&EPA
United States
Environmental Protection
Agency
        The EMPACT Beaches Project:
        Results From a Study on Microbiological
        Monitoring in Recreational Waters
        August 2005
                  »

-------
                                                     EPA 600/R-04/023
                                                         August 2005
                The EMPACT Beaches Project

       Results from a Study on Microbiological Monitoring in
                        Recreational Waters
                          Larry J. Wymer
                         Kristen P. Brenner
                         John W. Martinson
                          Walter R. Stutts
                        Stephen A. Schaub*
                          Alfred P. Dufour
               U.S. Environmental Protection Agency
                Office of Research and Development
              National Exposure Research Laboratory
                       Cincinnati, OH 45268
* U.S. EPA, Office of Water, Health and Ecological Criteria Division, Washington, DC
                                                  7~s   Recycled/Recyclable
                                                  j 7  Printed with vegetable-based ink on
                                                   ^"'\  paper that contains a minimum of
                                                  4\ )  50% post-consumer fiber content
                                                  (  V  processed chlorine free
                                                  V"

-------
Disclaimer: This document has been reviewed by the U.S. Environmental Protection Agency (EPA)
and approved for publication.  Mention of trade names or commercial products does not constitute
endorsement or recommendation of their use.
                                           11

-------
                               Acknowledgements
We convey our gratitude to the  following local officials for allowing us to conduct the fieldwork
in beaches under their jurisdiction and for their enormous help in the logistics of conducting the
fieldwork: Donald Hamel, Detroit Health Department, Environmental Health Services, Detroit, MI;
Rick Amador, San Diego County Department of Environmental Health; Kenneth Wilde, Diagnostic and
Public Health Microbiology, Maryland Department of Health and Mental Hygiene, Baltimore, MD;
Dale B. Engquist, U.S. National Park Service, Indiana Dunes National Lakeshore, Porter, IN; Andrea
Rex, Environmental Quality Department, Massachusetts Water Resources Authority, Charlestown
Navy Yard, Boston, MA.

We especially thank the following two individuals who, through their extensive  experience and
knowledge, helped overcome problems associated  with logistics and site selection: Steve Weisberg,
Director, Southern California Coastal Water Research Program, Westminster, CA; Richard Whitman,
Field Station Supervisor, U.S. Geological Survey, Lake Michigan Ecological Research Station, Porter,
IN.

Other individuals that we are indebted to for their participation and contributions during the conduct
of the study are Abdel El-Shaarawi of the National Water Research Institute, Burlington, Ontario,
Canada; Richard O.  Gilbert, independent consultant, Rockville, MD; Molly Leecaster, formerly of
the Southern California Coastal Water Research Program, Westminster, CA, and currently at Idaho
National Engineering and Environmental Laboratory, Idaho Falls, ID; David F. Parkhurst, Professor,
School of Public and Environmental Affairs, Indiana University, Bloomington, IN; William F. (Rick)
Hoffman, Office of Water, Office of Science and Technology, Washington, DC; Robin Oshiro, Office
of Water, Office of Science and Technology, Washington, DC; Mimi Dannel, Office of Research and
Development, Office of Science Policy, Washington, DC; Daniel Lewis, formerly of OAO Corporation
(now Lockheed Martin Information Technology), Cincinnati, OH; Sherie Brown, Shawn C. Siefring,
and Frederick P. Williams, U.S. EPA/National Exposure Research Laboratory, Cincinnati,  OH.

The fieldwork was organized and conducted by two contractors, Battelle Research Institute and
Lockheed Martin.  We are grateful to their staff for  their skill, cooperation, and diligence in this
study. The microbiological assays of water samples during this study were performed by contract
laboratories at various sites across the nation. We thank the staffs of these laboratories for their
diligent work and extraordinary cooperation: Battelle Research Institute Laboratory, Duxbury, MA;
Battelle Research Institute Laboratory, Edgewood,  MD; City of San Diego Laboratory, San Diego,
CA; Huron Valley Laboratory, Romulus, MI; Severn Trent Laboratory, Valparaiso,  IN. Finally,
we thank Katherine Loizos of Computer Sciences Corporation for preparing this document for
publication.
                                            in

-------
IV

-------
Table of Contents
Overview of the Study Results
Factors found to influence indicator levels
Introduction
Study Design
Sites
Sampling design
Sampling schedule
Data collection
Microbiological analysis
Other data
Factors Affecting Water Quality at a Beach
Introduction
Statistical methods
Criterion variable
Design variables
Covariates
Spatial factors
The physical significance of spatial factors
Findings from the EMPACT Beaches study with regard to spatial factors
Variation in target density associated with zones
Variation in target density associated with distance along the shoreline
Sampling depth
Temporal factors
The significance of temporal factors in sample design
Findings from the EMPACT Beaches study with regard to temporal factors
Day-to-day variability
Variation between morning and afternoon results
Environmental and bather effects
Statistical evaluation
Tides
Weather
Bather density
Sampling Variance
Sources of variation
Populations, sampling, variance, and components of variation
Sampling distribution
Components of spatial variation
Components of temporal variation
1
1
3
5
5
6
7
8
8
9
13
13
13
14
14
14
15
15
17
17
17
22
24
24
24
24
28
35
36
39
39
43
45
45
45
45
46
47

-------
     Estimates of variance components for the EMPACT beaches	48
     Spatial components of variation	48
        Small-scale (replicate) variance	48
        Variance among sampling depths	49
        Variance among zones	50
        Variance among locations within zone (among transects)	50
     Temporal components of variation	52
        Hourly variation	52
        Variance among days	52
        Variance of the change over a 24-hour period	52

Designing a Beach Monitoring Program	 54

   The DQO approach	54
     Defining the boundaries of the study: the population	55
        Sampling depth	55
        Distance from shore (Depth zone)	56
        Distance along the beachfront	56
        Time	58
     Specifying tolerable limits on decision errors	58
        A rationale for setting tolerance intervals	60
     Optimizing the design for obtaining data	61
        Stratification along the shoreline	64

Summary	 68

References	 71
                                             VI

-------
                             Index of Figures
Figure 1. EMPACT study beaches


Figure 2. Beach sampling grid	
Figure 3. Cross-section of beach sampling area
Figure 4. Freshwater beach sampling schedules 	  10


Figure 5. Point-in-time geometric mean indicator density per 100 mL by zone	  18

Figure 6. Indicator density per 100 mL by zone and transect	  20

Figure 7. Geometric mean density of surface samples vs. depth samples	  23

FigureS. Time series plots of geometric mean morning (9:00 am) indicator levels _  26

Figure 9. Change in mean loglO indicator density between the morning (9:00 am )
sampling and afternoon (2:00 pm) sample	  30
Figure 10. Geometric mean indicator density for morning (9:00 am) samples and
afternoon (2:00 pm) samples, by whether afternoon was sunny (<50% cloud cover)
or not sunny	 32
Figure 11. Geometric mean (GM) indicator density by time-of-day (9:00 am - 6:00
pm) for samples collected on hourly sampling days - ratio of GM on the hour to
overall GM for the day. Average ratio for each hour is shown by the heavy line	 34

Figure 12. Geometric mean enterococci density by tide stage at marine beaches	 40

Figure 13. Geometric mean indicator density for various weather conditions 	 41
Figure 14. Geometric mean indicator organism density by number of bathers on
the beach and in the water	 44

                                     vii

-------
Figure 15. The Data Quality Objectives Process (from "Guidance for the Data
Quality Objectives Process", U.S. EPA [2000b])	 54

Figure 16. Variance of indicator density per 100 mL vs. average separation among
sampling locations	 57

Figure 17. Correspondence between health effects criteria and indicator density	 60


Figure 18. Operating characteristic curves	 65
                                      Vlll

-------
                                Index of Tables


Table 1. Number of sampling visits and total number of samples collected

Table 2. Summary of laboratory methods	
Table 3. Data Collected at each sampling visit (in addition to microbial data).	12


Table 4. Comparisons of point-in-time geometric mean indicator density by zone	17


Table 5. Percentage of total samples collected that exceed the single sample limit as
recommended in EPA guidelines [U.S. EPA, 1986]	19


Table 6. Percentage of samples collected in knee-deep water that exceeded the single sample
limit as recommended in EPA guidelines [U.S. EPA 1986] (9:00 am samples)	21


Table 7. Comparison of geometric mean densityl of indicator densities by depth of water
from which the sample was collected.	22


Table 8. Percentage of total samples collected that exceeded the single sample limit as recommended in
EPA guidelines [U.S. EPA, 1986] 	23


Table 9. Ratio to the previous day of the geometric mean (GM) of all knee- and chest-depth
samples taken at the 9:00 am visit	27


Table 10. Serial correlations between log means separated by 1,2, or 4 days 	28
Table 11. Percentages of total samples collected that exceed the single sample limit [U.S.
EPA, 1986], classified by whether the sample from the same location was above or below the
limit  1, 2, or 4 days earlier	29
Table 12. Comparison of geometric mean density of respective indicator organisms at each
study beach between morning and afternoon visits — sunny vs. cloudy days	31
Table 13. Percentage of knee-deep samples collected at 2:00 p.m. that exceeded the single
sample limit as recommended in EPA guidelines [U.S. EPA, 1986], by whether the sample
from the same location was above or below the limit at 9:00 am.	33

                                         ix

-------
Table 14. Percentage of knee-deep samples that exceeded the single sample limit [U.S. EPA, 1986],
by time-of-day. Only hourly sampling results are used. For all but Belle Isle, there were 14 hourly
sampling days and 39-42 valid knee-deep observations (for Belle Isle there were 23 or 24 observations
over 8 days).	35
Table 15. Environmental characteristics and bather densities of EMPACT
study beaches	36


Table 16. Results of regression analysis on loglO density of indicator organisms	38
Table 17.  Comparisons of geometric mean entercocci density per 100 mL at low, mid, and
high water levels for marine beaches (based on 25th and 75th percentiles of water level).	39
Table 18.  Comparisons of geometric mean indicator density per 100 mL for various
weather-related variables	42

Table 19.  Comparisons of geometric mean indicator density per 100 mL. vs number
of bathers in the water and on the beach, for the afternoon (2:00 pm) sampling visits	44


Table 20.  Standard deviation and variance component estimates for the EMPACT beaches
	49
Table 21.  Standard deviation among locations (transects) within each separate dept zone
                                                                                    51
Table 22. Variance and standard deviation of 24-hour change in loglO indicator density and
of 24-hour change in loglO indicator density when modeled as in Table 16	53
Table 23.  Sample size requirements based on estimated overall sampling variance for the
geometric mean within a single depth zone in the water	62
Table 24. Width of tolerance intervals for various sample sizes based on sample variances
within depth zone at the study beaches	63


Table 25. Comparison of sample size requirements for equivalent precision with and
without compositing.	67

-------
                        Overview of the Study Results
Quantitative guidelines with respect to indicator organism densities in recreational waters that have
been established by the  U. S. EPA have been supported by human health studies, notably those
performed in the late 40's and early 50's by the U.S. Public Health Service, and in the 70's and early
80's by the U.S. EPA. Upper limits of 35 enterococci per 100 mL in marine water and 33 enterococci
per 100 mL in freshwater were recommended for the geometric mean of at least five samples taken over
a 30-day period [U.S. EPA, 1986]. An upper limit on the geometric mean of 126 E. coll per 100 mL
was recommended for freshwater environments only, since E. coll failed to show a good relationship
to swimming-associated  illnesses in marine waters.  A recommendation was also made that, for a
designated public beach,  no single sample exceed the 75th percentile of a lognormal distribution with
these respective geometric means. Lacking site-specific log standard deviations, this implied an upper
limit on any single sample of 104 enterococci per 100 mL in marine water (log standard deviation of
0.7), and 98 enterococci or 235 E coll per 100 mL in freshwater (log standard deviation of 0.4).

