12
(9-15)
11
(9-14)
12
(7-18)
12
(11-13)
Median
tank age
(in years)
12
11
11
10
11
11
11
(1)Does not include tanks at farms.
9-18
-------
Table 9-9. Estimates, by establishment type, of the mean and
median age (in years) of underground motor fuel
storage tanksI1' (95% confidence bounds about
the mean in parentheses)
Mean Median
tank age tank age
Type of establishment (in years) (in years)
Government & military 12
(10-14)
Gas stations owned by 12 11
major petroleum (9-14)
companies
Gas stations owned by 14 13
other companies (13-16)
Other fuel-related 12 11
establishments (10-13)
Large non-fuel- 10
related establishments (9-12)
Total 12 11
(11-13)
not include tanks at farms.
9-19
-------
Table 9-10. Estimates, by survey region, of the mean and median
capacity (in gallons) of underground motor fuel
storage tanks^1' (95% confidence bounds about the
mean in parentheses)
Survey region
1
Northeast
2
Southeast
3
Midwest
4
Central
5
Mountain
6
Pacific
Total
Mean
tank capacity
(in gallons)
4,583
(3,662-5,503)
5,744
(4,931-6,557)
5,710
(5,064-6,357)
5,176
(4,751-5,601)
5,866
(5,711-6,020)
6,180
(4,954-7,406)
5,405
(5,026-5,783)
Median
tank capacity
(in gallons)
4,000
6,000
5,000
4,000
5,000
6,000
4,000
not include tanks at farms.
9-20
-------
Table 9-11.
Estimates, by establishment type, of the mean and
median capacity (in gallons) of underground motor
fuel storage tanks* ' (95% confidence bounds about
the mean in parentheses)
Type of establishment
Mean
tank capacity
(in gallons)
Median
tank capacity
(in gallons)
Government & military
4,342
(3,059-5,626)
2,000
Gas stations owned by
major petroleum
companies
6,821
(6,120-7,522)
6,000
Gas stations owned by
other companies
5,093
(4,513-5,674)
4,000
Other fuel-related
establishments
5,687
(4,963-6,410)
4,000
Large non fuel-
related establishments
5,261
(4,374-6,149)
4,000
Total
5,405
(5,026-5,783)
4,000
' 'Does not include tanks at farms.
9-21
-------
•d
CU
M
O
4J
to
4-1
CO
rC
4-1
M x-v
W
-!,„ 0)
<" tn
•^ 0)
C jQ
« t;
w C
01
Sn r4
s? <«
03 o.
i-l
0 C
J-1 -H
Cfl
__1 M
^ T3
2 c
3 3
4H o
L, -°
r4
0 cu
w o
i c
E 01
_. T)
? 'H
U n |
g C
S °
fe0
s *
•S ^
? ON
c ^
3
ti, r*
«« cfl
o a)
, 1 >N
4-1
% ^
0 °
P 'H
^ H
OJ ^
CXc
... 0)
^ £•
0 4J
r-" C
g .H
0 j-
•r< 4J
60 .H
4) .j
»-l *
. to
J"N 01
v a
£ >^
5 >->
3
W rH
. CD
>•> 3
J3 i^
" T3
m <"
* -H
tJ ^
* .H
E o
'Jl «
™ a
" Cfl
w
CM
rH
1
ON
0)
rH
,0
03
i— 1
4-14-1 Cfl 0)
O 03 CU 3
rC P.1*-!
4-1 4-1 0) P^,
C (-1 4-> S-l
CU 10 O O
O ril! 4-1 rJ 4J
rJ C Cfl (U O
01 cO ,£ E
CM 4J 4-1
O <4-4
O
U-l 4J
O 03
-C rH
4J 4-1 CU O
C r-l f.
0) Cfl O O
O J >
03
rH
4-1 4J 0)
0 cfl 3
f, 4H
4J 4J a)
C r4 rH
o) tn o cu
O Jrf 4J W
»H C CO 0)
01 CO MH
D-i 4J T3
C4H 4->
O 03 T3 01
A
C4H 4J
O CO 0)
rC T3 C
4J 4J 0) 01 >H
C rW "d rH
cu w o rt o
o ^ 4J cu tn
1H C Ul rH 03
CU CO bO
O-i 4J
>N C
0) 0
> -H
rJ bO
3 01
e/> >H
X-N X-X
tff> OP dP
rH rH dP -J
CM i
V V rH
x-^*
CO s-*
^ dP
dP diP CM
O CM i
rH
s.x'
X-s X— v
dP dP
d* CO d* co
CM i CM i
rH rH
•^s ^_x
X^ X— x
dP dP
CM ON
di» co diP CM
co i co i
CM CO CM OO
rH rH
S** x— s
dP dP
ON ^3-
d* in d* -t
OO i ON i
-* VO CO -J-
co co
^^ N.X
x^ ^^
dP dP
VO VO
d* co d* CO
00 I CO I
CM ON CO O
rH CO
N_x ^x
4-> 4J
tn tn
tfl 03
01 01
rH JS CM J2
4-1 4-1
r4 3
0 0
Z on
X— s
dP
diP VO
CO
V
v»x
x~s
dP
di° vo
co
V
***s
X-N
dP
diP -J
vo i
co O
rH O O
V
N^X
X~N X— S X-V
dP dP dp
d*> in d«P oo dP -J-
CO I in i rH
r-l CO V
S^ N-X V^-
x-s x— s
dP dP x-x
O i-l dp
dP co dfl co dP oo
co i in i vo CM
CM VD CM ON rH i
rH fH ^f
*^x ^— X \^f
X— v X— N y — s
dP dP dP
ON -J
co i ~3- i r-i
co t— I co co co co
CO CM CO
^_x »~-x ^^x
c
rH -iH O
03 03 -H
(-1 4J <4H
-3-4-1 in C VO T-I
C 3 O
CU O CO
U S CM
x^
dp
diP CM
r-l 1
r-l
N~X
x— v
dP
t*° CM
r-l I
O
•^*
x^
dP
dP co
CM i
r-l
\^s
X-N
dP
VO
dt> CM
r-l 1
CM r-
r-l
v^
x^
dP
VO
Of co
CO i
CO O
CO
s^x-
1-4
03
4-1
O
H
r-l
0)
3 co
<4-l T3
C
cu 3
C 0
0 rO
C 0)
co o
rC C
4-1 01
•o
0) -i-l
r-l U-l
0 C
E 0
O
0)
S-l O
0 C
4-1
tn
01
O o
4-> C
o>
•d K
cu
tn
3
r-l
CU 01
r4 3
CU 4-4
S
C4-I
CO O
o.
C a)
CO P4
4-> >N
4-1
01
E to
O -H
CO rC
4J
CU
en bO
3 C
CO -H
0 IH
"CO O
& • 4J
>-, to
4-1 01
C > 01
0) rJ r-l
u 3 a.
!-i tn E
a) 03
a. a) ca
rC
O 4J CO
O
r-l 0 C
4-1 O
• C "H
tn to r-i bO
e j: o v
rJ 4J vH )M
03 (H
in cu fr. w
r-l -r-l
C 0 tfl ,C
0 E rC 4J
4-1
tfl O C C
r« 4-1 O -l-l
C E
cfl E T>
4-) 3 CM C
tfl r-l 3
cu O
•d >1 0> 4H
3 co £, -a
rH E 4J 0) Ol
O >-l 4J
C 4-1 bO CU to
•-H C C S E
0) "H >r-l
4-> O (-1 « 4-1
O rJ 3 J 4J cu
a> s a- i-i
O O (x, O 0)
Q oi 4-1 z s
rH CM CO
S«X ^rf' — rf*
9-22
-------
.
*\
cu
^
o
J— )
W
4J
cd
.•C
4-*
M
H
CO
**
^ ,^-s^
5 W
0)
bo £
(fl fj
n C
2 «
w b
D-
""""^
c
lii *r^
LJ W
O *O
4J C
O
-O
"2 r^
cfl ^
p >H
5 o
4J 1>
W ^
CO
I—I
ON
0j
,— t
«0
H
r-l
M-l 4-> W 3
rC CXM-I
C V-i JJ )-i
w
V-i C w nj
0) 03 txO
P-i 4-»
rH
0)
tW 4J d
O cfl M-i
4J 4J 0) C
C VJ 0
0) tO O -r-t
fj frj 4_l 4,1
M C W Cfl
PM 4-* K*
H3
,__!
U-l 4J 0)
o (D d
(-^ t[ (
4J 4-> a)
C rl r-l
a) w o a>
U ^ 4J I/l
l-l C 10 V
4-> 01 T3 -H
C M (fl r-l
0) in o a> o
CJ J«! 4J r-l W
ri C w C n)
0) (fl 3 bO
PLI 4-1
4-1 4J
O (fl 0)
x: -o c
4-1 4-1 (!) ^i
X) 4-1
(fl
4J
in
Uj
^-y
^— x CO x™v
dP — dP
dip oo c*P ovo CM
-1-0 r^
_v _v
^X /^~S ^~N
dp dP dp
rfp CM tf£ VO df CO
r—l CO r-l
V V V
X-N /*^i /^N
dp c^> co
f^O ,— j (^p ^>
v£) i O O
.—1
*+*/
s-^
rfP x—v x~v
CM c*P c*P
(^o ro eft° ^ cW r**
O\ ' O i CO i — 1
CM m ^ *-t •
CM ^ ON
N^ ^ N«X
X— v X~S X~N
e»P dP dP
i— 1 CM in
oN° PO CCP ^o eft0 LD
r* i o i o •
CM CM \o r^ m vo
CM m OO
C*° t
co r*"^ co CD co co
CM CO CO
^— ' s^- v_^
V < T)
a) o in a)
C l-i 0) C
T3 3 4J T-I 3
C owe o
(fl Od (fl
>-> e (x c to
•Ut-l OME OS-JO)
C (fl "i-lOO >rl 01 -r4
D4J 4-l-r-,0 4-1XIC
E -rl (0 (fl (fl 4-1 (fl
Cr-l 4->EE 4JOD.
l-i -H w 3 in E
0) S >, 0) X 0
> in xi r—i in xi o
O (fl (fl
O O o
/-^
x-x CO
dP *-s
tfft v£> d^°
co i O
O
^-*
X-s,
CO
/ X ^^
<*0 dP 6>P
r-l r-l O
V V
dP /^
E
(0 x: fl x:
l-i r-H tn a) r-i in
0) d) -rl bO O> "rl
£ S-l r-l S-J V-I r-l
4J (T3
0 rJ
^— V
dP
dr° CM
r-l <
r-l
*~-s
^^
dP
a)
IX rl
>-, 0)
4J 3
r-l W
O> 13
3 C
O
0) XI
CJ1
O U
o
c c
(fl O>
x: -o
4J T-l
4-1
o> C
S-i O
0 O
g
o
tt) C
o
4-> 0)
tn o
c
o X
•o
CD
0) r-l
3 OJ
3
a) 4-i
i_i
0> 4-1
3 o
in 01
•/ p.
C ^
(fl 4-1
4J
in
O> T-l
e x:
O 4-1
in
bO
0) C
in -H
3 l-i
(fl O
0 4J
v a) in
0)
4J a
C X
O) 4-1
O
1-4 4-1
~. XI
• (fl 0) (fl
m x* r> 4-i
E 4J s-i in
s^ 3 oi
(fl O! W
4-iS-i in
o 01 T-I
c E x: x:
O 4-1 4J
o
10 4-1 O C
^ 4-1 -H
c i
(fl 3 l-l T)
4-1 (0 O C
.H 3
0) >•> l-l O
T3 (fl CX4-I
3 E
r-l CO 0)
o in x: l-i
C 4J 4-1 0)
•H C C 3
0) O
4J O E W
Or! .id
C 0) CM C
p, r—l (fl
in 4-1
11 S o>
o o x; o
Q Ql 4-1 Z
x-^ /— N ,^v
i— 1 CM CO
9-:
-------
i-C
^~
*^
3
H
I
U)
4-J
to
C
4-1 s
hfl ^
!H d>
t '"i
4-J
s s
w 5H
c^j
1 \ PI
c2 c
'H
o W
•o -°
S Q)
3 y
O r*
KT\ Q)
~^ 'O
»H
•i«
S o
3 o
C4H ,,,,
-. dp
0 in
C ^
^ UN
0 0
" >l-l
QJ j_j
CX yj
•r-l
.S £
4-i
4H °
0 5H
r* Cfl
n j-
2 °
•H
sj
_j_j
0) ^
& '—|
y ^j
•^ 4-J
W W
£
?^ ! 0
4-i tfl UH
XI "O SH
MH 4-1 0) Z>
O 5H CO
co a)
4-1 E > T)
C cu o a)
a> 4-> o >
O 01 cfl
5n >-, a) PM
a> w JH
P-l Cfl Cfl
a)
tfl 4J a) rH tfl
4-1 Cfl r-H fx, a> 4J
,£ r-l rH 4-1
144 4-> Cd rH a) M
O 4-1 Cfl rH a>
01 CO -r-l CL, 4J
4-1 E C 4J B cfl
C Q> -H IH O |3
d) 4J Cfl O
o 01 a) ex ;?