Providing numerical guidelines that are supported by scientific data was a great advance.  Lacking,
however, was  scientific  data of comparable  quality to  support  monitoring schemes that would
provide the most meaningful information about whether the guidelines are being  met. The EMPACT
(Environmental Monitoring for Public Access and Community Tracking) Beaches project has attempted
to identify those characteristics of a beach environment that  have significant impact on monitoring.
This project examined five beach environments to determine the factors that most influence the
measurement of beach water quality. Two ocean beaches, an estuarine beach, a Great Lakes beach,
and a riverine beach were  selected to provide as broad a representation of beach environments as
possible. This approach was used to develop model protocols that can be applied to similar beaches.

Factors found to influence indicator levels

For each of the five study beaches, the greatest single determinant of microbial indicator level was
found to be the depth zone, or, roughly, distance from the shoreline at which the sample was collected.
Bacterial densities became substantially lower as one moved  from ankle-deep to knee-deep to chest-
deep water. This fact has important implications for sample design as well as for public health.

In contrast, this study found no  significant difference in indicator level among samples  that were
taken  at different depths below the surface, such as between those taken the standard 0.3 M from the
surface and those taken from near the bottom.  At three of the five beaches, no substantial systematic
differences were found among samples taken at different points parallel to the beachfront, which
spanned a distance of 60  M, as long as they came from water of the same depth. Minor "hot spots"
and "cold spots" were found at two of the beaches, although the differences were not dramatic and
were not consistently found at all times.

Significant declines in indicator densities from the morning to the afternoon (9:00 a.m. to 2:00 p.m.)
were observed at four of the study beaches. This effect was seen only on sunny days at one freshwater
beach, but was observed to be independent of sunshine at three others, a freshwater beach and two
marine beaches. Indicator levels  at the remaining beach, a West Coast marine environment, tended to
be very low at all times.

From  one day to the next, the geometric mean indicator density among all samples that were collected
changed by a factor of 2 (doubling  or halving) or more about half the time at  each studied beach.
                                            1

-------
There was virtually zero correlation between the indicator level on any given day and the indicator
level more than two days later.  The geometric mean of five samples taken over a 29-day period (that
is, on the same day of the week) did not accurately represent actual conditions on any given day during
the 29-day period.

Findings from this study with regard to spatial and temporal variation suggest the following implications
for a beach monitoring program:

    •   Time and location of sampling must be carefully considered.
            o  Depth zones from which samples are collected are likely to have great effect on the
               resulting estimate of indicator density.  Sampling in knee- to waist-deep water would
               seem to offer a reasonable, but still conservative, approach to estimating water quality,
               particularly  given that  health effects are  based on quality of water at waist-depth or
               deeper.
            o  Sampling at 0.3 M below the surface is justified based on exposure considerations. This
               study failed  to discern differences at lower collection depths.
            o  Sampling in the morning will likely be a conservative practice, in addition to perhaps
               being a convenient time to sample.
            o  Sampling should be performed as close as practical to the day on which a decision is to
               be made regarding beach closure or advisement.  Preferably this should be the day before,
               given current conditions of a one day turnaround for the results.

    •   A number of samples should be  collected from different points in knee- to waist-deep water.
            o  Results from the  EMPACT Beaches Study can be used as a guideline for the  initial
               determination of sample size requirements.  For the beaches in this study, sample sizes of
               from 3 to 6 would be adequate to allow for 95% certainty of detecting a 0.3 log exceedence
               from the action level for the geometric mean, equivalent to a health risk of 3 to 3.5 cases
               of HCGI per 1000 swimmers.
            o  As data are gathered, the  sampling  plan should be refined and ultimately based on a
               variance estimate that is uniquely associated with the subject beach.

    •   "Hot spots" within the bathing  area, which may be known  a priori or discovered as a result of
        sampling, should be considered  as separate strata for sampling purposes.
            o  These should be sampled independently and weighted appropriately in the final result.
            o  If indicator density within the problem area is very different from that elsewhere, sampling
               should be limited to this area. A proposed rule-of-thumb is  a two-fold difference.
            o  Lesser differences would warrant a combined, stratified estimate, weighting the "hot
               spot" in proportion to its extent relative to the rest of the beachfront.

    •   Composite sampling may be used as a cost-efficient technique, enabling better sample coverage at
        minimally increased  cost.
            o  In the initial stages of a monitoring program, composite sampling should not be used in
               order to develop data that is necessary in estimating the appropriate sampling variance for
               a particular beach.
            o  A composite sample estimates  an  arithmetic mean, which  would require adjustment in
               order to equate this to standards based on a geometric mean.
            o  If resources  are available to collect the data necessary for developing a predictive model
               for the change in indicator density over a 24 hour period, this will likely result in much-
               improved assessment of water quality, given the  1-day lag in obtaining results from
               membrane filtration assays.

-------
                                    Introduction

Protecting the health of swimmers who use surface waters for recreation has been of interest to public
health officials in the United States since  1930.  It was well recognized at that early date that human
excreta discharged to surface waters posed a health hazard to those who used the water for recreation.
Although the relationship between swimming-associated health effects and feces-contaminated water
used for swimming had not been defined, microbial limits based on coliform bacteria were used in
many states, particularly when there was physical evidence of sewage contamination.  The limiting
values selected by responsible authorities were based more on attainment rather than on risk of illness.
Thus, there was little uniformity among states regarding what level of coliforms constitute waters safe
for swimming. Several states chose 1,000 coliforms per 100 ml as a measure of good quality water,
but there was not much uniformity among states regarding what level of coliforms was a safe level.
There was, however, a general understanding that fecal contamination of surface water posed a risk to
those exposed to  the water, and that the risk might be limited by setting a level of fecal contamination
above which exposure would be unacceptable. The manner in which water samples were taken, the
frequency of sampling, and the number of samples were usually not described in the early literature.
It is  probable, however, that multiple samples  from several points along a beach were not taken.
Early monitoring schemes seldom involved several samples.  Even now the Environmental Protection
Agency (EPA) recommends at least 5 samples taken over a 30-day period, a sampling scheme that
was first suggested in 1968. That number of samples will not characterize a body of water as to its
true quality.  Similarly, weekly sampling, no matter how accurate, is an unlikely scheme to alert risk
managers of poor quality water between sample dates. Taking one daily sample might solve the latter
problem, but still would not provide an accurate characterization of the water.

In the late 1940's and early 1950's a series of studies were carried out by the U.S. Public Health Service
to determine the  relationship between swimming-associated health effects and bathing water quality.
The results of those studies were reported by Stevenson (1953).  This data was used by the National
Technical  Advisory Committee (NTAC), commissioned  by the Federal Water Pollution Control
Administration in 1968 to establish guidelines  for recreational waters.  The NTAC recommended
that a geometric  mean of 200 fecal coliforms per 100 ml from  5 samples  collected over a 30-day
period be used as an upper limit which was believed to be protective of public health for swimmers
[NTAC, 1968].  In the early 1970's the U.S. Environmental Protection Agency initiated  a series of
epidemiological studies to determine the relationship between swimming-associated illness and water
quality as  measured with multiple microbial indicators that are commonly  found in feces [Cabelli,
1983; Dufour, 1984]. These studies showed that, of all the microbes examined, enterococci had the
best relationship  to swimmers' illness in marine waters.  Thus, as the water quality became poorer,
i.e., the enterococci density increased, the swimming-associated illness rate increased. In fresh waters
E. coli and enterococci were found to be effective for relating water quality to swimmer illness.  EPA
recommended an upper limit on geometric mean enterococci density of 35 per 100 ml  for marine
waters and 33 per 100 ml for fresh waters  [U.S. EPA, 1986]. The means were to be calculated from at
least 5 samples taken over a 30-day period. E. coli was not effective in marine waters, but did show
a good relationship to swimmer-associated illness in freshwater environments.  An upper limit of
126 E. coli per 100 ml was recommended for fresh waters [U.S. EPA, 1986]. These epidemiological
studies conducted by the U.S.  Public Health Service and the U.S. Environmental Protection Agency
resulted in guidelines that limited the number of fecal indicator bacteria in a water body and, thereby,
the number of swimming-associated illnesses to a tolerated level.

Providing  numerical guidelines that are supported by scientific data was a great advance; however,
these guidelines cannot stand alone.  Monitoring schemes that will provide meaningful information
                                            3

-------
about whether the guidelines are being met must also be available. The manner in which monitoring
is carried out at any particular beach is not likely to be the same for a marine beach with tidal cycles as
for a freshwater beach, where other factors will significantly affect the level of indicator bacteria.

Since microbes almost universally have been recommended as the measure of water quality for
recreational waters  some  of the  characteristics that limit their usefulness should be addressed.
Bacterial indicators frequently die  off more rapidly than some of the pathogens found in feces [Fattal
etal., 1983]. When this happens, potential pathogens associated with feces might be present and pose
a health hazard. Another issue with indicator microbes is that they usually take up to 24 hours to grow
to visible colonies on solid media or turbidity in liquid media. This lengthy time lag creates a situation
where exposed individuals are informed of risk of illness long after the fact [Bartram, 1999].

Physical factors may affect the measurement of water quality in ways that are not controllable.  For
instance, microbial indicators are affected by solar radiation, which cause many indicators to die off
at variable rates [Fujioka and Narikawa, 1982], and neither the amount nor the intensity of sunlight
is constant.  These have a tendency to change from hour to hour and day to day.  Rainfall is another
factor that significantly influences the measurement of indicator microbes in beach waters [Olivieri et
al, 1977], but it is more predictable than other variables.  Wind direction may also affect the quality
of beach waters either by driving contamination toward  the beach or away from the beach. Swimmer
density can influence the quality of beach water as well.  Swimmers shed high numbers of indicator
bacteria from their bodies.  Tub experiments have shown the millions of indicator bacteria are shed in
very short periods of time [Smith and Dufour,  1993]. In marine waters, tides significantly affect the
measurement of indicator bacteria, in addition to other factors mentioned [Cabelli et al., 1974].

It is difficult to determine which of the above factors has the greatest effect on the monitoring of beach
waters.  It is likely that each beach will have  its own characteristics based on predominant factors
active at any one time. The EMPACT (Environmental Monitoring for Public Access and Community
Tracking) Beaches project has attempted to define  which characteristics  are most significant with
regard to monitoring approaches.  This  project examined five beach environments to determine the
factors that most influence the measurement of beach water quality. Two ocean beaches, an estuarine
beach, a Great Lakes beach,  and a riverine beach were  selected to provide as broad a representation
of beach environments as possible. This approach was used to develop model protocols that can be
applied to similar beaches.

-------
                                   Study Design
Sites
Beaches selected for this study are all located in or near partner cities in the Environmental Monitoring
for Public Access and Community Tracking (EMPACT) grants program.  The study was jointly
conducted by the EPA Office of Water, Office of Science and Technology, and Office of Research
and Development with the goal of helping communities collect, manage, and present environmental
information to their residents. The following were the participating sites in this study (see Figure 1):

   •   West Beach, Indiana Dunes National Lakeshore, Ogden Dunes, Indiana, a freshwater beach
       on the shores of Lake Michigan. The EMPACT partner city is Gary.
   •   Belle  Isle Park, Detroit, Michigan, a freshwater beach on the Detroit River between Lake St.
       Clair and Lake Erie.
   •   Wollaston Beach, Quincy, Massachusetts, a marine beach in Quincy  Bay.  The EMPACT
       partner city is Boston.
   •   Imperial Beach,  Imperial Beach,  California, a marine beach on the  Pacific Ocean.  The
       EMPACT partner city is San Diego.
   •   Miami Beach Park, Bowley's Quarters, Maryland, an estuarine beach on Chesapeake Bay
       near Middle River. The EMPACT partner city is Baltimore.
                          Figure 1. EMPACT study beaches.

-------
Sampling design
Samples were collected from each of nine locations within the limits of the beach's bathing area, each
location being determined by an associated transect and zone, as illustrated in Figure 2.  A transect
is denned as an imaginary line through a fixed point on the beach and forming  a right angle with
the  shoreline.  A zone is defined as a contour line  of equal water depth.  As illustrated in Figure 2,
each sampling location is defined by the intersection of transect and zone on a grid comprising three
transects and three zones projected on the water's surface. A random point along the shoreline within
the  recognized beach area was selected to define the first transect (the leftmost transect in Figure 2).
The middle transect was, then, determined as the parallel to the first transect at a distance of 20 meters,
and the remaining transect, as the parallel to the middle transect at an additional distance of 20 meters,
or 40 meters from the first transect.
                             Figure 2.  Beach sampling grid.
    Zone 3
    Zone 2
    Zone 1
                             -20m-
              Transect 1
                -20 nv-
Transect 2

  Beach
Transect 3
Intersections of sampling zones with these transects were determined by the depth of the water at
the time of sampling.  Zone 1 was determined at each transect as the point at which the water first
attained a depth of 0.15 meters ("ankle-deep"); zones 2 and 3 were determined as those points along
each of the transects where the water first attained depths of 0.5 meters ("knee-deep") and 1.3 meters
("chest-deep"), respectively. Buoys demarcating the swimming area in Belle Isle were so located that
the water within this boundary never attained a depth of 1.3 meters; at this beach, therefore, the buoys
themselves, at which the water was approximately 1 meter deep or about waist level, demarcated zone
3. The bathing area at Miami Beach Park was similarly restricted, the water depth along zone 3 being
about 1.1 meter.  Low tide at Wollaston Beach resulted in shallows extending quite far out from the
beach. A 100 meter limit was established on the distance  an individual would go to collect samples,
resulting in waist-deep water at zone 3 at these times.

-------
Water samples were collected from a depth of 0.075 meters below the surface in ankle deep water and
0.3 meters below the surface in knee-deep and chest-deep water.  These sampling depths are shown
schematically in Figure 3. Note that the actual sample locations (1-9 in Figure 2) may vary from one
visit to the next, in terms of both absolute geographical location and relative distance from shore.
in accordance with variations in water level and/or changing contour of the floor, particularly at the
marine beaches.

                    Figure 3. Cross-section of beach sampling area.
               = sampling location
Water samples  were collected in 500-mL pre-sterilized, polypropylene bottles.   Care was taken
in collecting samples near the bottom to avoid introducing additional sand or other solids into the
samples. Following collection, samples were maintained on ice until analysis.

Sampling schedule

On each day, one of four sampling plans, referred to as  "basic", "hourly", "replicate", and "depth"
sampling, was followed according to the schedules shown in Figures 4a and 4b. The actual schedule
at West  Beach most closely followed the planned schedule, which was to have been identical at all
five beaches.  It was anticipated, and was, in fact, the case in all cities, that events such as inclement
weather and rough water would force revision of the schedules. At Belle Isle and Imperial Beach,
major activities necessitated changes in schedule and cancellation  of certain sampling visits. The
total number of sampling visits and total samples collected under each of these plans is summarized
in Table 1 for each beach. Following are descriptions of each sampling plan.

    Basic sampling
    On each basic sampling day, samples were collected  on two separate visits, at 9:00 a.m. and
    2:00 p.m. A single sample was collected at each of the nine sampling locations (1 through 9
    in Figure 2) on each  sampling visit.

    Hourly sampling
    On each hourly sampling day, sampling was performed on ten separate occasions, hourly on
    the hour, the first set being collected at 9:00 a.m. and the last at 6:00 p.m. A single sample was
    collected at each of the nine sampling locations on each occasion.

-------
   Replicate sampling
    On each replicate sampling day, samples were collected on two separate visits, at 9:00 a.m.
    and 2:00 p.m.  On each sampling visit, ten replicate water samples were collected at the
    central sampling location (location 5 in Figure 2), and two replicate samples were collected
    at each of the eight remaining locations.  Initially, because it was anticipated that its location
    on a river would make the collection of true replicates difficult, Belle  Isle was to be an
    exception to the general replicate sampling scheme, with two samples being collected at all
    nine sampling points.  However, after the first two replicate sampling visits, ten replicates
    were taken from the central location at this beach as well, when it became apparent that water
    flow was negligible within the protected confines of the bathing area.

   Depth sampling
    On each depth sampling day, samples were collected on two separate visits, at 9:00 a.m. and
    2:00 p.m., with samples being collected at each of the nine sampling locations in Figure 2, the
    same as on a basic sampling visit. In addition to the standard sampling depths (0.3  m for knee
    and chest deep water), a sample was collected at a depth of 0.425 meters (0.075 meters, or 3
    inches, from the bottom) at each of the three sampling points in the knee-deep zone. At each
    of the three chest deep zone sampling points, two additional samples were collected, one from
    mid-depth (normally, 0.65 meters from the surface) and another at a depth of 0.075 meters, or
    3 inches, from the bottom. Thus, a total of nine additional samples were collected beyond the
    nine samples collected in the basic sampling plan.

     Table 1.     Number of sampling visits and total number of samples collected.
Location
West Beach
Belle Isle
Wollaston
Imperial
Beach
Miami Beach
Park
Basic sampling
Visits
69
75
71
68
72
Samples
610
671
638
612
643
Hourly
Visits
139
80
138
140
140
Samples
1162
705
1242
1254
1257
Replicate
Visits
16
16
16
16
16
Samples
416
334
407
416
416
Depth
Visits
8
8
8
8
8
Samples
144
138
144
144
144
Data collection

Microbiological analysis

Freshwater samples, collected at Belle Isle and West Beach were assayed for Escherichia coll by the
modified mTEC agar membrane filter method [U.S. EPA 2000a], while the marine and estuarine water
samples, collected at Imperial Beach,  Wollaston Beach, and Miami Beach Park, were analyzed for
enterococci by the mEI agar membrane filter method (Method 1600) [Messer and Dufour, 1998 U.S.
EPA1997b,U.S.EPA2000a]. The methods are summarized in Table 2. Using volumes of 100,10,and
1 mL of each sample, nitrations were begun within six hours of collection and completed within eight
hours of collection [Code of Federal Regulations 2000]. When concentrations were very high on the
previous day's sample(s), 0.1 mL volumes of the water samples were also analyzed.  Bacterial counts,

-------
volumes, and other critical information, such as analysis start time and incubation start and completion
times, were recorded on specially-designed data sheets, faxed daily to EPA (NERL-Cincinnati), and
entered into the EMPACT database for subsequent audits of data quality and data analysis.

                         Table 2. Summary of laboratory methods.
Method
Used
E. coli
(EPA 1603)
Enterococci
(EPA 1600)
Medium
Used
mTEC2
agar
mEI
agar
Incubation Time and
Temperatures (°C)
2 hours @ 35 + 0.5
20-22 hours @ 44.5 + 0.2
24 + 2 hours @ 41 +0.5
Detection Limits
(Colonies
Per Plate)
1-200
1-200
Ideal # of
Colonies Per
Membrane
20-80
20-60
The  general microbiology  laboratory quality control  procedures  used in this study have  been
previously described [American Public Health Assoc. 1998, Bordner et al. 1978, U.S. EPA 1997a,
U.S. EPA 2000a]. In addition, each preparation of modified mTEC agar and mEI agar was pre-tested
for performance with known positive (target) and negative (non-target) cultures, and membrane filter,
phosphate-buffered dilution water, and agar controls were performed with each set of samples that
was analyzed (For example, all of the samples taken at the nine locations on the three transects for a
specific time and date would be considered a set of samples.)

Other data

Other data that were collected at each sampling visit are shown  in Table 3.  Hourly and high and
low water tidal data were obtained from the National Ocean  Service of the National Oceanic and
Atmospheric Administration for the reference station nearest to each respective study beach.

-------
               Figure 4a.  Freshwater beach sampling schedules.
West Beach
 Sun.  Man.  TUB.  Wed  Thur.  Fri.   Sat.     Sun.  Men.  TUB.  Wed Tnur.  Fri.   Sat.
Belle Isle
 Sun.  Hon. TUB.  Wed  Thur.  Fri.   Sat.     Sun.  MOIL  TUB.  Wed Tnur.  Fri.   Sat.
   Basic sampling day
   Hourly sampling day (9:00 a_m-6:00 p.m.)
  ] No sampling visits on this day
   I Replicate s ampling day
 [3 Depth s ampling day
AH Only 9:00 am samples were collected on this day
PH Only 2:00 pm samples were collected on this day
                                         10

-------
      Figure 4b. Marine and estuarine beach sampling schedules.
Wollastoii Beach
 Sun.   Man.  Tues.  Wed  Thur.  Fti.   Sat.     Sun.  Man. Tues. Wed  Thur.  Fri.   Sat.
Imperial Beach
 Sun.   Men.  Tues.  Wed  Thur.  Fri.   Sat.
                                           Sun.  Mon. Tues. Wed  Thur.  Fri.   Sat.
Miami Beach I';irk
 Sun.   Man.  Tues.  Wed Thur.  Fri.   Sat.
                                           Sun.  Mon. Tues. Wed  Thur.  Fri.   Sat.
  1 Basic sampling day
   Hourly sampling day (9:00 a_m.-6:00 p.m.)
  ] No sampling visits on this day
 _J Replicate sampling day
 [3 Depth sampling day
AM Only 9:00 am samples TOT e collected on this day
PM Only 2:00 pm samples WET e collected on this day
                                        11

-------
Table 3.  Data collected at each sampling visit (in addition to microbial data).
Measurement
Date and time
Air temperature
Water temperature
Cloud cover
Rainfall since last
visit
Wind speed
Wind direction
Current direction
Wave height
Bather density in
the water
Bather density on
the beach
Boats
Animals
Debris
Total Suspended
Solids (TSS)
pH
Turbidity
Description & units of measurement
Date and time of day.
Measured by thermometer at a fixed location every
visit.
Measured by thermometer at the central sampling
location at appropriate depth for thermometer.
Sunny, Mostly Sunny (20-50% cloud cover),
Cloudy (50-70% cover), Mostly Cloudy (70-99%
cover), Overcast
Measured by rain gauge; collected each sampling
visit.
Sustained speed measured by wind gauge.
Compass direction to nearest semi-quadrant leeward
on wind gauge.
Described in relation to shoreline facing out.
Meter stick measurement at central sampling point.
Number of bathers in the water, within the sampling
area.
Number of bathers on beach, within outer transects.
Number of boats in the water within 500 m of
sampling area.
Animals within 20M of the sampling area in the
water or on the beach; also includes number of fowl
or other birds in the air near the sampling area.
Description of debris floating in the water or
washed on shore within the sampling area.
Measured from the sample taken at the central
sampling location, per Standard Methods (APHA,
1998).
Each sample measured after microbiological
analysis, per Standard Methods (APHA, 1998).
Each sample measured by nephlometer after
microbiological analysis processing, per Standard
Methods (APHA, 1998).
Units

°C
°C
S, MS, C, MC, O
Inches
Miles per hour
N, ME, E, etc.
Descriptive:
(onshore, right, etc.)
Meters
Categorical: <20, 20-100,
100-200, >200
Categorical: <20, 20-100,
100-200, >200
Categorical:
<20, 20-100, 100-200, >200
Type & number of animals
Categorical; "None," "Very
Little," "Little," "Lots;" and
description
mg/L
pH units
NTU
                                 12

-------
               Factors Affecting Water Quality at a Beach


Introduction

In this section, we will  discuss various factors that were seen to have some association with the
indicator density that was observed in our samples.  This will provide answers to questions such as:

    •  At what locations in the water, if any, does one tend to see higher or lower levels of indicator
       organisms?
    •  Are any temporal cycles evident, such as a time of the day or a day of the week when indicator
       levels tend to be higher or lower than they are at other times?
    •  Can we expect generally higher or lower sample results under certain weather conditions?
       What about tidal influences at marine beaches?