S-i >, M s-i o
CD M Cfl O rH
CM v— ' a)
C o
Cfl 4-1 -H
4-> Cfl "d C
Xi 0 0
MH 4-1 x: -ri
O 4J 4-1
W cfl O
4-i E o a)
C a> 4J
a> 4-J a> o
o en > M
rl >-, Cfl CO,
a> 01 xl
^ cd
Xi w o
UH 4-1 CO O
0
CO 0) 4-1
4J E VH 0
C a) rt pi
0) 4-> Xi
O CO
5n >~, a) a)
a) co 5n •
P-t tfl >rt
AS
C CO
cfl 4.1 in
4J Cfl tfl
XI r-H
C|H 4-1 00
O 5H
01 a)
•w E Xi
C a) T-I
0) 4-1 4-1
O CO
SH >-, a>
a) co SH
PM cfl
r^l M
a) o
5H bO
3 a>
CO 5H
^-s y^N X—x
dP dp dP
rH O in
dip ON d<9 OO dtP OO
co i I-H i m i
r- m r^ co vo <)•
in vo r*^ dP oo
00 1— 1 OO rH OO VO
c CO I
-* V oo
^^ ^x ^^s
X— s X—v X— V
dP dp dP
dP ON dP m dP O
VO CM CSl i— 1 O CM
i-H i i-H i i-H i
CM OO rH
••^X V ' V X
c
I-H T-I O
cfl tfl -H
SH 4-1 4H
<)• 4-J in C vo T-I
C 3 0
a) O cfl
U g PM
x— s
dP
O
dP oo
CO 1
r^ vo
vO
v-^1
^^
dP
m
dP CM
i-H i
csi r~-
r-l
\ '
x*-s
dP
dP vO
in i
CO
, — ^
dP
CSl rH
r-l 1
r-^
N— X
x— s
dP
dP m
rH rH
rH i
r~.
V '
rH
Cfl
4_)
O
H
co
^
C
tfl
4-1
Q
)_l
cfl
4H
0)
t3
3
1—^
O
C
•rH
4-1
0
c
01
a>
0
Q
rH
9-24
-------
i-C
4_>
•rH
£
H
CO
E
CD
4J
w
en
C
cfl ^
•^ eo
-. 0)
1> eo
bQ n\
*r* cu
rj
4-1
C
•^ CD
eo ^_j
r-l *
CD ^
3 d
4-1 .;J
£ w
S-2
O g
S o
^Q
;2
C CD
8 c
^
Q-" t4~|
r* ci
H °
p O
*0 ^
c ^
® en
^ o
5-i .i
/i\
CD ^j
ft en
•r-l
rS ^
4J
ILJ °
n ^
j^-t
Ccfl
^
•H
^ cfl
2: "-1
£ ""*
w w
. C
r*-! .,_)
r^
T3
m 0)
CD jj
4-1 Q
2
C
CD
O
r-l
CD
PM
^
cfl
4-1
4-1
O
4J
c
CD
O
CD
CM
V^J
C
Cfl 4-1
4-1 cfl
f-|
4-1 4-1
O
4-1 £
C a)
CD 4-1
O en
5-1 r-^l
0) CO
PM
A:
a -o
cfl 4-1 CD
4-1 cfl 4-1
,£ O
4-1 4-1 01
0 C
en C
4J E 0
C CU 0
CD 4->
O CO CD
5-1 >~, 5-1
CD CO cfl
PM
C
cfl
4J
4-4
O
4-1
C
CD
O
5-1
CD
4-1
cfl
f-|
4J
01
a
0)
4J
CO
fXj
eo
4J
cfl
4J
co
S
0)
4-1
w
CO
•o
CD
CD
4J
CD
e
CD
£>
cO
f~j
0)
J_l
*J
CO
co
Q)
ft
cfl
o
4-1
4J
rC
4J
CO
e
CD
4J
en
en
•d
C^
cfl
r~]
(
rrj
c
CM
CD
M
CD
1
CH
r-l
0)
en
CD
5-1
CD
*
bO
C
CO
c
0)
ft
en
•r-4
•d
5^"
CD
>
•H
r— 1
CD
•d
bO
C
ft
^
cx
1
•l— 1
q
cO
E
cfl
C
•H
0)
^1
cfl
r^-l
CD
>
rj
w
-d
CD
i— 1
,__!
cO
4-1
CO
C
•r-l
e
0)
•d
CD
r— 1
! — |
CO
4J
en
C
'
en
S
CD
4-1
CO
r^l
CO
eo
e
CD
4J
en
£x,
en
E
CD
4J
en
^
en
•d
CD
•d
rH
O
4-1
c
O
'bO
CD
5-1
s~^
^ &> s-^
dP r-l dp
CAP r^ cAP t — 1 CAP
CM r-l r-l
^-* **^ **-s
x-x
s-^ dP /~.
dP CO dP
eW r~- eAP CO cAP OO
CO CM in i OO CM
r-l i CM r- r-l i
O i— I oo
^s ^s ^^
f^
<& s-^ ^~.
O dP dP
O CO O '
C^ OO 00
^~\ *^*\ x-s
dP dP dp
CO CM O
VO 1 CT* I r-l i
CM O CM in co csi
r-l CM CM
V S V_^ ^^S
dP dP dP
r-l 00 l~-
cAP cAP CM
f-- r-l CO r-l
V V
V *~s ^~^
• — ^
x-x
dP x-v s- ,
r~- dP dP
CAP VO cAP OO cAP CM
-* i VO r-l CM oo
OO
x-v
dP
^J-
CAP CO
OO i
CM CM
CM
v^x
dp
f^
CAP CM
CO i
CM OO
i— 1
N '
1— 1
cfl
4J
0
H
.
en
e
cfl
4->
E
5-i
cfl
4-1
CD
*d
,__!
CJ
•H
4J
O
C
CO
CD
O
Q
r— 1
9-25
-------
H
g
CU
4J
Ul
CO
C
cfl
4-1 ^
01
cu cu
bO CO
ij i^*<
O 4-1
4-> C
Cfl CU
r-l Cfl
CU CX
3
•fH
ij
o w
4J "d
0 g
e s
O
13 XI
g 8
S-i e*
bO CU
S-4 *d
(U %r-l
"o 'y
c c
3 0
CJ
4-1
o <#>
in
M
CU
O 01
S-i 0
cu -i-i
CX 4-1
CO
0) •rl
f- 1 J^
4J CU
4-1
f| | Q
o cd
i-i
tfl rH
•H "H
CU 4J
U 01 CU
S-I r^N S-I
cu o) cfl
CM
,i(5
cd
4-1
4H
O
4J
c
p
cfl
4-1
4H
0
4-1
C
0
}_j
CU
0>
>-> o
4-1 X) cfl
cd 4H
XJ "d S-i
4J CU 3
S-I CO
01 cu
e > -d
cu o cu
4J 0 >
01 Cfl
K*S, cu ex
01 s-i
cfl cfl
cu
x-s rH
K*"N XI
r-H Cfl
>, CO 4-1
i-H 4-1
i-H CU SH
Cfl r-l CU
4J £ cfl
S-i O ^
cfl O
ex s
S-I O
O rH
's-/ CD
XI
4J 0
Cfl -iH C
x: T) o
4J O -iH
p« 1 \
01 4J CJ
g cfl CU
CU CJ 4J
4J O
01 CU S-I
>•, > ex
01 Cfl
rfl
y-X
•d
rH CU
4-1 CU 4-1
Cfl CU Cfl
Xi 4J 0
4J 01 O
01 CU 4J
S SH 0
o> n) C
4J X)
01
f*. cu a)
01 s-i
Cfl -H
^
^
fi 01
Cfl 4-1 0}
4J Cfl Cfl
Xi rH
4H 4J bO
0 S-i
CO CU
4J e x)
c; cu -H
CU 4J 4H
CJ 01
CU co SH
CM cfl
4J
C
4-4 S
o .2
Type ,
establisl
x~\ s~*\ s~*»
dP dP dp
OO ON -^
dP in d^ ON dP ON
vo i in i -3-1
\^s
X-s x-x x-s
dP dp dp
^" ON
dP cO d*° CO dP CM
f» ' •* ' rH I
CM ON CM -it CM CM
i-H i-l rH
^*- ' °*-s s^x
X-s
dp x-x X-N
rH dP dp
dP r-l dP ON d* CO
r~. i in i in i
CO CM i-H
^-S S»X N_x
x-s x—s x-s
dp dP dP
dP CM dP CM dP in
in CM p* i-H O rH
rH 1 1 r-l 1
CO CO VO
S— ' N« ' S^ /
X^s
x-s dP x-s
dP ~* dp
dfl O dP CO 1 01 •!-) 01 X!
S-I C Cfl C 4J
cd o S o o
4-1 4-> «rl S 01 -iH 01
C-r-l 4J>-,3cU 4J>-,CU
cu i— i cfljacu-iH cflja'r-i
g .iH 4J i-H C 4J C
s-iE oi'docfl w-dcd
cu cu M ex cu ex
>*d w p u E tops
OP cd?0)O cfl3O
Cjcd OOCXO OOO
s~>*.
dp
p«x
dP r*^-
oo '
^l ^^
VO
N^*
X-X
dP
VO
CftB *^t*
CO i
cn o
Csl
S^r'
X**i
dP
dP 1^
CO
N^X
y-s^
dP
0\ r^
cfl
1 4-1
rH CO
CU 0) CO
Other fu
related
lishment
X-s
dP
r-l
dP fs.
rH 1
VO I-H
m
s«x
X-N
dp
dP O
O CM
rH
V
X-s
dP
dP m
CM
\/
x-s.
dp
dP 00
PS. fO
rH
\/^
x"s
dp
dP OO
O i-H
rH I
CM
S^X
1
i-H I
CU X)
3 cfl
4H 4-1
I CO
C in
rH r-H
rH I
r-
^— ^
i-H
cd
4-1
O
H
CO
r^
C
cfl
4-1
e
S-i
cfl
4H
CU
•d
!— 1
o
•iH
4J
O
01
cu
o
Q
1
i— i
9-26
-------
-H
CO
e
cu
4J
W
w
C
4J
cu
W)
cfl
I-l
O
4J
W
i— 1
cu
3
4H
I-l
O
4J
O
s
•o
a
3
o
IH
b£
IH
cu
•0
C
3
4-1
0
4J
C
cu
CJ
^
CU
(X
CU
4J
4H
0
cu
Qj
r*>
4J
^J
C
CO
•H
rH
rO
Cfl
CO
CU
>s
CO
cu
4->
Cfl
a
•H
4-1
co
W
•
in
rH
ON
CU
rH
"rt
H
CO
CU
CO
CU
4->
c
CU
cfl
(X
C
CO
*d
C
o
,£
CU
CJ
C^
CU
•d
•H
4H
B
0
CJ
dP
m
ON
sw'
Cfl
Q)
•rl
j_>
CO
•iH
IH
CU
4->
CJ
cfl
cfl
JM
u
fj
o
4J
cfl
rH
rH
Cfl
4->
CO
C
•H
•d
CU
4-1
u
01
i-H
Cfl
•g
•iH
|j
x-s
•d
4«J
c
o
u
B
cfl
4->
4H
O
4->
C
cu
CJ
CU
CM
^
C
cfl
4J
4H
0
4J
C
cu
CJ
IH
cu
CM
*M
B
cfl 4J
4J Cfl
^fj
f| | |j
O
CO
4J gj
rj CU
CU 4J
CJ CO
IH >,
CU Cfl
CM
^
C -d
Cfl 4-> CU
4-) Cfl 4->
*C O
4H 4-1 0)
0 C
CO B
4-> a o
C CU U
CU 4-*
CJ W CU
CU CQ Cfl
CM
•a
cfl
4J
<4H
O
4->
CU
CJ
CO
a
CU
4J
CO
>1
CO
4J
cfl
f-j
4J
CO
a
cu
4->
CO
^s,
CO
•d
CU
rl
CU
4J
CU
CU
>
cfl
rC
Cfl
£
CO
a
cu
4J
CO
>1
Cfl
'O *0
B cu
Cfl H
i 'cfl
•d 4J
B co
CM B
CU
V-f C
0) 0)
S ,Cj
&
1
4H -d
rH CU
0) rH
CO rH
Cfl
CU 4J
IH co
cu C
^ *fH
bO
B co
•H a
CO CU
C 4J
CU Cfl
(X >-i
CO Cfl
•H
-d
^Sj
^4
d)
>
rH CO
o> a
•d cu
4J
bO Cfl
H ^>
•rH W
CL>
(X
1
•H a
C cu
cfl 4J
a co
Cfl CQ
C -d
•H CU
•d
CU i-H
rl O
Cfl 4H
4-1
CU
o » (fp *^
s* I 0 rH ON rH
•st ON I— 1 1 1
CM r-l -tf
X-N S*\ X-S
dP dP dP
00 0
dP ON dP O dP O
1^ i O rH CO rH
OO ro O i ON i
00 rH ON vo
s-' ON ON
s--' s^x1
X"V X*^
X-s dP dP
dp O ON
dP VO dP PS. dP CO
ON rH 00 i CM i
i m r^ co vo
ro -st CM
s»x s^x s^x"
dp dP dP
CM rH rH
dP CM dP ro dP ro
*st 1 *sT 1 ^ 1
rH r^ CM vo CM r~
s-' rH i-H
<^ S_X
IH M
O CU
^> W *r-) Cfl J^
cfl O 8 O O
4J 4-1 -rH a W -iH CO
C-rl 4J>,3£U 4->>-, -o w c 4J a caca
OB cfl^cuo cfl 3t O
cj cfl cj o ex o o o o
dP
CM
dP
rH
dP
oo
dP
CM
CM
dP
O
CO
,
rH
0)
4H
IH
0)
O
X-s
dP
^
i
rH
s«^
X— s
dP
00
CM
i
VO
X-s
dp
oo
00
O
00
x-s
dP
rH
CO
l
VO
i-H
s^*
dp
ON
CO
i
rH
CM
l
r\
cfl
4->
Cfl
CU Cfl
4J
•d c
cu cu
4-> I
Cfl rC
rH Cfl
CU -rl
rl rH
dP
O
rH
dP
O
CM
dP
O
00
dP
s*
rH
dP
CM
i
rH
CU
4H
1
c
g
fl)
bO
3
x-s
dP
m
rH
i
^~
S^X
s~^
dP
i-H
ro
1
00
X-N
dP
ON
00
i
i-H
>»x
^r-S
dP
i
CO
^^
dP
m
CO
i
CO
i
fN
4J
co
cu ca
4J
•d C
0) 0)
4J §
cfl J
i-H Cfl
« •*
IH rH
x— \
dP
dP vo
CM
S^ X
x-s
dP
VO
dP CM
O i
CM -it
i-H
x-s.
dP
ro
dP ON
rH i
ON 00
oo
x-s
dp
eo i
CM CM
CM
x-s
dP
dP CM
eo i
CM 00
i-H
i-H
cfl
4J
0
H
CA
•a
cfl
fi
cfl
<4H
-8
3
rH
g
•rl
4J
0
B
to
cu
O
o
rH
9-21
-------
of the survey regions, ranging from 12 percent in the Pacific
region to 20 percent in the Mountain and Central regions. The
distributions of tanks at gas stations in the two ownership
groups show differences particularly in the Pacific survey
region. While 18 percent of the tanks in all survey regions are
found at gas stations owned by major petroleum companies, in the
Pacific survey region, 29 percent of the tanks are at this type
of establishment. The Mountain survey region has the lowest
proportion of tanks in this group (11%). And while 32 percent of
the tanks in all survey regions are found at gas stations owned
by other companies, in the Pacific survey region only 16 percent
of the tanks are at gas stations owned by other companies. Tanks
at large non-fuel-related establishments account for 33 percent
of the tanks in the Pacific survey region, and 14 percent of the
tanks in the Central survey region, while they account for 22
percent of the tanks across all survey regions.