These refer to systematic variability, as opposed to random variability, which will be discussed in the
next section. Systematic variability implies predictable differences, albeit with a degree of uncertainty
as indicated in the phrasing of these questions. The variability that is left unexplained, the uncertainty
in our prediction, is random variability.

Knowledge about systematic variability is important for two reasons:

    •  Systematic differences that are found to exist may be exploited in the sample design, suggesting
       where and when to take samples.
    •  Because results from sampling are not available until the next day, a predictive model must
       be used to assess the current "state of the beach" (Of course, one possible model is to assume
       that nothing has changed). Even if results could be obtained instantaneously, it would still be
       desirable to have some lead-time for informing the public of the results.

Results from each of the five study beaches  are considered individually.  Any  observed spatial,
temporal, and environmental relationships with indicator density strictly apply only to that beach.
While  we have found similarities among the study  beaches with regard to certain parameters, this
should not be construed to mean that these relationships would apply to all beaches in general. Nor
should relationships found to exist at a particular study beach be assumed to apply to a general class
of which  it is representative, such as an estuarine, Great Lakes, or west coast marine beach.  Certainly,
however, a relationship found to exist at any of these study beaches may apply to other beaches, and
some of these characteristics may be true for many or most of a given class of beach.  In the course
of a regular beach monitoring program ample data can be generated to enable one to verify these
relationships and even to discover factors that are unique in one's own environment.

Statistical methods

A distinction is made between design variables and covariates. Design variables are those factors
over which we have control and were specifically accounted for in the design of the study.  Samples
were specifically collected  in ankle-, knee-, and chest-deep water, thus, "zone" is a design variable.
Likewise, three fixed transects were employed to define locations at each depth zone where samples
were to be taken, so that the variable of "transect", position along the shoreline, is also designed into
the study. Another design  variable is time-of-day, mainly being either 9:00 a.m. or 2:00 p.m. on a
given day, but also hourly from 9:00 a.m. to 6:00 p.m. on selected days. Depth below the surface from
                                            13

-------
which the sample was collected is also a design variable, since we occasionally sampled from depths
below the standard 0.3 meters from the surface.

"Covariates" include all factors that were outside of our control which were not a part of our sampling
plan, such as rainfall, cloud cover, tidal movements, currents, and bather load.  In general, these
constitute a "messier" set of observations than do design variables. We may not be able to attribute an
effect due to one of these variables  at a given beach simply because there was too little variability in
that factor at that beach. Such was the case at Imperial Beach, for example, where rainfall occurred
only twice during the course of our study.  In addition, many of these covariates may be interrelated
("co-linear" or "confounded" variables), a potential example being temperature, cloud cover, prior
rainfall, and bather load. Are the increased counts due to the greater number of bathers at the beach
or are they due to the high temperatures that bring these bathers  out?  Do we fail to see an increase
in indicator density with greater numbers of bathers simply because there are more bathers on sunny
days when, at the same time, the sunlight increases the rate of bacteria die-off?

Criterion variable

Log10 indicator density per 100 mL is used as the criterion variable (the "dependent" or "response"
variable) in the analyses that follow, thus having a direct correspondence with models that relate
mean Iog10 indicator density to swimming-related health effects [Cabelli et a/., 1982; Cabelli, 1983;
Dufour, 1984].  Existing U.S. EPA guidelines [1986], based on these models, specify limits for the
geometric mean - the antilog of the mean Iog10 density.  The implications of these health effect models
with respect to a sampling will be discussed  later, in the section Designing a  Beach Monitoring
Program.
                                    Single sample limits

  Another criterion recommended by the EPA [U.S. EPA, 1986] is the single-sample limit.  Over
  1500 samples were collected at each of the study beaches (see Table  1), giving us a multitude
  of "case studies" for evaluating the performance of the single-sample limit vis-a-vis any given
  factor.  These case studies will be presented in parallel to the analyses on log means in a boxed
  format such as this.
Design variables

Relationships between Iog10 indicator densities per 100 mL and design variables were investigated
via an analysis of variance (ANOVA) model. Factors used in the AN OVA were zone (ankle-, knee-,
and chest-deep), transect (labeled 1-3, as in Figure 2), time-of-day (9:00 a.m. through 6:00 p.m.), and
weekend (Friday through Sunday).

Covariates

The initial analysis of the dependencies of Iog10 indicator density per 100 mL on the ancillary variables
recorded in this study (Table 3) was conducted via a stepwise regression.  Because these  variables,
such as air temperature, cloud cover, etc., apply to the  entire sampling visit, the dependent variable
is the average log density for the respective visit.  This is calculated as a weighted average, by first
averaging replicate observations, if any, within a location, then averaging over transects within a zone,
and finally averaging over zones. Only samples taken from the usual sampling depth of 0.3 M below
the surface are used in this average.  So that whatever conditions happen to exist on hourly sampling
                                             14

-------
days do not disproportionately influence the results, only data from the 9:00 a.m. and 2:00 p.m. visits
are used.

Some ancillary data were transformed before being used as candidate covariates. In some cases, a
single variable was expressed in more than one form in the search for best predictor.  Many were
expressed as dummy variables, that is, variables that can take on only two values, zero  and one,
where one indicates the presence of some condition and zero, the absence of that condition. For
example, cloud cover determined the value of a dummy variable, "Sunny?", which took on the value
1 when cloud cover was coded as "Sunny" or "Mostly Sunny", and 0 when cloud cover was coded as
"Cloudy", "Mostly Cloudy", or "Overcast." Some of the variations on covariate that were considered
are:

    •  Temperature (air and water) was used in °C, as originally measured.  In addition, this was
       expressed as a high-temperature  dummy variable (greater than the median for the respective
       time-of-day) and as 1st quartile and 4th quartile dummy variables (among lowest/highest 25%
       for the respective time-of-day).
    •  Rainfall was used in inches, as  originally measured, and in the form of dummy variables
       indicating whether it had rained in the past 24 hours or in the past 3 days.
    •  Wind speed and wind direction were combined to produce a wind vector variable (the
       component of wind velocity that is onshore, which is negative if the wind is blowing away
       from the shore).  Also, dummy variables indicate onshore wind and offshore wind.
    •  Wave height was used in meters, as originally measured, and as a "high wave" dummy
       variable.
    •  Bather densities (on the beach/in the water) were considered as dummy variables (l=less than
       20, 0=20 or more).  They were also  combined in order to consider cases where there were
       more than 20 bathers both on the beach and in the water. (Note: Categories of "100-200" and
       "over 200" were too infrequently observed to be considered.)

One covariate that was included in all models was the previous day's Iog10 indicator density at the
same time-of-day.  This was either considered  as an independent variable or subtracted  from the
current day's Iog10 indicator density to create a dependent variable equal  to the change in log density.

Several regression model-building techniques were attempted, including stepwise and best R2 searches,
along with considerable trial-and-error.   In the  end, however, it was a matter of judgment  among
several competing models, each of which was as good as the other in terms of degree of complexity
and the percentage of variation explained (R2). A model for a given beach that is presented here is only
one of several possible models, any of which may be a reasonable alternative.  In general, however,
these models do tend to be similar in that some form of a major effect is common to all models with
only minor contributions being made by disparate effects. For example, rainfall and tidal stage, in one
form or another, may be common to all the models, while some models may include a bather density
effect, others, a temperature effect.

Spatial factors

The physical significance of spatial factors

We defined the physical location from which a sample  is collected in terms of transect, zone, and
sampling depth.

                                            15

-------
       Transect - an imaginary line extending at a right angle to the beach's orientation from
       a given point along the shoreline, and the water column below it.  Specifying transect
       fixes a location with respect to its distance along the shoreline.

       Zone - the water column  above a contour line on the floor of the bathing area,
       corresponding to constant water depth, that is, ankle-deep, knee-deep, or chest-deep
       water. Where the water depth is constant or decreases as one moves away from shore,
       as when the floor is level or undulates, our zones refer to the first occurrence of the
       specified depth. Roughly, "zone" means "distance from shore", at least in the sense
       that going from ankle-deep to chest-deep water takes one farther  from shore.   The
       actual distance from shore, however, may vary between two locations that are in the
       same zone.

       Sampling depth - the depth below the surface of the water from which a sample is
       collected.

For this study, samples were collected along parallel transects located 20 meters apart, each transect
being defined by two points on the shore that were fixed for the duration of the study. Samples were
collected in zones described as "ankle-deep", "knee-deep", and "chest-deep" water, but in  actuality
fixed by measurement to determine where the mean water depth was 0.15, 0.5, and 1.3 M, respectively.
Samples were collected from approximately 0.3 M below the surface (0.075 M in  ankle-deep water),
and, on certain days, additional samples were collected at greater depths.

Barring some rather unusual beach  topography, there are several  common factors that  ought to
differentially affect water quality among depth zones in qualitatively the same way at practically any
beach:

    •  Contamination on the shore or in the swash zone ought to have, if anything, a greater impact
       on water quality in ankle-deep water than in chest-deep water.
    •  The shallower the water, the more the water quality at or near the surface will be affected by
       any disturbance of the sediment, such as might be caused by bathers or wave action.
    •  Indicator organisms that come directly from bathers, not from sewage contamination, should
       be more common in shallower water due to children and a higher bather density  in general.
    •  Deeper water, on the other hand, may be exposed to any offshore point source of contamination
       to a greater degree than shallower water.

In contrast, there are few, if any, such generalities one can make with regard to  transects-different
locations along a shoreline.  Currents along or near the shoreline  may have negative or positive
effects on contamination levels downstream, and may have significant impact  at some beaches and be
negligible at others. Their impact may vary even at the same beach.  A physical barrier in or near the
bathing area, such as a sandbar, reef, or pier, may effect a gradient in the level of contamination parallel
to the shoreline at some beaches, but such effects will be absent at others.  Lacking such  generalities,
systematic variability along the shoreline must be evaluated with respect to a specific beach.
                                             16

-------
Findings from the EMPACT Beaches study with regard to spatial factors

Major findings on spatial variation are:

    •  In every case, the zone from which the sample was collected was found to have the greatest
       predictable impact on microbial indicator densities of all factors investigated in this study,
       spatial or temporal. Bacterial densities become progressively lower as one moves from ankle-
       deep to knee-deep to chest-deep water.
    •  Two of the study beaches, Belle Isle and Miami Beach Park, exhibited some form of systematic
       spatial variation that was not adequately accounted for by zones alone.  It may or may not be
       a coincidence that both of these beaches are associated with river systems.

Variation in target density associated with zones

For each of the five study beaches, the greatest single determinant of microbial indicator level is the
zone, or, roughly,  distance from the shoreline at which the sample was collected. Bacterial densities
become substantially lower as one moves from ankle-deep to knee-deep to chest-deep water.  Figure
5 and Table 4 illustrate the relative magnitudes of measurements taken 0.3 meter (0.075 M for ankle-
deep water) from the surface in these respective depth zones during the 9:00 a.m. and 2:00 p.m. visit
for each beach. Although the  effect appears to be relatively small at Imperial Beach, even here this
was the single most important influence on sample indicator density.

               Table 4.  Comparisons of point-in-time geometric mean indicator
                        density by zone.
Location
West Beach
Belle Isle
Wollaston
Imperial Beach
Miami Beach Park
9:00 a.m.
Ankle
34
2113
43
6
426
Knee
19
487
24
5
221
Chest
14
11
12
4
15
2:00 p.m.
Ankle
41
754
36
6
231
Knee
19
224
12
4
67
Chest
9
4
7
3
4
The  relationship between distance from shore and indicator density is important in terms of its
implication for sample design.  First, if the bather population exposed to contamination in one zone
essentially differs from those exposed to the contamination in another zone, then an overall estimate
that  combines indicator  densities from both zones may not adequately reflect exposure for either
population.  Given that the usual route of exposure for gastrointestinal effects is through ingestion,
an example of two different bather populations would be small children, who are likely to be exposed
to contamination only in shallow water, and adults and older children, who are most likely to be
exposed only in deeper waters. Second, if combining indicator densities from different depth zones
is appropriate for a given bather population, then the  sampling scheme should take  these known
systematic differences between zones into account.

Variation in target density associated with distance along the shoreline

In contrast to  indicator densities from samples taken at different distances out from the
shoreline,  samples that were collected in water of the same depth but at different locations
relative to the beach front (that is, on different transects) generally are not that different
                                            17

-------
from one another.  This is evident in Figure 6, which shows, by depth zone for each of the
study beaches, the geometric mean densities along each transect at 9:00 a.m. and 2:00 p.m.
Any systematic differences that may exist along the parallel-to-shore dimension are minor,
particularly in comparison to the obvious differences between depth zones, as seen previously
in Figure 5.
        Figure 5. Point-in-time geometric mean indicator density per 100 mL by zone.
     J
      1
     I
      o
      —•
      a
      u
     =3
      c
        10,000
         1,000
          100
10
                          Morning
                                                                    J?
                                                 Afternoon
There are, however, two instances in which we note  a statistically significant "transect effect"
- a persistent  difference in microbial  density along one transect relative to another.  The more
pronounced of these is at Belle Isle, where E.  coll densities in knee-deep water are consistently
lower along the leftmost transect  (Transect I  in Figure 6),  than  in  the  other two  knee-deep
sampling locations. At Miami Beach Park, where the other transect-related difference is indicated,
the situation is not so  clear and the relationship is comparatively weak.  Furthermore, analysis of
variance  reveals an interaction  between sampling location and  time-of-day, indicating that  any
such difference varies  between the 9:00 a.m. and 2:00 p.m. visits. As seen in Figure 6, values in
ankle-deep water were somewhat higher along  one transect, the leftmost transect in the morning,
but the rightmost transect in the afternoon.  We can also discern an increasing trend, particularly
in the morning, as one moves left to right (i.e., from Transect 3 to Transect I) in chest-deep water.
                                           18

-------
Whether these observed differences in indicator levels among transects have any practical significance
can be determined only by a detailed sanitary survey of these respective beaches.  It may be noted,
however, that at Belle Isle, where we see the only truly substantial transect effect, a drainage ditch
empties very near the vicinity of the rightmost transect from an area adjacent to a water park  and
bathhouse. This might be hypothesized as a point source capable of affecting
Single sample limits
and depth zones
Zones ("distance from shore") from which to collect samples will
designing a monitoring program


be a critical consideration in
How far from shore should one go? Two important points:
• If zones are not accounted for, it may be impossible to compare monitoring results from two
different days, or even to interpret the results of a single sample.
• Our current knowledge of human health effects is based on studies that have measured
indicator levels in approximately
waist-deep water.
Among the five study beaches, those samples taken in ankle-depth
exceed single-sample criteria [U.S. EPA, 1986],
water were the most likely to
and, except at Imperial Beach where only one
sample failed the criterion, samples taken in chest-deep water were

Table 5.
least likely to do
so.
Percentage of total samples collected that exceeded
the single sample limit as recommended in EPA













Location

West Beach
Belle Isle
Wollaston Beach
Imperial Beach
Miami Beach Park

guidelines [U.S.
EPA, 1986]

% of samples exceeding the single sample limit:
Number of exceedences / total number of samples
(total number of samples in parentheses)
Ankle
21.9 (183)
97.0 (167)
24.0 (183)
1.6 (183)
80.1 (186)

Knee
12.0 (183)
88.5 (174)
14.2 (183)
0.0 (183)
61.2 (185)

Chest
9.3 (172)
2.3 (174)
6.6 (183)
0.5 (183)
12.9 (186)











contamination levels in a manner consistent with observations.  Mainly, we might see higher levels
in the knee depth zone at the rightmost and middle transects, which are upstream in this riverine
environment, simply because the plume is more concentrated at these points.  Other features of the
observed pattern at Belle Isle are consistent with this hypothesis - particularly the elevated microbial
levels in chest-deep water at the leftmost transect (downstream from the ditch), and in ankle-deep
water near the outlet of the ditch at transects 2 and 3 (although this is clear only in the afternoon).
                                             19

-------
             Figure 6. Indicator density per 100 mL by zone and transect.
                   West Beach
                               2:00 p.m.
             Belle Isle
                                                 1 II.IKKI
                                                           9; 00a.m.
                                                                            2:00 p.m.
                VVollaston Beach
         Imperial Beach
                                               §
                                               iu
                                               s.
                                               *£
                                               I
                                               5
                                               _
                               2:00 p.r
                Miami Beach Park
    1,000
          Legend:
Transect 1   Bs Transect 2    Transect 3
                               2:00 p.m.
The Public Health Laboratory Service  (PHLS) Water Surveillance team (PHLS, 1995) examined
multiple  parameters  in  eight inland recreational waters.   Water samples  were collected  at  10
predetermined sites on each water body  on four consecutive weeks. Unlike the results in the current
study, the PHLS study results showed significant differences between results from the ten sites even
when the sites were very near each other.  The sampling sites in the PHLS studies were, in general,
further apart than the 20 meter distances in our study and, in addition, some of the sampling sites
were at the inlet and outlet points of the lakes. However, on two of the lakes, six of the sampling
points were in a grid one meter apart and the results from these points showed differences greater than
random variation.  The differences between our data and the PHLS data probably indicate that transect
parameter cannot be generalized from one location to another.
                                            20

-------
                       Single sample limits and shoreline location

We offer two suggestions with regard to where to sample along the beach.  The first is obvious.
The other may entail extra effort (and funding) to implement.

•     Naturally, we want to sample in areas where bathers tend to congregate.
•     Other than this, "hot spots" that deserve special consideration may exist. These may only be
      determined by knowledge of peculiarities at a particular beach or by exploratory sampling.

Consider the exceedences  found in the five study beaches among samples taken in knee-deep
water, again based on the 1986 single-sample criteria [U.S. EPA, 1986]. We chose the knee-depth
zone since this is likely to be an area that is sampled and yields a fair number of exceedences
among our study beaches (with the exception of Imperial Beach, where few exceedences were
seen among any of the samples).

                     Table 6.   Percentage of samples collected in knee-deep
                               water that exceeded the single sample limit as
                               recommended in EPA  guidelines  [U.S. EPA,
                               1986] (9:00 a.m. samples)
Location
Belle Isle
West Beach
Wollaston Beach
Imperial Beach
Miami Beach Pk.
Belle Isle
p.m. samples
% of samples exceeding the single sample
limit:
Total # of exceedences/total # of samples
(total number of samples in parentheses)
Transect 1
81.0 (58)
11.5 (61)
14.8 (61)
0.0 (61)
61.3 (62)
54.4 (57)
Transect 2
93.1 (58)
13.1 (61)
11.5 (61)
0.0 (61)
64.5 (62)
73.7 (57)
Transect 3
91.4 (58)
11.5 (61)
16.4 (61)
0.0 (61)
57.4 (61)
82.5 (57)
For the most part, there are no meaningful differences in the rate of exceedence among the three
transects.  A possible exception is Belle Isle, where fewer exceedences were observed at the 1st
(leftmost) transect. While the p-values for the observed differences among transects for the other
three beaches that had exceedences were all greater than 0.70, indicating that the magnitude of the
differences shown in Table 6 could very likely result from simple random variation, the p-value
for Belle Isle's inter-transect differences in exceedence was 0.08, indicating a somewhat unlikely
result unless there are truly systematic differences among these locations. As further evidence
of some transect effect at this beach, afternoon results for Belle Isle only are shown (afternoon
results for the other beaches do not indicate any different variability than the morning results).
The difference in the afternoon is even more dramatic for Belle Isle, with a p-value of less than
0.01.  Managers at Belle Isle would be well advised to pay particularly close attention to the area
defined by transects 2 and 3, and to investigate the cause for this phenomenon with an eye toward
possible remediation.
                                          21

-------
Sampling depth

One of the more interesting outcomes of these beach studies was the lack of a clear gradient in the
indicator density with respect to the depth from the surface at which the sample was collected (Table
7).   Such a difference  with  depth might be expected from the fact that water near the bottom is
more likely to contain suspended particles from the sediment, which is presumably relatively rich in
bacteria. Others have observed depth effects [Public Heath Laboratory Service Water Surveillance
Group, 1995].

Bottom samples (0.075 m from the bottom) were collected in knee-deep and chest-deep water and
mid-depth samples (0.65 m from the bottom) were collected in chest-deep water. Geometric mean
densities over the duration of the study and over all transects for each type of depth  sample are
compared to corresponding surface densities (from only those samples collected at the same time as
depth samples) in Figure 7.

The effect of depth on indicator  density results was examined in one  other study [PHLS, 1995].
The findings of that study indicated  that paired comparisons between surface, 30 cm and 100 cm
depth samples, showed a significant tendency for higher fecal streptococci counts at the surface. A
similar trend was shown for E. coll, but the differences were not statistically significant. Although the
procedure for taking surface samples was not described, an accumulation of microbes at the water-air
interface could have accounted for the differences observed in the PHLS studies.

              Table 7.  Comparison of geometric mean density1 of indicator densities
                        by depth of water from which the sample was collected. "P0"
                        is the "p-value"  for the difference (Iog10 means) between the
                        surface and respective depth means.
Location
West
Beach
Belle Isle
Wollaston
Beach
Imperial
Beach
Miami
Beach Park
Knee deep water
Surface2
32
154
31
3
79
Bottom3
36
142
47
3
87
Po4
0.237
0.324
0.040
0.716
0.499
Chest deep water
Surface5
19
3
18
2
2
Bottom
22
4
16
2
3
PO
0.229
0.082
0.732
0.673
0.478
Surface5
20
o
J
18
2
o
3
Mid-
depth6
21
4
14
3
4
PO
0.954
0.331
0.315
0.292
0.045
 1 Geometric means are over all transects and sampling visits. For surface samples, only those samples for
 which a corresponding depth sample was taken at the same time are considered. Numbers of samples range
 from 18 to 30. 2 Samples collected 0.3 m below the surface. 3 Samples collected 0.075 m from the bottom.
 4 "P-value" for difference in mean Iog10 values between surface and the respective depth (paired samples).
 Generally, a value less than 0.05 is considered "significant"  5 The two values for surface samples from
 chest-deep water may differ slightly due to missing data among the depth samples. 5 Samples collected
 midway between surface and bottom.
                                             22

-------
     Figure 7.  Geometric mean density of surface samples vs. depth samples.
          "*
          o
          

  • -------
    Temporal factors
    
    The significance of temporal factors in sample design
    
    Current EPA guidelines refer to limits on indicator organisms based on the geometric mean of "not less
    than 5 samples equally spaced over a 30-day period" and further specify that, for designated bathing
    beaches, no single  sample exceed the  75th percentile of the corresponding log-normal distribution
    [U.S. EPA, 1986].  Based on health studies conducted by the EPA in the 1970's and early 1980's
    [Dufour, 1984; Cabelli, 1983], the time-averaged recommended maximum geometric means of 126 E.
    coll per 100 mL in freshwater and 35 enterococci per 100 mL in marine water correspond to marginal
    illness rates of 8 and 19 cases of gastroenteritis per 1000 swimmers, respectively.
    
    In current practice, a reasonable scenario for a beach monitoring authority might be to collect a sample
    every, say, Tuesday morning.  This would allow adequate time to obtain the results from that sample
    and, if necessary, take corrective action and resample before the weekend, or give sufficiently advanced
    notice to the public that the beach will be closed the following weekend.  Closure, or action to avoid
    closure, would be based on the above 30-day geometric means or single sample limits.
    