B. Selected Tank Characteristics. Analyzed by Region and
by Establishment Type
Selected tank characteristics, including tank age, tank
capacity, type of motor fuel held in the tank, and installation
characteristics, were analyzed by survey region and by
establishment type. These analyses are based on a file of 2411
tanks, weighted to represent the national estimate of 638,000
tanks. The results of the analyses of these characteristics are
displayed in Table 9-8 through Table 9-15.
9-28
-------
1. Tank Age
Tank age was calculated by subtracting the year of
installation from the year of interview (1985). If a tank was
used (i.e., second-hand) when it was installed, the age of the
tank at installation was added to the number of years since
installation, to obtain tank age. All information required to
compute tank age was present for 69 percent of the tanks
surveyed. Tables 9-8 and 9-9 are based on these tanks. They
show the mean (or average) age of tanks and the median (or 50th
percentile) age of tanks in years, by survey region and
establishment type. For tanks in all survey regions, the average
age is 12 years, and the median age is 11 years. The average age
within survey regions was also 12 years for all regions except
Mountain, where the average age was 11 years. The median tank
age also was similar for all survey regions.
More differences appear when the average age of tanks is
compared across types of establishments. Table 9-9 shows that
tanks at gasoline stations owned by other companies (i.e., those
not owned by major petroleum companies) tend to be older than
tanks at other types of establishments. The mean age of tanks at
gas stations owned by other companies is 14 years (compared to a
mean of 12 years, overall) and the median age of these tanks is
13 years (compared to a median of 11 years overall). Large non-
fuel-related establishments appear to have newer tanks than other
establishments, with an average tank age of 10 years, and a
median tank age of 7 years.
9-29
-------
2. Tank Capacities
Average and median tank capacities also varied within survey
regions and establishment types, as may be seen in Tables 9-10
and 9-11. (Nearly all tanks, over 99%, had known capacity.)
Across all survey regions and establishment types, the average
tank size is 5,405 gallons, and the median tank size is 4,000
gallons. Tanks in the Pacific survey region tend to be larger,
with an average tank size of 6180 gallons and a median tank size
of 6000 gallons. Tanks in the Northeast survey region tend on
average to be smaller, with a mean tank capacity of 4,583
gallons, and a median capacity of 4000 gallons.
Tanks at gas stations owned by major petroleum companies
tend to be larger, with an average capacity of 6,821 gallons
(compared to the national average of 5,405 gallons), and a median
capacity of 6,000 gallons (compared to the national median of
4,000 gallons.) The establishments with the smallest tank
capacity are the government and military establishments, where
the average tank size is 4,342 gallons and the median tank size
is 2,000 gallons.
3. Type of Motor Fuel Stored
Tables 9-12 and 9-13 show the estimates, by survey region
and by type of establishment, of the percent of underground tanks
that were used to store specific motor fuel types during the
prior year. All tanks had data on fuel type. There is not much
variation in the percent of tanks that store leaded gasoline
across survey regions. For all survey regions together, 33
percent of the tanks store leaded gasoline, and the proportions
within survey regions range from a low of 28 percent in the
9-30
-------
Northeast region to a high of 37 percent in the Pacific region.
Unleaded gasoline is stored in 42 percent of the tanks overall.
The regions with the lowest percentage of tanks that store
unleaded gasoline are the Midwest and Mountain, each with 36
percent. The region with the highest percentage of tanks that
store unleaded gasoline is the Northeast with 48 percent. Diesel
fuel is stored in 21 percent of the tanks nationally, but in a
slightly lower percentage of tanks in the Pacific survey region
(16%).
Tanks that store aviation fuel account for about two percent
of the tanks in the continental United States, while tanks that
store gasohol and tanks that store other fuel types and fuel
blends each account for about one percent of the tanks in the
country. Regional variations among these products do not appear
to be meaningful.
Some variation in the types of fuels stored within
establishment types is expected, since different types of
establishments use fuels for different types of vehicles. A
smaller percentage of government and military tanks are used to
store unleaded gasoline (27%) than the national percentage of 42
percent. Among gas stations, the percent of tanks that store
unleaded gasoline is substantially higher that the national
percentage, but interestingly enough, the percentage is higher
for gas stations owned by major petroleum companies (at 60%) than
it is for gas stations owned by other companies (at 51%).
Although it is risky to draw any conclusions from such small
subsamples, it is not surprising that government and military
establishments and other fuel-related establishments account for
the majority of tanks that store aviation fuel. (These two
categories of establishments include the government and public
sector airfields, airports and air services.)
9-31
-------
4. Material of Construction
Overall, 11 percent of the underground motor fuel storage
tanks in the continental United States are fiberglass. This
information was known for 92 percent of tanks, and the estimate
of 11 percent fiberglass is based on those with known material of
construction. The percent distribution is nearly equal for all
survey regions (ranging between 10% and 12% in each region)
except for the Central survey region, where the percentage of
fiberglass tanks is just slightly higher (16%). There are more
striking differences in the percentages of tanks that are
fiberglass within establishment types. The percentages of
fiberglass tanks at gas stations that are owned by other
companies, other fuel-related establishments, and large non-fuel-
related establishments are all close to the national percentage
of 11 percent. (These establishment types have 9%, 8%, and 10%
respectively.) Only 6 percent of the tanks at government and
military establishments are fiberglass, while 24 percent of the
tanks at gas stations owned by major petroleum companies are
fiberglass.
Steel tanks comprise 89 percent of the underground motor
fuel storage tanks nationally. According to establishment
operators, the majority (about 85%) of steel tanks are coated,
based on tanks for which this question was answered. (However,
establishment operators did not know if the tanks were bare or
coated for 39% of the tanks, and the material of construction was
unknown for 8% of the tanks.) Among tanks reported to be coated,
the coating material was not reported for 7 percent of the tanks.
Among tanks for which the coating material was reported, 58
percent were coated with asphaltic material, 34 percent with coal
tar epoxy, and the remainder with urethane (2%), "black coating"
(2%), fiberglass/epoxy (1%) or other coatings with less than 1
9-32
-------
percent each (and totaling 2% of responses) such as double-
wrapped tar tape, red oxide, paint, jenite (creosote and
asphalt), tar paint, rust primer and anti-rust paint, etc.
Based on available data, it is estimated that 12 percent of
the tanks nationally are bare (uncoated) steel tanks. These bare
tanks are found in differing proportions throughout the country.
Only five percent of the tanks are reported to be bare steel in
the Northeast survey region. However, 38 percent of the tanks in
the Pacific survey region are reported to be bare steel. The
remaining four survey regions are much closer to or equal to the
national proportion.
The variations in the distributions of tanks that are bare
steel across establishment types are less pronounced. The lowest
percentage is found among gas stations owned by major petroleum
companies, where 7 percent of the tanks are reported to be bare
steel. Large non-fuel-related establishments have the highest
percentage of bare steel tanks, with 17 percent. However, the
percentages for all establishment types are within five
percentage points of the overall percentage.
5. Cathodic Protection
Only five percent of all tanks were reported to have
cathodic protection systems of any type installed to protect them
against corrosion, based on the 87 pecent of tanks for which this
question was answered. Within survey regions, 17 percent of the
tanks in the Mountain survey region were reported to have
cathodic protection. There are no other particularly noteworthy
divergences from the national proportions across other survey
regions or across establishment types.
9-33
-------
6. Water Table Level
Establishment operators were asked to indicate how each of
the tanks at an establishment were situated in relation to the
water table ("completely above", "partially above and partially
below", or "completely below"). Establishment operators were
able to provide information on the water table level for about 70
percent of the tanks surveyed. Based on the tanks for which
responses were given, 21 percent of the tanks are reported to be
installed partially or completely below the water table. This
percentage varies somewhat across regions, with only 12 percent
of the tanks in the Central survey region installed in or beneath
the water table. In the Mountain survey region, the highest
percentage of tanks (27%) are reported to be installed in or
beneath the water table.
Among establishment types, large non-fuel-related
establishments have the lowest percentage of tanks reported to
be installed in or beneath the water table, with 10 percent. The
other fuel-related establishments reported 33 percent of tanks
installed partially or completely below the water table.
Government and military establishments also have a slightly
higher percentage, with 27 percent of tanks at these
establishments reported to be installed in or beneath the water
table.
7. Surface over the Tank
About 73 percent of the tanks nationally are covered by a
paved surface. (Paved surfaces are defined here to include
9-34
-------
asphalt surfaces, concrete surfaces and surfaces that are made of
these materials in combination. They exclude surfaces that are
"paved" with gravel or crushed rock. Surface over the tank was
known for nearly all tanks — over 99 percent.) Among the survey
regions, the Pacific region has the highest percentage of tanks
that are covered by a paved surface, with 90 percent of tanks
reported to have this trait. The Midwest survey region has 65
percent of its tanks covered by a paved surface, and this is the
lowest proportion among regions. The remaining four regions are
all within 2 percentage points of the national percent.
Among establishment types, gas stations have the highest
percentage of tanks covered by paved surfaces, with 95 percent of
the tanks at gas stations owned by major petroleum companies
covered by paved surfaces, and 84 percent of the tanks at gas
stations owned by other companies covered by paved surfaces.
Government and military establishments have the lowest proportion
of tanks under pavement, with 46 percent. Tanks that are under
unpaved surfaces may or may not have the same amount or type of
traffic over the tank as tanks under paved surfaces. However,
tanks under paved surfaces will be more costly to excavate and
rebury, should there be a need to remove, replace or repair them.
8. Manifolded Systems
Twenty three percent of all tanks are in manifolded tank
systems. (This was known for 99.9% of tanks surveyed.)
Manifolded systems were defined, for the purposes of this survey,
as two or more underground motor fuel storage tanks that are
joined together by pipes or lines prior to the dispenser meters.
The average number of tanks per manifolded system is 2.36. The
survey region with the highest percentage of tanks in manifolded
9-35
-------
systems is the Northeast, with 29 percent of tanks in such
systems. The Mountain survey region has the lowest percentage of
tanks in manifolded systems, with 12 percent. The remaining four
regions are all slightly lower than, but very close to the
national percentage, ranging from 18 percent to 22 percent.
Other fuel-related establishments have a somewhat higher
percentage of tanks in manifolded systems (30%). The
establishment type with the lowest percentage of tanks in
manifolded systems was the government and military establishments
with 14 percent.
9. Types of Pumping Systems
The majority of tanks are reported to be connected to
suction pumping systems, based on the 95 percent of tanks for
which type of pumping system was known. Overall, 28 percent of
the tanks were reported to be connected to pressurized delivery
systems. All survey regions were close to the national
percentage on this trait, except the Mountain survey region,
where 42 percent of the tanks were reported to be connected to
pressurized delivery systems. Also, more than half (58%) of the
pumping systems at gas stations owned by major petroleum
companies are pressurized, which is nearly twice the national
percentage. Government and military establishments and large
non-fuel-related establishments have much lower percentages of
tanks connected to pressure pumping systems, with 9 percent and
13 percent respectively.
9-36
-------
10. Metered Dispensing Systems
It is virtually impossible to conduct inventory
reconciliation monitoring on tanks that are connected with
dispensing systems that do not have meters to record the total
amount of fuel dispensed from the tank. Fortunately, 91 percent
of the tanks nation-wide have metered dispensing systems, based
on 99.8 percent of tanks for which this question was answered.
The percentages within each survey region are all within five
percentage points of the national percent of tanks with dispenser
meters. Some variation in the percentage of tanks with dispenser
meters can be seen across establishment types. Nearly all tanks
at gas stations (100 percent at gas stations owned by major
petroleum companies and 98 percent at gas stations owned by other
companies) have metered dispensing systems. The remaining three
establishment types are all slightly lower than the national
percentage of tanks with metered dispensing systems. Large non-
fuel-related establishments have the lowest percentage, with 80
percent of the tanks at this type of establishment connected to
metered dispensing systems.
11. Self-Installed Tanks
The identity of the tank installer was known for over half
(54%) of tanks surveyed. Based on these tanks with known
installers, establishment owners reported that 20 percent of
underground motor fuel storage tanks were self-installed (i.e.,
installed by the establishment or its owners.) However, this
percentage varies both across survey region and across
establishment type. Percentages of tanks that were self-
installed across survey region range from a high of 44 percent in
the Central survey region to a low of 6 percent in the Mountain
9-37
-------
survey region. Among establishment types, 44 percent of the
government and military tanks were self-installed, which was the
highest percentage among the establishment types. The
establishment types with the lowest percent of self-installed
tanks are the gas stations owned by other companies (with 9%
self-installed) and gas stations owned by major petroleum
companies (with 10% self-installed).
12. Secondhand Tanks
A small percentage of the tanks surveyed were secondhand
when they were installed at their current location, based on the
81 percent of tanks for which this information was ascertained.