    Bearing this in mind, the daily, and sometimes hourly, collection of samples in both the morning and
    afternoon in this analysis will give insight into the following questions on sampling strategy:
    
        •  How well does a 30-day mean characterize recreational water quality on a given day?
        •  How does  an  observed exceedence (or non-exceedence) in today's sample correspond to
           water quality four (or three, two, one) days in the future?
        •  When is a good time of day to sample?
    
    Findings from the EMPACT Beaches study with regard to temporal factors
    
    Major findings on temporal variation are:
    
        •  From one day  to the next, the geometric mean indicator density over all sampling locations
           changed by a factor of 2 (doubling or halving) or more about half the time at each studied
           beach.
        •  There was a limited statistical relationship observed between levels on the sampling day and
           the following day at three out of the five beaches studied.
        •  Indicator densities tended to decline from morning to afternoon at three of the beaches.  This
           was true for both freshwater and marine indicators. Early morning results were fairly indicative
           of afternoon water quality,  realizing that the density may be lower as the day progresses.
    
    Day-to-day variability
    
    Time series plots of geometric mean indicator densities over all sampling locations in knee-deep and
    chest-deep  water at 9:00  a.m. are shown in Figure 8 for each of the five beaches.  From the analysis
    of spatial factors, depth zone was seen to have a substantial impact on indicator densities. These plots
    are representative of samples from zones in which adult swimmers would be exposed.
    
    The time series plots of Figure 8  afford us a rather accurate daily "snapshot" of each beach. At
    the same time, we may simulate a  realistic beach monitoring scenario. For illustration, we will use
    scenario described  in the previous section - a single Tuesday morning sample.  Having examined
                                                24
    

    -------
    other scenarios, however, we can offer assurance that the conclusions apply to a single-sample scheme
    for any day of the week or time of day, although specific details  in the behavior of the simulated
    monitoring results will, of course, vary.
    
    Based on this scenario, the four-week moving geometric mean is superimposed on these graphs for a
    sample collected on Tuesday morning, approximating a monitoring result based on 5 samples collected
    over a 30-day period (actually a 29-day period) as specified in U.S. EPA guidelines. Naturally, a four
    week moving average cannot be calculated until August 1, four weeks from the first Tuesday of the
    study.  From then on, the running geometric mean is based on an idealized Tuesday sample result - a
    single sample that matches the geometric mean of six knee- and chest-deep samples.
    
    In the first of these graphs, which is for West Beach, note that the only occasion during August when
    the point-in-time geometric mean for the morning exceeded 126 was on Sunday the 27th. On this day,
    the geometric mean reached 446, well above the single-sample limit of 235 E. coll per 100 mL in fresh
    water.  Not only does the 29-day geometric mean fail to predict this event, it is, in fact, at its lowest
    level of the study.  There appears to be very little possibility that any single sample that had been
    collected on the preceding Tuesday would have itself exceeded the single-sample limit (235), given
    that the highest single value actually obtained on that Tuesday morning of August 22 was 11 E. coll
    per 100 mL. Even the day before August 27, the highest observed single density was 23.  In fact, there
    is nothing from these data that would indicate this high indicator levels on Sunday, August 27 at West
    Beach, other than the monitoring results from that very day. The last rainfall prior to this day had been
    observed about four days earlier, and was fairly light (0.3"). There was a brisk on-shore wind on this
    particular Sunday, a factor which will be shown to have an impact at this beach, but stronger winds
    that were observed nine to eleven days prior to this failed to exhibit such a strong effect.
    
    This is just one contradiction of many that can be gleaned from these plots.  For example, at Wollaston
    Beach, the only beach at which there was not a preponderance of results on one side or the other of
    the recommended limit, only five out of eleven exceedences were captured by the 29-day Tuesday
    moving geometric mean.
    
    Aside from the geometric mean over 29 days, it is obvious that variations in the geometric mean at
    a point in time can be quite large in themselves from one day to the next.  In fact, for each of these
    beaches and considering the 9:00 a.m. results only, on at least half of the days during this study the
    change in geometric mean indicator level from the previous day was by a factor of 2 or more, that
    is, it more than doubled or halved from the previous day (Table 9). Day-to-day changes in indicator
    density in this study were large enough that the geometric mean over five Tuesdays, or, for that matter,
    any geometric mean of five observations over a 30-day period, will often not reflect true conditions at
    a beach at a given time.
                                                25
    

    -------
    Figure 8.      Time series plots of geometric mean morning (9:00 a.m.) indicator levels.
                   Heavy line indicates the 4-week geometric mean for samples collected on
                   Tuesday morning of each week.
                      1,000
                        07/M  OW1J  17/23   OMK   08,'IZ  0812!  KWI
                                  07/2.1   (HUD2   BK''12  IBW2
                        0711.1  07/13  07i'2.1  OHM
                                                              07W   17/13
                                                                                          M/tll
                                                         ~~ Geometric mean of daily samples at 9:00 a.m.
                                                         ~ Running geometric mean of samples over the
                                                           last 5 Tuesdays
                                                         * Tuesday sampling value for the running
                                                           geometric mean
                                                26
    

    -------
                  Table 9.  Ratio to the previous day of the geometric mean (GM) of all
                            knee- and chest-depth samples taken at the 9:00  a.m. visit.
                            Interquartile range (25th and 75th percentiles) and  frequency
                            with which the geometric mean changes by a factor of 2 or more
                            (ratio<0.5 or ratio>2) and 5 or more (ratio<0.2 or ratio>5).
    Location
    West Beach
    Belle Isle
    Wollaston
    Beach
    Imperial Beach
    Miami Beach
    Number
    54
    56
    59
    60
    61
    Interquartile range
    for change in GM
    25th %-ile
    0.32
    0.55
    0.39
    0.50
    0.33
    75th%-ile
    1.99
    2.34
    2.50
    2.00
    2.64
    Percent of days that the
    ratio to previous GM is:
    <0.5 or >2
    67%
    43
    66
    55
    69
    <0.2 or >5
    20%
    21
    25
    13
    28
    Table 10 compares the serial correlation between indicator densities at each beach using a 1-day, 2-day,
    and 4-day time lag.  This is simply the correlation between two time-related variables, the log mean
    indicator density over all samples on a given day at 9:00 a.m. and the corresponding log mean at 9:00
    a.m. 1, 2, or 4 days later. The higher this correlation, a perfect correlation being +1 (or -1), the better
    will be our ability to predict results 1,2, or 4 days in the future based on "today's" results. While none
    of these correlation values are particularly impressive, at least a positive relationship is indicated at
    most of the beaches between one day and the very next day.  Correlation becomes markedly weaker
    when the lag is two days, and negligible at four days.  These results do not support a scenario whereby
    water is sampled on Tuesday for a decision on whether it will be acceptable to swim next Saturday.
    
    Significant day-to-day, week-to-week, or month-to-month variation has been observed in several
    studies (Seyfried, 1973, Cheung, 1990, Brenniman, 1981, PHLS, 1995). The differences were as
    great as two orders of magnitude between days, weeks, or months. Similar differences were  observed
    in the current study.
                                                27
    

    -------
             Table 10. Serial correlations between log means separated by 1, 2, or 4 days
    Location
    Time lag between samples
    1-day lag
    2-day lag
    4-day lag
    Correlation coefficients1:
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach
    Miami Beach Pk.
    0.27
    0.18
    0.44
    0.45
    0.40
    0.04
    0.06
    0.18
    0.28
    0.30
    -0.03
    0.17
    0.13
    -0.09
    0.18
    P-values2:
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach
    Miami Beach Pk.
    0.039
    0.191
    0.001
    <0.001
    0.001
    0.789
    0.644
    0.176
    0.031
    0.021
    0.805
    0.233
    0.329
    0.514
    0.180
                    Correlations greater than about 0.24 are statistically significant
                    (ot=0.05).  Sample sizes are in the range of N=50-60 (days).
                    Probability associated with the sample correlation under the null
                    hypothesis that the true correlation is zero.
    
    Variation between morning and afternoon results
    
    A phenomenon that was observed to occur in three of the five beaches studied was a tendency for
    indicator levels to  decrease between the morning (9:00 a.m.) and afternoon (2:00 p.m.) sampling
    events (Figure 9). We use Iog10 of the indicator density per 100 mL as the  y-axis in these plots, since
    there are problems with scale in attempting to show the changes in geometric mean or percentage
    changes. Note that the predominant effect seen at Miami Beach Park, Wollaston Beach, and Belle
    Isle, the three beaches where indicator levels were relatively high, was a decline in the 2:00 p.m.
    reading as compared to the 9:00 a.m. value.
                                                28
    

    -------
    Consider the scenario of taking one Tuesday morning sample for a decision whether to close the
    beach on Saturday and those instances when a Tuesday sample from knee-deep water exceeds the
    EPA recommended limit for a single sample of 235 E. coll or 104 enterococci per 100 mL [1986].
    Had that sample been collected on Saturday instead, would the decision to close the beach or
    not, have been the  same?  How well does the Tuesday sample discriminate between acceptable
    and unacceptable Saturday results?  Does an exceedence on Tuesday indicate a  higher risk of
    exceedence the following Saturday?
    
    The first part of Table 11 examines each single sample from knee-deep water in terms of whether
    it exceeds the EPA limit and whether that sample taken from the very same location four days
    earlier had exceeded the limit.  In effect, this gives the probability of exceedence four days from
    now when (in column 1) today's sample does not exceed the limit or (in column 2) today's sample
    does exceed the limit.  One would hope that the probability of exceedence four  days hence is
    substantially higher when today's sample exceeds the limit than when it does not. That is, column
    2 should be substantially higher than column 1.
    
    This is not the case.  In fact, for the four-daytime lag, the percentages in both columns are remarkably
    similar except at Wollaston Beach.  It may well be that the validity of this sampling scenario varies
    by beach, and has  merit at Wollaston.  However, note that the difference at Wollaston is not
    statistically significant (p-value = 0.31).
    
    What about sampling on Friday morning, the day before a decision must be made?  We can see
    from Table 11 that a sample was generally more likely to exceed  the limit if that same sample
    had exceeded the limit the previous day ("1  day earlier").   Sampling two days earlier may be
    somewhat, but not much, better than sampling four days earlier. In fact, note that all exceedences
    at Wollaston, the only beach where the four-day lag appeared to hold promise, are  missed by the
    two-day sampling lag.
    
             Table 11.   Percentage of total samples collected that exceeded the single sample
                       limit I U.S. EPA, 1986], classified by whether the sample from the same
                       location was above or below thi" limit I, 2, or 4 davs earlier.
    Location
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach3
    Miami Beaeh Park
    
    \\ c-4 Beach
    Belle lsk
    \\ ollaston Beaeh
    Imperial Btadr
    Miami Beach I'.ttk
    Percent of samp)
    limit (total numb
    4 dajs earlier
    < Limit
    9.2",, (152)
    W.6 (32)
    13.X (152)
    0.0 (174)
    62 5 C")
    2 davs earlier
    «" Limit
    1 1 4",( (1*M
    '1 '* li"1!
    IhU (lN>(
    Oil (1MM
    52 X T2l
    cs that rvcmfrd the I'TA-recommended
    er of samples injmrenthcscs)1
    the sample was: 1 da\ earlier the
    > Limit < Limit i
    9.1% (22)
    91.5 (130)
    1K.2 (22)
    (0)
    62 4 (1 0 1 )
    the sample was:
    •* Limit
    10 il'-, (21)
    l»2 * (142)
    OH (22)
    (0)
    (A? ^UOIL
    9.3% (161)
    50.0 (32) I
    12.7 (158) |
    0.0 (183) 1
    52.8 (72) I
    single sample
    sample was:
    > Limit
    31.X% (22)
    93.2 (148)
    25.0 (24)
    (0)
    67.3 (110)
    
      For example, ul 152 samples found to be less than the single-sample limit (235 E. coli per 100
     mL) in West Beach, ^ 2l'» exceeded the limit 4 days later, while of the 22 samples that did
     exceed the limit, 9.1% were in exceedence 4 days later.
     • There were no single sample exceedences at Imperial Beach.
    