It is estimated that four percent of the tanks nationally were
installed secondhand. (As explained above, if a tank was
secondhand when installed, its total age was calculated to
include its age at the time it was installed.) Because of the
small percentage of tanks with this trait, it is risky to draw
many conclusions about the percent distribution of this trait
within survey regions and establishment types. It may be noted
with some caution that a higher percentage of tanks at large non-
fuel-related establishments (10%) were secondhand tanks, while a
lower percent of tanks at gas stations in both ownership
categories and tanks in the Pacific survey region (about 1% for
each category) were secondhand.
IV. LEAK STATUS OF UNDERGROUND MOTOR FUEL STORAGE TANKS
A total of 439 individual and manifolded tank systems at
business and government establishments were tested as a part of
the physical tank testing phase of the survey. The analyses in
this subsection are based on two related data files. The first
9-38
-------
is the file of 412 individual and manifolded tank systems for
which there are conclusive test results. The second data file
consists of the 383 conclusive test results which refer to
single-tank systems. The sample of individual and manifolded
tank systems, is analyzed in tables where the characteristics
being examined apply to entire systems, whether they consist of
individual tanks and their lines, or manifolded tanks and their
lines. The sample of individual tank systems, is analyzed in
tables where the characteristics being examined apply to
individual tanks in systems rather than manifolded tanks in
systems (such as, for example, tank capacity.)
A. Leak Status of Tank Systems within Survey Regions and
Establishment Types
Detailed analyses are given below for the leak status
distribution of tank systems. On an establishment basis, an
estimated 117,000, or 36 percent, of establishments have one or
more tank systems judged to be leaking under test conditions.
This is comparable to the percentage on a tank system basis.
Within the continental United States 35 percent of the
underground motor fuel storage tank systems at business and
government establishments are judged to be leaking under test
conditions, based on all tests with conclusive results. The
distributions of percentages of leaking tank systems across
survey regions and establishment types are displayed in Tables
9-16 through 9-18. The estimated number of tank systems judged
to be leaking is 189,000. This excludes tank systems for which
test results were inconclusive. (Therefore, this estimated
number, when divided by the estimated total number of tanks, will
not give the percentage shown.)
9-39
-------
Table 9-16. Estimated number and percent of tank systems 1/2
judged to be leaking under test conditions within
each survey region (95% confidence bounds in
parentheses)
Survey Region
1
Northeast
2
Southeast
3
Midwest
4
Central
5
Mountain
6
Pacific
Total
Sample
size3
98
104
82
48
39
41
412
Number of tank
systems judged
to be leaking
(in 1,000's)
57
(37-77)
42
(19-65)
34
(22-46)
26
(14-38)
11
(3-18)
20
(11-30)
189
(153-226)
Percent of tank
systems judged
to be leaking
39%
(32-45%)
36%
(22-50%)
28%
(19-37%)
46%
(22-70%)
33%
(17-48%)
28%
(18-38%)
35%
(30-40%)
1In this table tank system test results are reported for single
tank systems unless multiple tanks were tested as a part of a
manifolded tank system that was not broken apart. These
manifolded tank systems are included in this table.
2Does not include farm tanks.
3Includes all tank systems with conclusive test results.
Excludes tank systems for which test results were inconclusive,
-------
Table 9-17.
Estimated number and percent of tank systems1,
judged to be leaking under test conditions within
establishment types (95% confidence bounds in
parentheses)
Type of
Establishment
Sample
size3
Number of tank
systems judged
to be leaking
(in 1,000's)
Percent of tank
systems judged
to be leaking
Government
and military
Gas stations
owned by major
petroleum
companies
Gas stations
owned by other
companies
Other fuel-
related estab-
lishments
Large non-fuel-
related estab-
lishments
55
62
155
64
76
29
(5-54)
25
(11-38)
56
(40-71)
35
(25-45)
45
(17-71)
36%
(16-55%)
32%
(19-45%)
30%
(22-37%)
57%
(43-71%)
33%
(18-47%)
Total
412
189
(153-226)
35%
(30-40%)
1In this table tank system test results are reported for single
tank systems unless multiple tanks were tested as a part of a
manifolded tank system that was not broken apart. These
manifolded tank systems are included in this table.
2Does not include farm tanks.
3Includes all tanks with conclusive test results.
for which test results were inconclusive.
Excludes tanks
9-41
-------
Table 9-18. Estimates by survey region and establishment type of percent of
underground motor fuel storage tank systems -1-'2'3 judged to be
leaking under test conditions (95% confidence bounds in
parentheses)
Establishment type
Survey
region
1
Northeast
2
Southeast
3
Midwest
4
Central
5
Mountain
6
Pacific
All
regions
Govern-
ment and
military
60%
(40-80%)
17%
(0-34%)
0%4
29%
(0-76%)
60%
(36-84%)
20%
(0-61%)
36%
(16-55%)
Gas
stations
owned by
major
petroleum
companies
50%
(50-50%)
55%
(43-66%)
12%
(0-39%)
40%
(27-53%)
0%4
29%
(18-39%)
32%
(19-45%)
Gas
stations
owned by
other
companies
31%
(19-43%)
29%
(20-37%)
31%
(12-51%)
41%
(12-70%)
18%
(0-52%)
0%4
30%
(22-37%)
Other
fuel-
related
establish-
ments
44%
(4-85%)
60%
(37-83%)
68%
(49-88%)
60%
(40-80%)
38%
(33-42%)
50%
(0-100%)
57%
(43-71%)
Large
non-fuel
related
establish-
ments
32%
(12-52%)
36%
(0-73%)
0%4
75%
(38-100%)
50%
(12-88%)
30%
(0-75%)
33%
(18-47%)
All estab-
lishment
types
39%
(32-45%)
36%
(22-50%)
28%
(19-37%)
46%
(22-70%)
33%
(17-48%)
28%
(18-38%)
35%
(30-40%)
In this table tank system test results are reported for single tank systems
unless multiple tanks were tested as part of a manifolded tank system that
could not be broken apart. These manifolded tank systems tested together are
included in this table.
Does not include farm tanks.
Includes all tanks with conclusive test results. Excludes tanks for which
test results were inconclusive (sample size - 412 cases).
No eligible sampled establishments of this type were found in this survey
region, so the percent and confidence bounds on it were not estimated.
9-42
-------
As is shown in Table 9-16, the Central survey region has the
highest percentage of leaking tank systems, with 46 percent. The
regions with the lowest percentage of leaking tank systems are
the Midwest and the Pacific, each with 28 percent. The Northeast
survey region has the highest estimated number of leaking tank
systems (57,000) while the Mountain survey region has the lowest
estimated number (11,000).
Table 9-17 shows the number and percent of tank systems
judged to be leaking by establishment type. Fuel-related
establishments other than gas stations have the highest
percentage of leaking tank systems, with 57 percent judged to be
leaking. Gas stations owned by other companies have the lowest
percentage of tank systems judged to be leaking, with 30 percent,
but account for the largest estimated number (56,000) of tank
systems judged to be leaking.
The percent of tank systems judged to be leaking under test
conditions within each category of establishment type and survey
region is displayed in Table 9-18. Small sample sizes within
many of the cells of this table result in wide confidence bounds
and some difficulty in interpretation. The five cells with the
highest percentages of tank systems judged to be leaking were
government and military tank systems in the Northeast and
Mountain survey regions, tank systems at other fuel-related
establishments in the Midwest and Central survey regions, and
large non-fuel-related establishments in the Central survey
region. The five cells with the lowest percentages of tank
systems judged to be leaking were government and military tank
systems, tank systems at gas stations owned by major petroleum
companies, and tank systems at large non-fuel-related
establishments in the Midwest survey region, tank systems at gas
9-43
-------
stations owned by major petroleum companies in the Mountain
survey region, and tank systems at gas stations owned by other
companies in the Pacific survey region.
B. Characteristics Associated with Leaking Tanks
Selected tank characteristics, including tank age, tank
capacity, type of motor fuel held in the tank, and installation
characteristics were analyzed by tank system leak status. The
results of these analyses, including the sample size, estimated
number of tank systems leaking, and percent of tank systems
leaking, are displayed in Tables 9-19 through 9-23.
!• Tank Age and Material
Table 9-19 shows the estimated number and percent of tank
systems that are judged to be leaking under test conditions
within five categories of tank age. While overall, 35 percent of
tank systems are judged to be leaking, 57 percent of the tank
systems that are more than twenty years old are judged to be
leaking under test conditions. However, tank systems that are
thirteen through twenty years old form the age category with the
smallest percentage leaking. Note, however, that this difference
does not appear to be statistically significant. Each of these
arbitrary age categories accounts for a similar number of leaking
tank systems, ranging from an estimated 22,000 tank systems to
27,000 tank systems.
Table 9-19A provides a more detailed look at percent judged
to be leaking broken down by both age and tank material. There
9-44
-------
Table 9-19. Estimated number and percent of tank systems1'2
judged to be leaking under test conditions within
tank age categories (95% confidence bounds in
parentheses)
Tank age
categories
< 5 years old
5-8 years old
9-12 years old
13-20 years old
> 20 years old
Age unknown
Total
Sample
size
54
52
57
56
36
128
383
Number of
tanks judged
to be leaking
(in 1,000's)
27
(4-51)
25
(13-38)
26
(9-42)
22
(8-35)
26
(14-37)
50
(21-79)
177
(139-214)
Percent of
tanks judged
to be leaking
38%
(19-57%)
35%
(23-47%)
36%
(20-52%)
30%
(15-45%)
57%
(39-75%)
29%
(19-39%)
35%
(30-40%)
In this table, tank system test results are reported for single
tank systems. Tanks tested as a part of a manifolded tank
system that was not broken apart are not included in the table.
2Does not include farm tanks.
3Tank age, as calculated from year of installation and age at
installation (if second-hand), was missing for 128, or 33%
of the tanks.
4Includes all individually tested tank systems with conclusive
results. Excludes tank systems for which test results were
inconclusive, and tank systems tested as part of a manifolded
tank system.
9-45
-------
Table 9-19A.
Estimated number and percent of tank systems1'2
judged to be leaking under test conditions, by
age and material of tank construction3
(95% confidence bounds in parentheses)
Tank material
and age
categories
Steel:
< 5 years old
5-8 years old
9-12 years old
13-20 years old
> 20 years old
Fiberglass:
< 5 years old
5-20 years old
Sample
size
46
44
46
51
35
7
20
Number of
tanks judged
to be leaking
(in 1,000's)
23
(< 47)
21
(12-30)
18
(5-31)
22
(8-35)
26
(14-39)
2
(< 5)
10
(< 20)
Percent of
tanks judged
to be leaking
38%
(11-65%)
38%
(24-45%)
31%
(13-49%)
33%
(17-50%)
58%
(39-77%)
26%
(< 77%)
36%
(17-55%)
1In this table, tank system test results are reported for single
tank systems. Tanks tested as a part of a manifolded tank
system that was not broken apart are not included in the table.
2Does not include farm tanks.
3Respondents were unable to provide age or material of
construction for 33 percent and 8 percent, respectively, of
tested tanks.
9-46
-------
Table 9-20.
Estimated number and percent of tank systems 1/
judged to be leaking under test conditions within
tank capacity category (95% confidence bounds in
parentheses)
Tank Capacity
< 1100 gallons
1101 to 3999
gallons
4000 to 4999
gallons
5000 to 5999
gallons
6000 to 7999
gallons
8000 to 9999
gallons
10,000 to 11,999
gallons
> 12,000 gallons
Total
Sample
size3
82
58
61
17
50
33
60
22
383
Number of tank
systems judged
to be leaking
(in 1,000's)
29
(13-45)
21
(9-32)
30
(11-49)
13
(3-24)
25
(4-45)
15
(2-28)
34
(17-51)
11
(3-19)
177
(139-214)
Percent of tank
systems judged
to be leaking
23%
(13-34%)
27%
(18-37%)
39%
(24-55%)
61%
(35-86%)
37%
(14-60%)
34%
(11-58%)
45%
(29-62%)
43%
(30-57%)
35%
(30-40%)
In this table tank system test results are reported for single
tank systems. Tank tested as a part of a manifolded tank
system that was not broken apart are not included in the table.
2Does not include farm tanks.
3Includes all individually tested tank systems with conclusive
results. Excludes tank systems for which test results were
inconclusive, and tank systems tested as part of a manifolded
tank system.
9-47
-------
Table 9-21. Estimates by fuel types stored during the prior year
of the number and percent of tank systems1'2 judged
to be leaking under test conditions (95% confidence
bounds in parentheses).
Fuel types stored Samplf
during prior year size'
Number of tank
systems judged
to be leaking
(in 1,000's)
Percent of tank
systems judged
to be leaking
Leaded gasoline
Unleaded gasoline
Diesel fuel
130
171
97
32
(15-49)
72
(53-91)
73
(48-99)
18%
(9-26%)
33%
(25-41%)
57%
(47-67%)
In this table tank system test results are reported for single
tank systems unless multiple tanks were tested as a part of a
manifolded tank system that was not broken apart. These
manifolded tank systems are not included in the table.
f\
^Does not include farm tanks.
3Includes all tank systems with conclusive test results.
Excludes tank systems for which test results were inconclusive.
9-48
-------
Table 9-22. Estimated number and percent of tank systems1'2
judged to be leaking under test conditions for
single tanks and tank systems in manifolded systems
(95% confidence bounds in parentheses).
Number of tank Percent of tank
Sample systems judged systems judged
size to be leaking to be leaking
(in 1,000's)
Single tanks 338 140 31%
systems (not (105-174) (26-36%)
manifolded)
Tanks in mani- 74 50 54%
folded systems (19-80) (39-68%)
Total 412 189 35%
(153-226) (30-40%)
1In this table tank system test results are reported for single tank
systems unless multiple tank systems were tested as a part of a
manifolded tank system that was not broken apart. These manifolded
tank systems are included in this table.
2Does not include farm tanks.
3N1 = Tank systems with conclusive test results. Excludes tank
systems for which test results were inconclusive.
9-49
-------
Table 9-23.
Estimates number and percent of tank systems1 judged
to be leaking under test conditions for tank systems
with selected installation characteristics (95%
confidence bounds in parentheses).
Installation
characteristics
Fiberglass tanks
Bare (uncoated
steel) tanks
Coated steel tanks
Tanks with cathodic
protection installed
Tanks installed par-
tially or completely
below the water table
Tanks covered by a
paved surface
Tanks connected to
a pressure pump
delivery system
Tanks with metered
dispensing systems
Tanks that were
self -installed
Tanks that were 2nd-
hand when installed
National totals3
Sample
size2
30
33
192
10
40
263
71
351
42
13
383
Percent of
systems judged
to be leaking
31%
(15-48%)
32%
(14-49%)
38%
(30-46%)
60%
(0-100%)
47%
(25-69%)
39%
(32-45%)
41%
(27-55%)
36%
(31-40%)
23%
(31-40%)
52%
(26-78%)
35%
(30-40%)
Number of systems
judged to be leak-
ing (in 1,000's)
12
(1-24)
14
(0-28)
97
(65-128)
9
(< 25)
23
(9-36)
136
(103-198)
37
(21-52)
164
(130-198)
14
(3-26)
9
(4-14)
177
(139-214)
1Does not include farm tanks.
2Includes all individually tested tank systems with conclusive
results. Excludes tank systems for which test results were
inconclusive, tanks tested as part of a manifolded tank system.
3Since the categories overlap, the national totals are given to
provide a reference point, not as the total of the rows alone.
9-50
-------
is some indication that the oldest steel tanks (> 20 years old)
may have a higher proportion leaking than younger steel tanks. A
comparison of percent judged to be leaking by tank material
cannot be made for tanks more than 20 years old, since no
fiberglass tanks that old were found in the survey. For age
categories where a comparison can be made by tank material, no
significant differences appear.
2. Tank Capacity
While the percentage of tank systems judged to be leaking
under test conditions nationally is 35 percent for all tested
tank systems, there is some variation in the percentage across
size categories. Table 9-20 displays the estimated number and
percent of tank systems judged to be leaking under test
conditions within eight arbitrary size categories. The
percentage of tank systems judged leaking is somewhat lower than
the national percentage in the two smallest size categories. An
estimated 23 percent of the tank systems that are 1100 gallons or
less are judged to be leaking, and 27 percent of the tank systems
ranging in size from 1101 to 3999 gallons are judged to be
leaking. The size categories with the largest percentages of
tank systems that are judged to be leaking are systems with tanks
that are 5000 to 5999 gallons, with 61 percent; systems with
tanks that are 10,000 to 11,999 gallons, with 45 percent; and
systems with tanks that are 12,000 or more gallons, with 44
percent. The size categories with the largest estimated numbers
of leaking tank systems are those, systems with tanks that are
10,000 to 11,999 gallons, and 4,000 to 4,999 gallons.
9-51
-------
3• Type of Motor Fuel Stored
There were interesting differences in the leak status of
tank systems depending upon the type of motor fuel they had
stored during the prior year, as may be seen in Table 9-21.
Among tank systems that store unleaded gasoline the percent
leaking under test conditions is 33 percent which is nearly the
same as the national percentage. However, among tank systems
used to store leaded gasoline, only 18 percent, or about half of
the national percentage, are judged to be leaking. Among tank
systems storing diesel fuel, the percent judged to be leaking is
57 percent, or nearly two-thirds more than the national
percentage. Very small numbers of tank systems used to store
other fuel types such as aviation fuel and gasohol were tested,
so reliable estimates of leak status can not be provided for
tanks storing these fuel types.
4. Manifolded Systems
A somewhat higher proportion of the tank systems that are
part of a manifolded tank system are judged to be leaking under
test conditions, as is displayed in Table 9-22. Among tank
systems that are a manifolded system of two or more tanks, or are
individually tested members of such a system, 54 percent of the
tank systems are leaking, compared with single tank systems,
where 31 percent of the tank systems are leaking.
5. Material of Construction
Single-tank systems with tanks made of fiberglass, coated
steel and bare (uncoated) steel were analyzed separately, with
9-52
-------
the results displayed in Table 9-23. The percent of tank systems
judged to be leaking under test conditions within each category
of material of construction was nearly the same as the overall
percent: compared to 35 percent of tank systems judged to be
leaking overall, 31 percent of the systems with fiberglass tanks
are leaking, 32 percent of the systems with bare steel tanks are
leaking, and 38 percent of the systems with coated steel tanks
are leaking. Systems with coated steel tanks account for the
majority of all tank systems, and therefore not surprisingly
account for the largest number of tank systems judged to be
leaking within construction material type.
6. Cathodic Protection
Only a small percentage (about 5% nationally) of tank
systems have cathodic protection, and a small number of these
systems were tested. The results of these tests for ten systems
with cathodic protection are displayed in Table 9-23 and show
that 60 percent of cathodically protected tank systems are
estimated to be leaking. However, the confidence bounds on this
estimate are not tight. In fact, the 95 percent confidence
limits range from 0-100 percent. A tighter bound can be
constructed by requiring a lower confidence: for example, the 50
percent confidence bounds are 33-87 percent.
This result should not be taken to imply that cathodic
protection, when properly installed, maintained, checked by the
local establishment operator, and covering the entire tank system
is not an effective leak prevention system in principle.
However, it does raise a question that would bear further
investigation, namely, how feasible it is to achieve the above
conditions in practice. A possible explanation for the high
proportion judged to be leaking among cathodically protected tank
9-53
-------
systems is that such protections are only judged worth the
installation expense for tank systems believed to be at high risk
of leaking. Thus, the figure would be more a description of the
tank system universe to which cathodic protection is applied than
a measure of what would happen were it to be adopted on a wider
basis.
7. Water Table Level
Systems with tanks that are installed in or beneath the
water table may be more susceptible to corrosion than tanks that
are installed above the water table. Based on an analysis of
forty tank systems with tanks reported to be installed in or
beneath the water table, it is estimated that 47 percent of the
tank systems with tanks installed in or beneath the water table
are leaking under test conditions. This is a somewhat higher
percentage than the national percentage of 35 percent of systems
leaking.
8. Surface over the Tank
Tank systems with tanks that are covered by a paved surface
might be expected to be more protected than tank systems with
tanks that are not covered by pavement. However, 37 percent of
the tank systems with tanks covered by paved surfaces were judged
to be leaking under test conditions, which is nearly identical to
the overall percentage of tank systems that are leaking (35%).
9-54
-------
9. Type of Pumping System
Conclusive test results were obtained for 71 tanks with
pressure pumping delivery systems. The results of the analysis
of these tests are displayed in Table 9-23. Under test
conditions, 41 percent of tank systems that include a pressure
pumping delivery system are judged to be leaking. This
percentage is only slightly higher than the overall percentage
(35%) of tank systems that were judged to be leaking.
10. Metered Dispensing Systems
As shown in Table 9-23, based on tests of 351 tank systems,
36 percent of tanks systems with metered dispensing systems are
judged to be leaking under test conditions. This percentage is
nearly identical to the overall percentage of tanks judged to be
leaking.
11. Self-Installed Tanks
Forty-two of the tank systems for which conclusive test
results are available were reported to be installed by the
establishment itself. The leak status of these self-installed
tank systems is displayed in Table 9-23. Among self-installed
tank systems, 23 percent were judged to be leaking, which is
twelve percentage points lower than the overall percentage of 35
percent.
9-55
-------
12. Secondhand Tanks
About 4 percent of the tanks overall were secondhand when
installed. Thirteen systems with secondhand tanks were included
in the tank testing program and provided conclusive test results,
which are displayed in Table 9-23. Based on this small number of
conclusive tests, 52 percent of the systems with secondhand tanks
were found to be leaking under test conditions.
V. LEAK RATES OF UNDERGROUND MOTOR FUEL STORAGE TANK SYSTEMS
This section provides an analysis of the estimated mean and
median leak rates for tank systems that were judged to be leaking
under test conditions. Analyses of mean (average) leak rates are
based only on those tank system test results which provided valid
system leak rates. Tank systems with leak rates that were coded
by the test crew as "unquantifiably large" are excluded from the
calculations of the mean leak rates. These comprise 20 percent
of the tank systems judged to be leaking. The largest quantified
leak rate (after adjustment) was 3.04 gallons per hour. Tank
systems that were judged to be leaking and have unquantifiably
large leaks were included in the analyses to derive the median
leak rates, however. The leak rates reported in this section,
whether mean or median, are rates in gallons per hour as adjusted
to operating pressures. The adjustment procedure is described in
detail in Section 8 and Appendix D of this report.
Each table describes in footnotes the set of test results
each column is based on. The medians are based on more cases
than the means, since the unquantifiably large leaks are included
in calculating the median but not in calculating the mean.
Further, some tables include results from manifolded systems not
9-56
-------
separated for testing, and others, based on characteristics of
individual tanks (such as tank age), do not. The total mean and
median leak rates are different in these two groups. The overall
mean leak rate, based on all quantified tank systems judged to be
leaking (118 cases) is 0.32 gallons per hour (0.24-0.39 gallons
per hour). The mean for the tests which are of single tanks and
associated piping (109 cases) is 0.31 gallons per hour (0.23-0.39
gallons per hour). The median leak rate for all tests (149
cases) is 0.25, while the median for single tank system tests
(138 cases) is 0.21.
Mean leak rates for tanks with quantified leak rates are
calculated as the weighted arithmetic mean, using sample weights,
which allows us to project from the sample to the universe. It
is presented as a commonly used and well understood measure of
"average" value. However, the mean has two drawbacks which the
median does not suffer from: data from unquantifiably large
leaks cannot be incorporated in calculating the mean (which
biases the mean towards zero) ; and a single large quantified
value can affect the mean greatly (pulling it away from zero),
while having no impact on the median. The median is the "middle
value" — i.e., half the values are larger, and half are smaller
(including the unquantifiably large leaks at the large end of the
scale). It is calculated on an unweighted basis. In a data set
with a non-symmetric distribution and significant outliers, the
mean and median measure different things, and a discrepancy
between them indicates skewness or outliers in the data, which
should be kept in mind in undertaking any interpretation.
The overall mean leak rate for tanks leaking under test
conditions is 0.32 gallons per hour, and the median leak rate is
0.25 gallons per hour. The distribution of observed leak rates
(adjusted to operating conditions) of the 149 tank systems judged
9-57
-------
to be leaking is shown below. Note that the intervals are not of
equal width.
Proportion of
test results
Interval in interval
Up to 0.0549 11%
(0.0550-0.149) 30%
(0.150-0.249) 10%
(0.250-0.349) 5%
(0.350-0.449) 5%
(0.450-0.549) 5%
(0.550-1.49) 10%
Quantified, >1.49 3%
Unquantifiably large 21%
A. Mean and Median Leak Rates within Survey Regions and
Establishment Types
Within survey regions, mean leak rates for leaking tank
systems ranged from a low of 0.24 gallons per hour for the
Northeast survey region to a high of 0.48 gallons per hour for
the Central survey region (Table 9-24). The Midwest and Pacific
survey regions have the lowest median leak rates (of 0.13 gallons
per hour) and the Mountain survey region has the highest median
leak rate (0.44 gallons per hour) followed closely by the Central
survey region with 0.41 gallons per hour.
Within establishment types, leaking tank systems at
government and military establishments, gas stations owned by
other companies and large establishments have leak rates that
range between 0.24 and 0.26 gallons per hour, which is below the
overall average of 0.32 gallons per hour (Table 9-25). Leaking
tank systems at gas stations owned by major petroleum companies
9-58
-------
Table 9-24
Estimates by survey region mean and median leak rates
among tank systems1'2 judged by be leaking under test
conditions (95% confidence bounds in parentheses)
Survey
Region
1
Northeast
2
Southeast
3
Midwest
4
Central
5
Mountain
6
Pacific
Total
Sample
Size
(N3)5
29
28
23
17
9
12
118
Mean
adjusted3
leak rate
(gph)
0.24
(0.23-0.26)
0.30
(0.17-0.43)
0.32
(0.06-0.57)
0.48
(0.26-0.71)
0.34
(0.27-0.41)
0.33
(0.02-0.69)
0.32
(0.24-0.39)
Sample
size
(N4)5
38
39
26
22
12
12
149
Median
adjusted
leak rate
(gph)
0.22
0.34
0.13
0.41
0.44
0.13
0.25
1In this table tank system tests results are reported for single tank
systems unless the tanks were tested as a part of a manifolded tank
system that was not broken apart. These manifolded tank systems are
included in this table.
2Does not include farm tanks.
3Leak rates of leaking tank systems were adjusted to operating
pressure.
Calculation of median adjusted leak rate included tank systems judged
to have unquantifiably large leaks.
5N3 = Number of tank systems judged to be leaking under tests
conditions that had quantifiable leak rates.
N4 = N3 + those tank systems judged to be leaking under test
conditions that had unquantifiably large leaks.
9-59
-------
Table 9-25. Estimates by establishment type of mean and median leak
rates among tank systems 1/2 judged to be leaking under
test conditions (95% confidence bounds in parentheses)
Establishment
Type
Sample
size
(N3)5
Mean
adjusted^
leak rate
(gph)
Sample
size
(N4)5
Median4
adjusted3
leak rate
(gph)
Government
and military
Gas stations
owned by major
petroleum
companies
Gas stations
owned by other
companies
Other fuel-
related estab-
lishments
Large non-fuel
related estab-
lishments
14
18
30
33
23
0.26
(0.06-0.47)
0.42
(0.18-0.68)
0.24
(0.13-0.34)
0.45
(0.20-0.71)
0.25
(0.14-0.36)
20
21
46
36
26
0.27
0.29
0.28
0.32
0.14
Total
118
0.32
(0.24-0.39)
149
0.25
1In this table tank system tests results are reported for single tank
systems unless the tanks were tested as a part of a manifolded tank
system that was not broken apart. These tank systems are included in
this table.