                                              29
    

    -------
    Figure 9.     Change in mean Iog10 indicator density between the morning (9:00 a.m.)
    
                 sample and afternoon (2:00 p.m.) sample.
       -2.0
                 r-
                 e
                                              2.0
                                                Imperial Beach
                                         30
    

    -------
    Previous studies [Sieracki, 1978] have demonstrated the die-off of E. coli after exposure to sunlight,
    an effect supported by these data when cloud cover at the time of the afternoon sample is considered.
    As seen in Table 12, and illustrated by Figure 10, indicator levels tended to be lower in the afternoon
    relative to the same morning on sunny afternoons (i.e. less than 50% cloud cover) at both freshwater
    beaches.  At West Beach, lower E.  coli densities were observed in sunny or mostly sunny afternoons,
    but higher levels were seen in the afternoon on cloudy days.  E. coli at Belle Isle generally dropped in
    the afternoon even on overcast days, but decreased even more on those occasions when cloud cover
    was less than 50%.
    
             Table 12.  Comparison of geometric mean density of respective indicator
                       organism at each study beach between morning and afternoon
                       visits - sunny vs. cloudy days.
    Location
    West
    Beach
    Belle
    Isle
    Wollaston
    Beach
    Imperial
    Beach
    Miami
    Beach
    Cloudy days
    (Greater than 50% cloud cover)
    Number
    19
    19
    32
    30
    32
    9:00 a.m.
    24
    202
    27
    7
    149
    2:00 p.m.
    31
    125
    14
    4
    80
    Sunny days
    (Less than 50% cloud cover)
    Number
    41
    41
    25
    31
    30
    9:00 a.m.
    22
    235
    21
    4
    84
    2:00 p.m.
    17
    60
    15
    4
    20
    Although enterococci have been shown to be relatively more resistant to solar radiation than are E.
    coli [Sieracki 1978], significantly lower densities of this indicator were noted at Miami Beach Park in
    the afternoon relative to corresponding morning levels on days when there was less than 50% cloud
    cover at the time of the afternoon visit. Neither of the other two saltwater beaches exhibited this effect
    to a significant degree.  In our evaluation of changes between morning and afternoon, other effects,
    such as tidal stage, rainfall, air and water temperature, and bather load were included as co-variable
    with cloud cover, but failed to account for the apparent effect of sunlight at the Miami Beach Park
    location.  Of course, this research only indicates association between indicator levels and cloud cover,
    not causation of die-off of either E. coli or enterococci because of exposure to sunlight. There may be
    other, unmeasured associations with cloud cover that influence indicator levels at any of the beaches.
    
    Some studies in the literature have reported significant differences between the indicator densities
    obtained in the morning when compared with results obtained in the afternoon [Seranno et a/., 1998;
    Brenniman, 1981; Seyfried, 1973]. Indicator densities were generally highest in the early morning
    and lowest in the mid-afternoon.  These authors point out that this time interval is closely related
    to the period  of greatest solar radiation intensity.  What we may be seeing  is a confirmation of this
    phenomenon.
                                                31
    

    -------
    Figure 10.     Geometric mean indicator density for morning (9:00 a.m.) samples and
                   afternoon (2:00 p.m.) samples, by whether afternoon was sunny (< 50%
                   cloud cover) or not sunny.
              1,000
           _
           E
           ^"   100
           —
           u
           1
           O
    10	i
    Results from the hourly sampling that was performed on several days at each study beach are also
    available for ascertaining whether there were any predominant diurnal patterns in indicator density.
    We can make at least a qualitative assessment of this by examining Figure 11, in which indicator levels
    observed on the hour (9:00 a.m. through 6:00 p.m.) relative to the overall geometric mean indicator
    level for that day are shown. The heavy lines in these graphs indicate the average (geometric mean)
    of this ratio for each hour.
    
    With exception of West Beach, the tendency is for elevated indicator levels during the first sampling
    event of the day, at 9:00 a.m. This is particularly evident at Miami Beach Park, where average values
    for the first two hours were substantially higher than the overall daily mean, after which values tended
    to be much lower. As can be seen from the actual hourly results, deviations from this overall trend can
    be substantial in any case.  These data are quite noisy; the best indication of overall trend still comes
    from the  normal 9:00 a.m. and 2:00 p.m. daily sample results, which cover the entire two-month
    period.
                                               32
    

    -------
                              Single sample limits and time of day
    
    Again, consider the single-sample exceedence rate for those samples taken at 9:00 a.m. from knee-
    deep water, using the EPA limits of 235 E. coll or 104 enterococci per 100 mL.  On those occasions
    when the morning sample exceeded the limit, at three  of the study beaches a sample taken from
    the same location in that same afternoon was much more likely also to be in exceedence, when
    compared to those occasions when the morning sample  did not exceed the limit. This is illustrated
    in Table 13. At West Beach, for example, of the 22 knee-depth samples that were over the 235 per
    100 mL standard for E. coli at 9:00 a.m., 40.9% were also higher than this limit at 2:00 p.m., as
    compared to 7.3% of the  "acceptable" 9:00 a.m. samples that exceeded 235 by 2:00 p.m.
    
                         Table 13.  Percentage of knee-deep samples collected at
                                  2:00 p.m. that exceeded the single sample limit as
                                  recommended in EPA guidelines [U.S. EPA,1986],
                                  by whether the sample from the same location
                                  was above or below the limit at 9:00 a.m.
    Location
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach
    Miami Beach Park
    
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach
    Miami Beach Park
    Percent of samples above the EPA recommended
    single sample limit at 2:00 p.m.
    (total number of samples in parentheses)
    < Limit at 9:00 am
    7.3 (164)
    21.9 (32)
    7.5 (160)
    1.1 (186)
    19.2 (73)
    > Limit at 9:00 am
    40.9 (22)
    74.8 (151)
    11.5 (26)
    - (0)
    52.2 (113)
    % above the limit:
    At 9:00 a.m.
    11.8 (186)
    82.5 (183)
    14.0 (186)
    0.0 (186)
    60.8 (186)
    At 2:00 p.m.
    11.3 (186)
    65.6 (183)
    8.1 (186)
    1.1 (186)
    39.2 (186)
    In total, single sample exceedences were much more likely to occur in the morning at both Belle
    Isle and Miami Beach Park. At West Beach, the relative number of exceedences was about the
    same at the morning and afternoon visits. Considering both parts of Table 13, it can be seen that a
    morning sample from a given location in the water that exceeds the recommended limit is nearly
    four times as likely also to exceed that limit in the afternoon compared to what may be expected
    by chance (40.9% compared to 11.3%).
                                              33
    

    -------
    Figure 11.     Geometric mean (GM) indicator density by time-of-day (9:00 a.m. - 6:00
                    p.m.) for samples collected on hourly sampling days - ratio of GM on the
                    hour to overall GM for the day.  Average ratio for each hour is shown by
                    the heavy line.
           9:00  10:00
                        12:00  13:00 14:00  15:00  16:00 17:00  18:00
                              Time
                                                          9:01)
        10:00 11:00  12:00 13:00 14:00  15:00  16:00 17:00  18:00
                       Time
           9:00  10:00 11:00  12:0
                            13:0(1 14:00  15:00  16:00 17:00  18:00
                              Time
    9:00  10:00 11:00  12:00 13:00 14:00  15:00  16:00 17:00  18:00
                       Time
                                  9:00 10:00  11:00 12:00 13:00  14:00  15:00 16:00  17:00  18:00
                                                     Time
                                                    34
    

    -------
                                  Single sample limits by the hour
      Hourly single-sample exceedence rates among knee-deep water samples are shown in Table 14.
      Exceedences were lower by mid-afternoon at Wollaston and Miami, confirming what we saw
      earlier when comparing only 9:00 a.m. and 2:00 p.m. results. Belle Isle, which showed a tendency
      to decline between 9:00 a.m.  and 2:00 p.m. when all data were considered, does not show this
      tendency when hourly results alone are considered.  Then too, because of logistic problems, only
      eight days of hourly  sampling were completed at this beach, compared with a full two weeks at
      the other sites.
    
                     Table 14.   Percentage of knee-deep samples that exceeded the single
                               sample limit [U.S. EPA, 1986], by time-of-day. Only hourly
                               sampling results  are used.  For all but Belle Isle, there
                               were 14 hourly sampling days and 39-42 valid knee-deep
                               observations (for Belle Isle there were 23 or 24 observations
                               over 8 days).
    Time of
    day
    09:00
    10:00
    11:00
    12:00
    13:00
    14:00
    15:00
    16:00
    17:00
    18:00
    Percent of knee-deep samples that were in exceedence
    West Beach
    2.4
    4.8
    4.8
    7.1
    0
    2.6
    4.8
    0
    0
    2.4
    Belle Isle
    79.2
    52.2
    39.1
    54.2
    58.3
    75.0
    58.3
    75.0
    66.7
    79.2
    Wollaston
    Beach
    23.8
    16.7
    14.3
    33.3
    26.2
    7.1
    14.3
    5.1
    20.5
    19.0
    Imperial
    Beach
    0
    0
    0
    0
    0
    4.8
    0
    0
    0
    2.4
    Miami
    Beach Park
    78.6
    63.4
    52.4
    45.2
    40.5
    35.7
    26.2
    51.2
    38.1
    50.0
    Environmental and bather effects
    
    A summary of environmental and bather density data collected reveals some striking differences
    among the  study beaches (Table 16).  West Beach was by far the most heavily  used, followed by
    Imperial Beach.  The water at Wollaston Beach was particularly turbid in comparison to the other
    beaches with  an  average of 56 NTU.  Note that turbidity at any of the study beaches was highly
    variable, with the standard deviation of the turbidity values approximately equal to its mean in every
    case (data not shown).
    
    Particularly noteworthy is that weather conditions at Imperial Beach were fairly uniform throughout
    the  period of the  study. Most of the afternoons were sunny or mostly sunny, only trace amounts of
    rainfall were recorded, and the onshore wind was ever present.  This constancy obviously limits any
    conclusions that may be made regarding weather related effects at Imperial.
                                               35
    

    -------
               Table 15.   Environmental   characteristics  and  bather   densities   of
                           EMPACT study beaches.
    