^
zDoes not include farm tanks.
3Leak rates of leaking tank systems were adjusted to operating pressure.
Calculation of median adjusted leak rate includes tank systems judged
to have unquantifiably large leaks.
5N3 = Number of tank systems judged to be leaking under test conditions
that had quantifiable leak rates.
N4 = N3 + those tank systems judged to be leaking under test conditions
that had unquantifiably large leaks.
9-60
-------
and other fuel-related establishments have mean leak rates that
are higher (0.42 and 0.45 gallons per hour respectively) than the
overall average. Leaking tank systems at large non-fuel-related
establishments have a median leak rate that is lower (at 0.14
gallons per hour) than the overall median rate of 0.25 gallons
per hour.
B. Mean and Median Leak Rates for Tank Systems with
Selected Characteristics
Tables 9-26 through 9-30 below display analyses of the mean
and median leak rates for leaking tank systems with selected
characteristics. The characteristics include tank age, tank
size, type of product held in the tank, and various installation
characteristics.
1. Mean and Median Leak Rates within Tank Age
Categories
Table 9-26 shows the mean and median leak rates for the
tanks judged to be leaking under test conditions. Leaking tanks
that are less than five years old and leaking tanks of unknown
age have the highest mean leak rate (0.46 gallons per hour).
Leaking tanks that are less than five years old also have the
highest median leak rate.
The lowest mean and median leak rates were found among
leaking tanks that are five to eight years old and leaking tanks
that are more than 20 years old.
9-61
-------
Table 9-26. Estimated mean and median leak rates among tank systems1'2
judged to be leaking under test conditions within tank
age categories. (95% confidence bounds in parentheses)
Tank Age
Categories
< 5 years old
5-8 years old
9-12 years old
13-20 years old
> 20 years old
Age unknown
Total
Sample
size
(N5)5
14
17
20
10
17
30
109
Mean
adjusted
leak rate
(gpn)
0.46
(0.20-0.71)
0.14
(0.09-0.18)
0.27
(0.12-0.42)
0.22
(0.10-0.34)
0.15
(0.11-0.20)
0.46
(0.21-0.71)
0.31
(0.23-0.39)
Sample
size
(N6)5
21
20
21
18
21
36
138
Median4
adjusted3
leak rate
(gph)
0.71
0.11
0.18
0.52
0.16
0.39
0.21
1In this table tank system test results are reported for single tank
systems. Tank systems that were tested as a part of a manifolded tank
system that was not broken apart are not included.
2Does not include farm tanks.
3Leak rates of leaking tank systems were adjusted to operating pressure.
Calculation of median adjusted leak rate includes tank systems judged to
have unquantifiably large leaks.
5N5 = Number of tank systems judged to be leaking under tests conditions
that had quantifiable leak rates.
N6 = N5 + those tank systems judged to be leaking under test conditions
that had unguantifiably large leaks.
9-62
-------
Table 9-27. Estimated mean and median leak rates among tank sytems ^-
judged to be leaking under test conditions within tank
size categories. (95% confidence bounds in parentheses)
Tank Size
Categories
< 1100 gallons
1101 to 3999
gallons
4000 to 4999
gallons
5000 to 5999
gallons
6000 to 7999
gallons
8000 to 9999
gallons
10,000 to 11,999
gallons
> 12,000 gallons
Sample
Size
(N5)5
16
13
18
6
10
8
28
10
Mean
adjusted3
leak rate
(gpn)
0.14
(0.08-0.21)
0.26
(0.09-0.43)
0.20
(0.13-0.28)
0.15
(0.11-0.18)
0.35
(-0.07-0.77)
0.53
(0.30-0.76)
0.30
(0.18-0.43)
0.83
(-0.05-1.72)
Sample
size
(N6)5
21
16
24
10
17
12
28
10
Median4
adjusted3
leak rate
(gph)
0.10
0.18
0.21
0.23
1.24
0.89
0.12
0.37
Total
109
0.31
(0.23-0.39)
138
0.21
^•In this table tank system test results are reported for single tank
systems. Tank systems that were tested as a-part of a manifolded tank
system that was not broken apart are not included.
2Does not include farm tanks.
3Leak rates of leaking tank systems were adjusted to operating pressure.
Calculation of median adjusted leak rate includes tank systems judged to
have unquantifiably large leaks.
5N5 = Number of tank systems judged to be leaking under test conditions
that had quantifiable leak rates.
N6 = N5 + those tanks judged to be leaking under test conditions that had
unquantifiably/large leak.
9-63
-------
Table 9-28. Estimates by fuel types stored during the prior year of the
mean and median leak rates among tank systems1'2 judged to
be leaking under test conditions (95% confidence bounds in
parentheses)
Fuel types
Sample
size5
(N3)
Mean
adjusted3
leak rate
(gph)
Sample
size
(N4)5
Median4
adjusted3
leak rate
Leaded gasoline
120
Unleaded gasoline 42
Diesel fuel
50
0.22
(0.14-0.31)
0.36
(0.25-0.47)
0.27
(0.17-0.37)
24
60
56
0.22
0.49
0.14
1In this table tank system test results are reported for single tank
systems unless the tank systems were tested as a part of a manifolded
tank system that was not broken apart. These manifolded tank systems
are not included in the table.
2Does not include farm tanks.
3Leak rates of leaking tank systems were adjusted to operating pressure.
Calculation of median adjusted leak rate includes tank systems judged to
have unquantifiably large leaks.
5N3 = Number of tank systems judged to be leaking under test conditions
that had quantifiable leak rates.
N4 = N3 + those tank systems judged to be leaking under test conditions
that had unquantifiably large leaks.
9-64
-------
^ I
Table 9-29. Estimated mean and median leak rates among tank systemsx/
judged to be leaking under test conditions for tanks in
manifolded systems and single tanks. (95% confidence
bounds in parentheses.)
Tank type
Sample
Sizec
(N3)5
Mean
adjusted3
leak rate
(gpn)
Sample
size
(N4)5
Median4
adjusted
leak rate
(gph)
Single tanks 93
(not manifolded)
Tanks in mani- 25
folded systems
0.27 109
(0.20-0.35)
0.49 40
(0.22-0.75)
0.16
0.71
Total
118
0.32
(0.24-0.39)
149
0.25
1In this table tank system test results are reported for single tank
systems unless the tanks were tested as a part of a manifolded tank
system that was not broken apart. These manifolded tank systems are
included in the table.
2Does not include farm tanks.
3Leak rates of leaking tanks were adjusted to operating pressure.
Calculation of median adjusted leak rate included tank systems judged to
have unquantifiably large leaks.
5N3 = Number of tank systems judged to be leaking under test conditions
that had quantifiable leak rates.
N4 = N3 + those tank systems judged to be leaking under test conditions
that had unquantifiably large leaks.
9-65
-------
Table 9-30.
Estimated mean and median leak rates among tank systems1'2
judged to be leaking under test conditions for tanks with
selected installation characteristics. (95% confidence
bounds in parentheses.)
Fiberglass tanks
Bare (uncoated
steel) tanks
Coated steel tanks
Tanks installed
partially or com-
pletely below the
water table
Tanks covered by
a paved surface
Tanks connected to
a pressure pump
delivery system
Tanks with metered
dispensing systems
Tanks that were
self-installed
Total
Sample
Size
(N5)5
110
8
55
16
79
23
101
9
109
Mean
adjusted3
leak rate
(gpn)
0.42
(0.22-0.61)
0.60
(0.12-1.08)
0.26
(0.19-0.32)
0.49
(0.09-0.89
0.29
(0.22-0.37)
0.25
(0.12-0.39)
0.32
(0.23-0.40)
0.26
(-0.05-0.56)
0.31
(0.23-0.39)
Sample
size
(N6)5
10
10
75
19
100
29
129
11
138
Median4
adjusted
leak rate
(gpn)
0.14
0.39
0.23
0.20
0.20
0.18
0.22
0.13
0.21
1In this table tank system test results are reported for single tank
systems. Tanks that were tested as a part of a manifolded tank system
that was not broken apart are not included.
2Does not include farm tanks.
3Leak rates of leaking tank systems were adjusted to operating pressure.
Calculation of median adjusted leak rate included tank systems judged to
have unquantifiably large leaks.
5N5 = Number of tank systems judged to be leaking under tests conditions
that had quantifiable leak rates.
N6 = N5 + those tank systems judged to be leaking under test conditions
that had unquantifiably large leaks.
9-66
-------
2. Mean and Median Leak Rates within Tank Size
Categories
Table 9-27 shows the mean and median leak rates for eight
size categories among tank systems judged to be leaking under
test conditions. Systems with tanks that are 12,000 gallons or
more in size have the largest mean leak rate (of 0.83 gallons per
hour), and systems with tanks in the smallest size category (1100
gallons or less) have both the smallest mean leak rate (0.14
gallons per hour) and the smallest median leak rate (0.10 gallons
per hour). The tank size category with the largest median leak
rate is the 6000 to 7999 gallon category, where the median was
1.24 gallons per hour, due in large part to the fact that seven
of the seventeen tanks tested in this category have
unquantifiably large leaks.
3. Mean and Median Leak Rates by Types of Motor Fuel
Stored
Table 9-28 shows mean and median leak rates for leaking
tanks storing different types of motor fuels. Leaking tank
systems that store unleaded gasoline have the highest mean (0.35
gallons per hour) and median (0.49 gallons per hour) leak rates.
Leaking tank systems that store leaded gasoline have the lowest
mean leak rate (of 0.22 gallons per hour), but leaking diesel
tank systems have the lowest median leak rate (of 0.14 gallons
per hour).
9-67
-------
4. Mean and Median Leak Rates for Manifolded
Tank Systems
Table 9-29 shows the mean and median leak rates for leaking
tanks that are in single tank systems and those that are in
manifolded tank systems. Both the mean and the median leak rates
are higher for leaking tank systems that are manifolded. Leaking
Tank systems that are manifolded have a mean leak rate of 0.49
gallons per hour (which is nearly 50% higher than the overall
mean leak rate of 0.31 gallons per hour) and a median leak rate
of 0.71 gallons per hour, which is nearly three times greater
than the overall median leak rate of 0.25 gallons per hour.
5. Material of Construction
Table 9-30 shows the mean and median leak rates for leaking
tank systems that have tanks which are fiberglass, bare
(uncoated) steel, and coated steel. Leaking tank systems with
tanks that are bare steel have the highest mean (0.60 gallons per
hour) and median (0.39 gallons per hour) leak rates, but the
category also has a small sample size. Leaking tank systems with
coated steel tanks have the lowest mean leak rate, of 0.25
gallons per hour. Although there are a large number of tank
systems with unquantifiably large leaks in this category, the
median leak rate is only 0.23 gallons per hour.
6. Water Table Level
The mean leak rate for leaking tank systems installed in or
beneath the water table is 0.49 gallons per hour, based on 16
9-68
-------
tank systems tested (see Table 9-30). The median leak rate among
these tanks is 0.20 gallons per hour.
7. Surface over Tank
As is shown in Table 9-30, the mean and median leak rates
for tank systems where the tank is covered by a paved surface is
nearly equal to the overall mean and median leak rates.
8. Mean and Median Leak Rates for Tank Systems with
Pressure Pump Distribution Systems
Leaking tank systems with pressurized distribution systems
have lower mean and median leak rates than leaking tank systems
overall, as is shown in Table 9-30. The mean leak rate for tank
systems with pressurized distribution systems is 0.25 gallons per
hour, and the median is 0.18 gallons per hour.
9. Mean and Median Leak Rates for Tank Systems with
Metered Dispensing Systems
Leaking tank systems that have metered dispensing systems
have mean and median leak rates that are nearly identical to the
overall mean and median leak rates. (See Table 9-30.)
9-69
-------
10. Mean and Median Leak Rates for Tank Systems that
Were Self-Installed
Table 9-30 shows the mean and median leak rates for leaking
tank systems which were installed by the establishment itself.
Only a small number of tank systems in this category were tested,
in view of which the mean and median rates may be regarded as not
substantially different from the overall means and medians.
VI. STATISTICAL ASSOCIATIONS OF LEAK STATUS AND LEAK RATE WITH
OTHER VARIABLES
A. Introduction
In this section we present the results of statistical
correlations found between leak status and each of 49 possible
explanatory variables. Similarly, results are reported for leak
rate.
In addition, multivariate and logistic models were developed
to identify significant predictor variables. Because these
models resulted in rather low predictive power, they are
considered preliminary, and further research is required. These
models and their development are detailed in Appendix I.
B. Simple Correlations
No single explanatory variable, among the 49 examined, had a
strong correlation with either leak status or leak rate — i.e.,
no correlation coefficient was larger than .34 (Table 9-31). The
9-70
-------
Table 9-31. SIHPLE CORRELATION OF LEAK STATUS AND LEAK RATE WITH EXPLANATORY VARIABLES
Explanatory
Variable
X1
X2
XJ
X4
X52
X6
X7
X8
X9
X10
X11
X12
X13
X15
X16
X17
X18
X19
X20
X21
X23
X24
X25
X26
X27
X28
X29
X30
X31
Meaning
Gas Station
1 Underground tanks
Tank capacity
Average low fill level'3'
(Age of tank)2
Leaded gasoline
Diesel fuel
Aviation fuel
Gasohol
Other
Suction pump
Depth buried
Water level
Tank tested
Years since test
Tank material
Tank lined
Tank coated
Passive cathodic protection
Impressed current cath.
protection
Other protection
Previous tank leak
Previous line leak
Frequency of delivers
Sand fill
Gravel fill
Concrete pad
Packed earth pad
Oist. to nearest tank or
structure
Definition
1 = Yesj 0 = No
Number at facility
Gallons
As fraction of tank capacity
in (years)
1 = yes 0 = No
1 = Yes 0 = No
1 = Yes 0 = No
1 = Yes 0 - No
1 = Yes 0 = No
1 = Yes 0 - No
Inches from surface
to top of tank
Inches from surface
to water table'4'
1 if tested after placed
in service; 0 otherwise
Since most recent test
1 = steel; 0 = fiberglass
1 = Yes; 0 = No
1 = Yes; 0 = No
1 = Yes; 0 = No
1 = Yes; 0 = No
1 = yes; 0 = No
1 - Yes; 0 = No
1 = Yes; 0 = No
Number per year
1 = Yes; 0 = No
1 = Yes; 0 = No
1 = Yes; 0 = No
1 = Yes; 0 = No
(feet)
Correlation'1' with Y1,
Leak status
(1 = Leak; 0 = No Leak)
-.08
.12
.14
-.05
.11
-.26
.24
.13
-.07
.08
.003
.10
-.15
.03
.06"
.02
.07
-.01
.10
0
-.08
-.05
.05
-.05
.03
.006
.07
.03
-.04
Correlation'1' with Y2,
Leak rate (gal/Hr),,
among leaking tanks'^)
-.06
.10
.34
-.07
-.20
-.11
-.08
.07
0
.29
-.12
-.006
-.005
.01
-.21
-.09
.02
-.25
.05
0
0
-.04
.23
-.003
-.10
.16
-.09
-.09
-.09
Pearson's correlation coefficient; Kendall's Tau-8 was also calculated for all Y1 correlations and found to be the same for
nearly every variable.
Using data only from individual leaking tanks with quantifiable leaks.
I.e., just before product is added.
A
At time of test.
9-71
-------
Table 9-31. SIMPLE CORRELATION OF LEAK STATUS AND LEAK RATE WITH EXPLANATORY VARIABLES (Continued)
Explanatory
Variable
X32
X33
X34
X35
X36
XT3
XT4
XT18A
XT19
XT 20
XT 36
XB5
X819
XC7
XC8
xriA
XG20
XG2E
XG2F
XG2G
Meaning
Interaction: age i material
Interaction: gaaohol &
material
Permit to install
Permit to store
Average high fill leveiC6)
Average fuel delivery
Max. ever stored
Attached to other tank
Tank proximity to water
table
Manway with tank
Not self-installed
Remote gauge
Log of deliveries
Any abandoned tank(^)
Definition
(X5 ) (1-X17)
X9 (1-X17)
1 = Yes; 0 = No
1 = Yes; 0 = No
As fraction of tank capacity
in gallons (to one tank)
gallons
1 = Yes; 0 = No
1 = above; 2 = partially
above; 3 = below; 4 s other
1 = Yes; 0 = No
1 = Yes; 0 = No
1 = Yes; 0 = No
1 = Yes; 0 = No
1 s Yes; 0 = No
1 Abandoned tanks | (coded as zero if none)
Corrosion prevention equip./
mat.
Trained to check pump
Trained to check line leaks
Trained to check leak
prevention
Trained to check leak
monitoring
1 = Yes; 0 = No
1 = Yes; 0 s No
1 = Yes; 0 = No
1 = Yesj 0 = No
1 = Yes; 0 = No
Correlation^) with VI,
Leak status
(1 = Leak; 0 = No Leak)
-.03
0
.12
.02
-.06
.15
.11
.22
.13
.19
.12
-.005
-.03
-.03
• .12
-.02
.14
.10
.10
.15
Correlation^1) with Y2,
Leak rate (gal/Hr),
among leaking tanks'''
-.07
0
.17
.09
-.09
.23
.29
.24
.28
.13
.12
.05
.002
.03
-.09
-.12
.24
.18
.15
.17
At that facility.
I.e., Oust after product is delivered.
9-72
-------
highest correlations with "leak status" (i.e., whether or not a
tank is leaking) were found with the following variables:
Correlation
Variable Coefficient
Leaded gas -0.26
Diesel fuel +0.24
Tank manifolded +0.22
Tank has manway +0.19
Among leaking tank systems, the leak rate had stronger
correlations. Eleven variables had a correlation coefficient
larger than .20, as shown below:
Correlation
Variable Coefficient
Tank capacity +0.34
Maximum ever stored +0.29
Other fuel type +0.29
Proximity to water table +0.28
Tank coated -0.25
Attached to other tank +0.24
Trained to check pump +0.24
Previous line leak +0.23
Average fuel delivery (gal.) +0.23
Years since tested -0.21
(Age of tank)2 -0.20
The last variable in the list above is "age squared." It
was used rather than "age" because of the non-linear relationship
suggested by the plotted data. The sample size for most of these
correlation coefficients is about 380. Due to missing data, some
sample sizes are smaller, but all had a sample of 200 or more
except X13 (N = 89), X16 (N = 72), and X18 (N = 173).
Correlations with leak rate were calculated using all tanks that
9-73
-------
were tested for leaks. Somewhat different correlations might be
found if only leaking tanks were included.
While this represents a considerable number of variables
with non-trivial correlations with leak status and leak rate,
none of the correlations would be considered strong and,
therefore, no single variable will have a strong predictive
ability.
Some variables of possible interest, such as soil
characteristics, were not included because they were not
collected in this study. However, data were included on the back
fill material where used (sand, gravel, etc.).
C. Multiple Regression and Logistic Models
Multivariate models were developed to explain leak status
and leak rate in terms of various predictor variables. Although
some statistically significant relationships were found, the
overall predictive ability of the models was low. The models
could account for only 8 percent of the variance in leak status
and 20 percent of the variance in leak rate. Therefore, further
research in this area is still required. Appendix I describes
the model development method and specifies which variables were
statistically significant. Logistic and regression models
identified several of the same variables as significant.
9-74
-------
SECTION 10
INVENTORY RECONCILIATION TECHNIQUES
I. INTRODUCTION
Analysis of properly collected motor fuel inventory data
offers the potential for an inexpensive, readily available
approach to detecting tank-product losses, including those
resulting from leaking tanks and piping. Inventory data is
collected typically at the close of each day of operation by the
tank establishment operator, records (1) volume of fuel in each
tank as measured by metering stick or gauge, (2) delivery
volumes, and (3) dispensing meter readings. From these
measurements, the volume of fuel metered through a tank system's
dispensers is reconciled with the physical measurement (based on
stick readings and delivery volumes) of product gone from the
tank.
Rarely are these two measurements of volume of daily
through-put (physical versus dispensing meter) numerically equal,
even when there has been no loss of product. Rather, they will
show some variance either as an "overage" (a numerical excess of
product in the tank) or an "underage" (a numerical loss of
product). Some part of the daily variance will be due to random
errors of measurement; however, inaccurate gauging and metering
devices add to the variance, as do. temperature-induced product
shrinkage and expansion, vapor loss, theft, and leakage of
product from (or of water into) the system. To further
complicate interpretation of inventory reconciliation data,
several of the above factors may contribute both to overages and
underages.
10-1
-------
There are currently available several commercially developed
computerized models which can be used to identify and quantify
factors contributing to daily inventory variances. These models
are proprietary and, in any case, are much too sophisticated to
be implemented and interpreted by a typical tank owner or
operator. To monitor inventory using these models, a tank owner
must contract for the services of one of the firms which has
developed such a model. To provide tank owners and operators
with a simple, inexpensive procedure for monitoring inventory,
EPA has developed a procedure based on counts of the number of
negative daily variances (underages) in successive monthly
periods.1 Because of its simplicity, the EPA method would not be
expected to be as accurate as the more sophisticated modeling
techniques. Tank tightness test data and inventory
reconciliation data collected during two phases of the National
Survey of Underground Motor Fuel Storage Tanks make possible a
determination of the extent of agreement between tank tightness
tests and the various methods of inventory reconciliation
analysis, as well as the extent of agreement among the latter.
At the time the inventory analysis was conducted, the survey
had provided complete, properly-collected inventory data on 855
tank systems. While this represents only the first portion of
the inventory returns, the inventory data collection and editing
effort resulted in 41 percent of the attempted cases providing
usable data. Of these 855 tank systems, 511 were analyzed using
the inventory reconciliation model developed by Warren Rogers
Office of Toxic Substances, "More about Leaking
Underground Storage Tanks: a Background Booklet for the
Chemical Advisory," (October 1984).
10-2
-------
Associates.5* In addition, in a smaller study, 18 tank systems
were analyzed by Entropy Limited. Tightness test results were
available for 189 of the 855 tanks for which usable inventory
data was available. The EPA inventory analysis method was
applicable, in modified form, to all 855 tanks; modification was
required since the EPA method was intended for application to on-
going monitoring programs and not to a single set of one-time
inventory data.4
The present section provides an analysis of the extent of
agreement of the above inventory reconciliation methods with one
another and with tightness-test results, based on data collected
in the survey. In addition, we report the results of a small
quality control study in which the various inventory approaches
were applied to a simulated set of inventory data for five tanks,
Mathematical techniques were used to simulate various
combinations of stick error, leakage and theft for the five
tanks. The results of the inventory analyses were then compared
with the true condition of the hypothetical tanks which in this
case, of course, was known exactly.
II. METHODS AND DATA
The survey protocol called for the collection of 30 days of
inventory data on each tank or manifolded tank system at the
2Warren Rogers Associates, Inc., "Inventory Reconciliation
System," (undated)
3Entropy Limited, "Precision Tank Inventory Control," (1984).
4USEPA, Office of Toxic Substances, "More about Leaking
Underground Storage Tanks: a Background Booklet for the
Chemical Advisory," (October 1984).
10-3
-------
sampled establishments. Many respondents were unable to supply
proper inventory data (see Appendix B for details), with the
result that complete, usable data was obtained for only 855 tanks
or manifolded tank systems. Of these tanks, 511 were analyzed by
Warren Rogers Associates (WRA). Eighteen of the 511 tanks were
also analyzed by Entropy Limited (EL). Of the 439 tank systems
in the tightness test (TT) sample, 189 had usable inventory data.
Table 10-1 shows the available sample sizes for all possible
pairwise comparisons between methods.
Table 10-1. Sample sizes for pairwise comparisons between
methods
EPA WRA EL
EPA 855 511 18
WRA 511 17
EL 18
TT
TT
189
106
17
439
The EPA-developed inventory analysis method is based on a
simple count of the number of days for which the inventory
reconciliation shows a negative variance, i.e., a numerical loss
of product. An excess of days with negative variance over those
with positive (or zero) variance may, under certain
circumstances, be interpreted as a loss of product due to leakage
or some other cause. The EPA method was developed and calibrated
for application to an on-going monitoring program, in which
cumulative month-by-month counts of negative variances would be
compared with statistically-derived "action numbers" to determine
whether there was evidence of a systematic deficit in inventory.
Calculations indicated that the method would be effective in
10-4
-------
detecting even relatively small leaks over a sufficiently long
period of monitoring, at least when reasonably good inventory
records were kept.5 The method was not intended for application
to a one-time collection of 30 days inventory and, indeed, would
not be expected to detect smaller leaks based on such a small
data set. For the present comparisons, the method was modified
and the following decision rule adopted: A tank system is
declared leaking if the 30-day record exhibits 18 or more
negative daily variances. Calculations detailed in Appendix E,
indicate that this rule has approximately a five percent false-
positive (or "false alarm") rate. That is, there is a five
percent chance that a tank system which is not leaking and whose
inventory record is subject only to random stick measurement
error, would be declared leaking. This is consistent with the
definition of a leak adopted for the tightness test procedure,
see Section 8. However, the modified EPA inventory analysis
method has significantly poorer detection capability than
tightness testing, even in optimistic scenarios where the
inventory record is subject only to stick measurement error and
not to other sources of discrepancy, such as delivery errors.
For example, the chance of detecting a leak of 0.1 gallons per
hour (2.4 gallons per day) in a typical tank is approximately 17
percent, as opposed to 95 percent for tightness testing. A
detection capability of 80 percent or greater was obtained using
the EPA method for leaks in excess of 0.37 gallons per hour (8.9
gallons per day). By substantially increasing the number of days
of inventory data, an 80 percent detection capability would be
possible for smaller leaks.
^David C. Cox, "Performance of the Chemical Advisory Inventory
Analysis Method Under Various Scenarios," Report from Battelle
Columbus Laboratories to EPA under Contract No. 68-01-6721
(April 1984)
10-5
-------
Both the WRA and EL procedures are proprietary, so that
details of the methodology and decision rules are available only
in sketchy form. A description of the methods, based on
literature provided by their developers, is given in Appendix E.
The literature mentioned is partly promotional in nature. Claims
made therein have not been investigated or verified by EPA,
except to the extent reported here, so that no endorsement of the
methods should be inferred. In order to place the WRA and EL
methods on the same footing as the EPA and tightness test
approaches, it is important that the false positive rate of the
WRA and EL methods also be five percent. According to the
developers of these methods, the inventory analysis results they
have provided meet this requirement.
The WRA, EL and tightness test methods occasionally fail to
produce a definite conclusion as to whether a tank system is
leaking or not leaking. This occurs for the inventory methods
whenever the data is excessively noisy due, for example, to large
stick errors or very frequent deliveries. For tightness tests,
various physical problems may lead to an indeterminate test
result (see Section 8.) Table 10-2 presents a breakdown of the
results for each method. It is reasonable that the tightness
test procedure reports the largest percent leaking; this method
should have the best detection capability. Likewise, the EPA
procedure should have the lowest percent leaking, as it does. In
the next section, we examine agreement between the methods, i.e.,
the extent to which they agree, not just on percent leaking but
on which tank systems are leaking and which are not leaking.
10-6
-------
Table 10-2. Number and percent of tank systems judged to be leaking,
judged not to be leaking, and providing inconclusive results
for inventory methods and tightness testing
Tank Systems
Judged to
Tank Systems
Judged to
Tank Systems with
Method
TT
EPA
WRA
EL
Sample
Size
439
855
511
18
be Leaking
Number Percent
152 35%
149 17%
160 31%
4 22%
be li
Number
259
706
294
11
•UdC 1
Percent
59%
83%
58%
61%
Inconclusive Result,
Number Percent
28 6%
0 0%
57 11%
3 17%
1May differ from national estimates because survey sampling weights are not considered here.
10-7
-------
III. COMPARISON OF METHODS
In comparing two methods of deciding whether or not a tank
system is leaking, one cannot focus simply on the degree of
agreement between the predictions of the methods. To see why,
consider two methods which give a correct prediction in 70
percent of cases. If the methods were completely independent,
one would expect them to agree in 49 percent of cases, according
to the rules of probability. This 49 percent represents the
degree of agreement expected purely by chance and not due to any
tendency for the methods to act in the same direction. In this
example, agreement in significantly more than 49 percent of cases
is required before one can conclude that the methods really are
in substantial agreement. In this section, a statistic, K, is
used to measure the agreement between methods above and beyond
what is expected by chance.6 The value K = 0 corresponds to
purely chance agreement, while K = 1 means perfect agreement.
Values of K between 0 and 1 may be interpreted on an ordinal
scale, i.e., the larger K is, the better the agreement.
Quantitative interpretation of K is more elusive, e.g., it is
difficult to determine just how much agreement is represented by
a value of, say, K = 0.5.
For each pairwise comparison between methods considered, we
carry out a statistical test to determine whether there is
sufficient evidence to conclude that K is positive, i.e., that
there is more than chance agreement between the methods. In
performing this test we have ignored inconclusive results for the
various methods. We have also treated the data as if it were
generated by a simple random sample of tank system, ignoring the
6Yvonne M. M. Bishop, Stephen E. Fienberg and Paul W. Holland,
"Discrete Multivariate Analysis: Theory and Practice," MIT
Press, Cambridge, MA (1975)
10-8
-------
survey sample weights. Tables 10-3 through 10-5 present
comparisons between the tightness test, WRA and EPA.
The two inventory methods show agreement with each other
beyond what one would expect by chance alone. However, each
inventory method exhibits only chance agreement with tightness
test results. This conclusion should be regarded as tentative
since the sample sizes for even the overall inventory —
tightness testing comparisons reported here are not very large.
More detailed analyses of the agreement between the methods would
not be statistically meaningful. It is, however, worth pointing
out that the agreement does not appear to be improved even if we
restrict attention to large leaks (as measured by tightness
testing). For example, consider quantifiable leaks exceeding
4 gallons per day. The EPA procedure found 6 out of 23 (26%),
for which a comparison was possible, to be leakers; WRA found 4
out of 7 (57%) leaking.
Statistically meaningful comparisons with the EL results are
not possible because of the very small number of tanks evaluated
by Entropy. However, the data confirm the above two findings in
a general way. Inventory methods agree with one another but not
with tightness test results. The extent to which inventory and
tightness testing may be measuring differing phenomena, as is
suggested by these results, is not clear. It is possible that
certain measured leaks may not represent operational leaks. For
example, leaks at the very top of the tank would occur in a
tightness test, but might not occur in practice if the tank were
never filled to the top. Likewise, a phenomenon such as theft
may be reported as a leak by the inventory methods while the tank
system tests tight. The resolution of this question will require
more detailed analyses of the survey data and, possibly,
collection of longer series (more than 30 days) of inventory
data.
10-9
-------
Table 10-3. Comparison of EPA Inventory Reconciliation Method
with Warren Rogers Associates Inventory Reconciliation
Method
WRA
Number of Number of
tank systems tank systems
judged to judged to
be leaking be tight Inconclusive
Number of tank
systems judged
to be leaking 61 19 10
Number of tank
systems judged
to be tight 99 275 47
Percent agreement = 74%
K - 0.36 (STATISTICALLY SIGNIFICANT)
10-10
-------
Table 10-4. Comparison of EPA Inventory Reconciliation Method
with Tightness Testing
TT
Number of Number of
tank systems tank systems
judged to judged to
be leaking be tight Inconclusive
Number of tank
systems judged
to be leaking 13 22 1
EPA
Number of tank
systems judged
to be tight 37 102 14
Percent agreement = 66%
K = 0.09 (NOT STATISTICALLY SIGNIFICANT)
10-11
-------
Table 10-5.
Comparison of Warren Rogers Associates Inventory
Reconciliation Method with Tightness Testing
TT
WRA
Number of tank
systems judged
to be leaking
Number of tank
systems judged
to be tight
Inconclusive
Number of
tank systems
judged to
be leaking
11
5
Number of
tank systems
judged to
be tight
28
38
8
Inconclusive
5
2
Percent agreement = 55%
K = 0.02 (NOT STATISTICALLY SIGNIFICANT)
10-12
-------
IV. QUALITY CONTROL SAMPLES
A limited quality-control study of the 3 inventory
reconciliation methods was conducted in order to compare
performance on a data set for which the true leak status could be
unequivocally determined. Inventory data for a total of 5 tanks
at 2 sites was generated using mathematical techniques to
simulate various combinations of stick error, leakage and theft.
To maximize realism, actual tank conversion charts were used to
simulate the effect of random measurement error due to sticking
the tank. The simulated data was provided blind to WRA and EL.
Table 10-6 shows the scenarios simulated. Table 10-7 shows the
results of the inventory analyses. In addition to EPA, WRA and
EL, we have added a simple t-test. The test is a standard t-test
with 29 degrees of freedom based on the 30 daily variances in the
inventory record. For consistency with the other inventory
methods, the false positive rate of the test is set at 5 percent.
Site 1 represents clean inventory data. There is no noise
in the record other than random measurement error due to sticking
the tank. The EPA method and the t-test correctly predicted leak
status for both tanks. WRA detected the leak but also classified
the non-leaker as a leaker. EL correctely classified the non-
leaker but reported the leaker as inconclusive (accurately
estimating the true leak rate, however).
Site 2 represent a more difficult test of the inventory
methods. Stick measurement error is unusually large. Moreover,
both the random pattern of theft in tank 2 and the relatively
small leak (3 gallons per day) in tank 3 would be expected to be
difficult to detect by any inventory method based on only 30
10-13
-------
Table 10-6. Simulated quality control inventory data
Site 1
Tank Size Product Description
1 10,000 Reg. unleaded 5 gals/day leak
2 10,000 Reg. leaded No leak, stick error only
Readings to nearest 1/2" on dipstick — typical random
measurement error of 14-19 gallons
Site 2
Tank Size Product Description
1 6,000 Reg. leaded Stick error only
2 6,000 Prem. unleaded Theft 15-20 gallons on
9 days
3 10,000 Reg. unleaded Leak, 3 gals/day
Readings to nearest 1" on dipstick — typical random measurement
error is 20-25 gallons for 6,000 gallon tanks, 35-40 gals for the
10,000 gallon tank
10-14
-------
8-
o
cu
CD
to
LU
4-1
_j
LU
3
LU
O£
1
V)
*i
1
§
0
s
2
4-*
C
I
CJ
s
to
_J
CJ
LU
CO
1
co
+-»
(0
•M
CO
CD
5 "S
.c t-
« 0
ro t-
E
(_ (A
CU C/)
.C 0
t— — '
NO
>d-
c,.
CO CO
cu c
.- .— CU
O CO D)
C O> t_
CO CO
a 73 -i ^
CU CD «-
t. C - ^
cj •— -^
CO CO -J- CO
73 "a. ^ o
X CO 1-
>~ 0 0 c_
C. C CO CU
CU 3 CO
> o ^
.— . _. CJ
0 CM t- •*-*
Q N-' O CO
CO
-
0
_J
V)
ro 03
"D D)
W N-
(D *~
0) II
-
^
1
-
o
Q.
CD
CD"
O)
CO
5 "S
^ t-
CO 0
(D L.
L- CO
x: o
i— — '
•
^
CO
CO 1-
01 O
••- £_
CJ 1-
C CD
CD
CU O
t_ '~-
0 4->
CO CO
"5 CD
0)
>^ 1-
C. CO
CU —1
CU CM CM
Q N^ *^
co
CM
1
in
z
(- C/]
o — <
L. ro
L. O
3
ro ro < o
0) —1 — .
•D - . l_
(D ^* *^ O
C ro «- t-
tD CU
— • « co
Q- a> t. .v
X CO O O
00 £_ •.-
C O L. 4J
Z3 — • CU CO
s
-
CO
-.
C- CO
o — •
C- CD
C- O>
a>
CM
^ CM
o X
t-> C
« ° b
-
CM
O
O
_J
C CD
CD D>
CD
CO c-
CO <. O
O) —
CU ^ O
C %O c_
ca CD
— • CO
CL CU ^
X CO CJ
C O Z
3 — • CO
ro
o
-
o
-.
CO
O
O •*- • V>
co x — • rvj
•M >. o ro CNJ
N*- ro t- en
cu "D Q. ii
^ Q. h—
h— ON
(M
ra
'
z
+-« *-•
— ' 3
"D <0 CO 4-»
C 3 CU C
W CD D) O"
C +-• C CU
.,- ro .^ i_
OJ Q C M-
o) ro
73 . i £ £
0 ^ CO
C «- . 73 0
.^ ^ CM
X CO 0 — •—
CU CO "O Q. — '
C O O X CU
3 — ' CJ 0 73
t
•x.
ro
x.
CO
>- ^
CO CO
73 O)
CO O
CO
0) II
ro V>
_i
K,
CM
10-15
-------
days' data. (The change of detecting the 3 gallons per day leak
using the EPA method is only about 10%.) Both EPA and WRA
predicted that the tight tank was leaking and that the leaker was
not; both found the theft case to be a leak, thus successfully
detecting the negative trend in the inventory, although
attributing it to the wrong cause. The t-test correctly
evaluated the non-leaker but also got the other two tanks wrong.
A surprising feature of both sophisticated methods (WRA and EL)
was a tendency to find effects in the data which were not in fact
present. Thus WRA found numerous unexplained gains and losses,
delivery discrepancies and large stick errors, while EL found
theoretical shrinkage and vapor loss. These spurious findings
apparently tended to obscure the true status of the tank systems.
It should be pointed out, of course, that a method of analysis
tailored to perform well on real-world, noisy data will of
necessity be less than optimal for unusually clean data such as
we have here. Moreover, the simulated data does not reflect the
effects of factors such as location and time-of-year that may be
very important to account for in real data analysis. The results
reported here must be interpreted in this light.
V. CONCLUSIONS
We have compared a number of inventory reconciliation techniques,
with each other and with the results of tank tightness tests,
using data from the survey, as well as a small set of simulated
inventory records. The sample sizes, especially for inventory
vs. tightness test comparisons, were somewhat small. Finally,
the data were analyzed as if they were generated by a simple
random sampling technique, rather than the sampling procedures
actually used in the survey. Thus, the conclusions reported here
10-16
-------
are tentative and require confirmation by further research. The
major conclusions are:
o Results of inventory analysis methods and tightness
tests exhibited only chance agreement;
o Inventory analysis methods agreed among themselves more
than would be expected by chance; and
o On simulated data, all inventory methods were
comparable in predictive power. However, the more
sophisticated methods may have some tendency to
"detect" noise in the data from effects that are not,
in fact, present.
10-17
-------
30272-101
REPORT DOCUMENTATION
PAGE
l._ REPORT NO.
EPA1 560/5-86-013
3. Recip4«nfs Accession No.
4. Title and Subtitle
Underground Motor Fuel Storage Tanks: A National Survey
5. Report Oat*
May, 1986
7. Author^ Dietz, Stephen K.;a Flora, Jairus D. , Jr.;B
Strenio, Judith F;a Carmen J. Vincent* «t al.
8. Performing Organization Rept. No.
9. Performing Organization Nam* and Address
*W«tat, Inc., 1650-Research Blvd., Rockville, MD 20850
bMid»««t Research Institute, 425 Volkar Blvd., Kansas City, M8 64110
Battalia Colunbus Division, Washington Operations, 2030 M Str««t. 1W,
Washington, D.C. 20036
Washington Consulting Group, 1625 Eye Street, NW, Washington D.C. 20006
10. Proiect/Task/Work Unit No.
Task 3
11. Contract(C) or Grant(G) No.
to EPA No. 68-02-4243
and
(G) EPA No. 68-02-3938
12. Sponsoring Organization Nam* and Address
Environmental Protection Agency
Office of Toxic Substances
Exposure Evaluation Division
401 M Street, S.W.
Washington, D.C. 20460
13. Typ* of Report & Period Covered
Final Report
Feb. '84 - May '86
14.
15. Supplementary Notes
1ft. Abstract (Limit: 200 words)
A nationally representative sample of 2,812 establishments were interviewed to
determine the presence of underground motor fuel storage tanks. This sample
represented establishments in fuel-related industries (1,612), large establish-
ments in all other industries (600) , and farms (600). A total of 890 of these
establishments were found to have a total of 2,445 underground motor fuel storage
tanks. Only 19 farms with 34 tanks were found. The following national estimates
were made: there are 796,000 underground motor fuel storage tanks at 326,000
establishments in the Unisted States -- 158,000 of these are on 79,000 farms. A
subsample of 218 establishments was selected for tank tightness testing, using a
modification of a commercially available test. The method over-filled the tank
system into a standpipe, and thus detected leakage anywhere in the system of tank
vessel, pipes, lines, joints, and fittings. Among the non-farm establishments
tested, the following estimates were made: 35 percent (189,000) of tank systems
were judged to be leaking under test conditions; the average leak rate of those
systems with quantifiable leak rate, adjusted for test pressure, was 0.32 gallons
per hour; half the leaks among all systems judged to be leaking were 0.25 gallons
per hour or less.
17. Document Analysis a. Descriptors
Establishment characteristics, leaking underground tanks, survey design, tank
characteristics, tank establishments, tank tightness testing, underground motor
fuel storage tanks
b. Identifiers/Open-Ended Terms
e. COSATI Field/Group
It. Availability Statement
19. Security Class (This Report)
Unclassified
20. Security Class (This Page)
11
21. No. of Pages
Vol I: 227
Vol II: 350
22. Price
(See ANSI-Z39.18)
Set Instructions on Raven*
OPTIONAL FORM 272 (4-77)
(Formerly NTIS-35)
Department of Commerce
------- |