    Sunny/mostly sunny days 1
    Avg. air temperature ("C)1
    Avg. water temperature(°C) J
    Total rainfall (in.)
    # days with rain
    Offshore winds l-2
    - Frequency
    - avg. wind speed (mph)
    Onshore winds u
    - Frequency
    - avg. wind speed (mph)
    Avg. wave height (M)
    Avg. turbidity (NTU)3
    Avg. tide range (ft.) 4
    # bathers on the beach 1
    -<20
    -20-100
    - 101-200
    ->200
    # bathers in the water 1
    -<20
    -20-100
    - 101-200
    ->200
    West
    Beach
    67%
    25.0
    23.6
    9.9
    14
    
    31%
    4.1
    
    59%
    8.3
    0.30
    6
    -
    
    6
    21
    20
    14
    
    19
    31
    9
    2
    Belle Isle
    44%
    24.7
    24.1
    13.2
    17
    
    47%
    4.5
    
    25%
    6.3
    0.02
    4
    -
    
    34
    21
    1
    1
    
    37
    18
    2
    0
    Wollaston
    Beach
    50%
    26.6
    22.2
    14.2
    14
    
    26%
    3.0
    
    39%
    5.8
    0.11
    56
    9.9
    
    52
    10
    0
    0
    
    62
    0
    0
    0
    Imperial
    Beach
    71%
    22.5
    19.8
    <0.1
    2
    
    0%
    -
    
    81%
    4.9
    0.36
    1
    4.2
    
    19
    37
    2
    1
    
    30
    29
    0
    0
    Miami
    Beach
    Park
    48%
    27.5
    27.4
    13.7
    22
    
    24%
    3.3
    
    56%
    4.6
    0.13
    7
    0.6
    
    50
    11
    1
    0
    
    52
    10
    0
    0
     1 at 2:00 p.m.  2 Wind at a direction of at least 45° with respect to shoreline
     3 as measured from the central sampling location in the water (9:00 a.m. & 2:00 p.m.)
     4 average difference in water level between succeeding low and high tides
    
    Statistical evaluation
    
    In evaluating environmental and bather effects, the goal was to develop a model for each beach that
    might be used in predicting log mean indicator density 24 hours later based on its current value, since
    with the culture methods used, "current" values are not known until the next day. Thus, the previous
    day's Iog10 density is present  in all of these models, even though it  proves to be non-significant in
    one case. A stepwise regression procedure  [Draper and Smith, 1981] was utilized on the potential
    predictor variables, with the significance level ("p-value")  for entry or removal from the equation
    set at 0.10. Trial and error was involved in selecting the "best" regression as well.  In some cases,
    forcing one or two variables into the equation, in addition to the previous day's density, resulted in
                                                36
    

    -------
    some improvement in its explanatory power with little or no increase in its complexity (number of
    independent variables in the model).
    
    These analyses are exploratory, indicating variables which might be useful predictors of log microbial
    densities.  There is a tendency for any technique designed to select a "best" regression model to
    overstate the fit, with regard to significance levels, partial or total R2, or any other measure. To the
    extent that the  same variables are suggested as being good predictors in a number of different beaches,
    this suggests that these variables really are important.  Several variables show up consistently in these
    models: rainfall, bathers on the beach, etc.  There is good theoretical rationale to believe that these
    variables make a difference in water quality.  Consequently further examination of these variables in
    more detail will be presented.
    
    Potential explanatory factors considered in  this section are not "design  variables",  in  contrast to
    the time-of-day, day-of-week, depth zone,  and transect of the previous  section.   Rather, we  take
    environmental and human factors as  they occur.  Often, these variables may be correlated with one
    another, and, thus, we do not observe an effect that is purely a result of a single factor. For example,
    it is probable that bather density at a  beach, the occurrence of rain in the past 24 hours, whether it is
    sunny or not, and the temperature are all interrelated. We cannot isolate their effects, because we did
    not, for example, prevent bathers from going to the beach on warm, sunny days.  Therefore, some
    variables may  be surrogates for others or may mask the effects of other variables.
    
    In addition, if some condition never occurred, or happened only infrequently, in the study, we cannot
    infer  an effect from that condition.  Such was the case for offshore winds and rainfall at Imperial
    Beach.  Rainfall was observed on  only two  occasions at this location.  It may well be that rainfall
    has an overwhelming effect on conditions at Imperial Beach,  but, if so,  we were not afforded the
    opportunity to observe this effect.
    
    Relationships found at the study beaches are specific to the respective location during the time period
    of the study; blindly generalizing these results to other beaches should be avoided. Such correlates
    with indicator levels as found in this section, however, may serve as a general guideline for related
    research at other sites.
    
    Table 16 summarizes the results of modeling for each of the study beaches. The regression models
    shown are not the only models possible, but are representative of a group of models, any of which
    might be appropriate.
    
    The corresponding log density of indicator organisms 24 hours earlier is included as a covariate in all
    models in Table 16.  In one case, that of Miami Beach Park, the relationship is seen to be non-significant.
    In this case,  there is a highly significant relationship between the 1-day lagged log indicator densities
    by themselves (Table 10 shows this), but the inclusion of other variables is sufficient to account for
    this relationship.
    
    The last column of Table 16 gives the coefficient  of determination, R2,  for each model.  This is the
    proportional of total variation (sum  of squared deviations from predicted value)  of the dependent
    variable that is accounted for by the regression.  A "partial R2" is also listed, which indicates the
    extra proportion of variation accounted for by the other covariates over that which is accounted for
    by knowledge of the previous day's  log density alone. The partial R2 is  a critical consideration in
    improving our prediction of water quality based on the previous day's microbial  levels, as will be
    discussed in the next section, Sources of Sampling Variance.
    
                                                 37
    

    -------
      Table 16.   Results of regression analysis on log  density of indicator organism.
    Factor
    West Beach (N=120)2
    Sunny?3
    Wind vector4
    Rain in past 3 days? 3
    PM & rain in past 24 hrs.?3
    Previous day Iog10 density
    Belle Isle (N=113)
    Sunny?3
    2:00 p.m.?3
    Rain in past 24 hrs.? 3
    <20 bathers on beach?3
    Previous day Iog10 density
    Wollaston Beach
    (N=121)
    2:00 p.m.?3
    Rain in past 24 hrs.? 3
    Low air temperature?3'7
    Onshore wind?3
    <20 bathers on beach?3
    Water level8
    Previous dav Iog10 density
    Imperial Beach (N= 1 1 8)
    <20 bathers in water?3
    Water level
    - if 9:00 a.m. sample9
    - if 2:00 p.m. sample9
    Previous day Iog10 density
    Miami Beach
    Park(N=122)
    Rain in past 24 hrs.?3
    Wind vector
    Water level8
    PM & sunny?3
    Water temp. > median?3
    -if 2:00 p.m. sample11
    Previous day log,n density
    Coefficient
    Int.=1.0322
    -0.329
    0.029
    0.229
    -0.326
    0.280
    Int.= 1.769
    -0.180
    -0.281
    0.173
    0.148
    0.212
    Int.=0.928
    -0.259
    0.396
    0.239
    0.195
    -0.334
    0.043
    0.234
    Int.=0.236
    -0.246
    0.158
    0.102
    0.221
    Int.= 1.326
    0.258
    0.041
    0.290
    -0.111
    -0.387
    0.071
    P-value1
    
    < 0.001
    < 0.001
    0.024
    0.032
    < 0.001
    
    0.005
    < 0.002
    0.016
    0.076
    0.010
    
    0.007
    0.001
    0.029
    0.049
    0.051
    0.003
    0.005
    
    0.005
    < 0.001
    0.00710
    0.010
    
    0.009
    0.004
    < 0.001
    0.031
    0.004
    0.371
    R2
    
    R2 = 0.345
    Partial R2 = 0.24 6
    
    R2 = 0.42
    Partial R2 = 0.28
    
    
    R2 = 0.37
    Partial R2 = 0.24
    
    
    
    
    R2 = 0.31
    Partial R2 = 0.21
    
    R2 = 0.42
    Partial R2 = 0.37
    
    1 "Significance level" of the variable - required to be < 0.10 for stepwise selection.
    2 N=# of observations, Int.=intercept  3 Dummy variable: value = 1 if the condition is true, 0
    if false.  4 Wind speed x direction vector, where the direction vector is +1 for a wind blowing
    straight onshore, -1 for offshore.   5 Proportion  of variation explained by the regression.  6
    Additional proportion of variation  explained by environmental and bather factors over that
    explained by previous day's log density alone. 7 Air temperature less than the 25th percentile
    for the respective time-of-day.  8 Water level in feet above mean lower-low tide mark. 9 Effect
    differs between morning  and afternoon.  10 P-value for the difference between morning and
    afternoon effects.   u Water temperature higher than the median (for that time of day) is a
    significant effect only at 2:00 p.m.
                                             38
    

    -------
    Tides
    
    Among the three beaches in this study subject to tidal movement, tide stage is seen to be a significant
    determinant of enterococci density within the swimming area; in fact, in terms of the p-values of Table
    16, tide stage has the strongest association with microbial contamination of all the environmental
    factors examined at these beaches.  We note that the water level, measured in feet above the mean
    lower-low tide mark (the lower of the two low tides in a 24-hour tidal cycle), was found to have a
    stronger association with bacteria densities than the actual tide stage itself, which we defined as high
    or low when samples were collected within  1.5 hours of the actual event.  Neither did it seem to
    matter whether the tide was incoming or outgoing.  Figure 12, data for which are shown in Table 17,
    illustrates the observed water level effect, comparing geometric mean enterococci levels for times
    when the water level was in the lower and upper quartiles  of its range at each particular beach for
    both the morning and afternoon visits.  Ranges of water level, at least over the times when sample
    collection took place at that beach, are also indicated in Table 17.
    
                    Table 17.  Comparisons of geometric mean enterococci density per 100 mL
                              at low, mid, and high water levels for marine beaches (based on
                              25th and 75th percentiles of water level).
    Location
    Wollaston Beach
    Imperial Beach
    Miami Beach Park
    9:00 a.m.
    Water level:
    Low
    12
    2
    85
    Mid
    28
    5
    101
    High
    37
    11
    140
    2:00 p.m.
    Water level:
    Low
    11
    4
    20
    Mid
    12
    3
    49
    High
    22
    6
    228
    In each case, enterococci densities are seen to be higher when the water level was in its upper quartile
    compared to densities obtained during lower quartile water stages; this is so even at the estuarine
    beach, Miami Beach Park, where tidal variations were relatively low. The fact that water level
    shows a stronger association with microbial levels than does tidal stage itself or whether the tide is
    incoming or outgoing suggests that suspension of sediment or debris from the beach may contribute
    to contamination, perhaps in tandem with waste washing in from deeper waters. Shoreline sediment,
    of course, would constantly mix with bathing waters, and may ultimately affect water of swimming
    depths whether the tide is coming in or going out, but sediment higher on the beach is more exposed
    to humans and animals as well as to air and may be richer in microbial content.
    
    Weather
    
    We had previously discussed the apparent  effect of sunlight on afternoon readings.  Other weather
    factors are also seen to play important roles in determining microbial  contamination. These vary
    among the beaches of this study, and include wind, particularly whether  the wind is blowing onshore
    or offshore, the occurrence of rain in the hours prior to sampling, and air or water temperature. Rainfall
    exhibited a strong association with water quality in three of the study beaches,  as did wind velocity
    and direction, while association with air or water temperature was found in two of the beaches.
                                                39
    

    -------
    Figure 12.    Geometric mean enterococci density by tide stage at marine beaches.
      -
       E
           1,000
                   Low   LJ Mid  LJ  High tide
             100
       o
       CJ
       o
    10
                      9 am     2  pin
                        Wollaston
     9 am     2 pm
    Imperial Beach
                                                          9 am      2 pm
                                                           Miami Beach
    These various weather effects are illustrated in Figure 13 and summarized in Table 18. Significant
    rain effects were observed at Miami Beach Park, Wollaston Beach, and Belle Isle.  In all three cases,
    as might be expected, the occurrence of rainfall within the past 24 hours is associated with higher
    indicator levels at the beach. Note that measurable rainfall occurred on only two occasions at Imperial
    Beach; thus, we have no real opportunity to discern rain effects due to rain  at this location. While
    previous 24-hour rainfall fails to be a significant factor at West Beach, there is an evident rainfall effect
    when we consider rain in the past three days, as indicated in Table 18.  In essence, then, rainfall was
    observed to have an effect on water quality in all cases where we had the opportunity to observe this
    effect.  A 24-hour window is arbitrary; in truth, rain effects on recreational water quality may likely
    vary from beach to beach, depending on such factors as the size and gradient  of the drainage area.
    
    Wind vector at the beach, which takes into account the wind direction as well as wind speed, is a
    significant factor at Miami Beach Park, Wollaston, and West Beach, in each case resulting in higher
    geometric mean indicator densities with a relatively strong onshore wind component and lower
    densities with an offshore component. Once again, we do not have the opportunity to gauge an
    effect at Imperial Beach, because offshore winds never occurred at this site; its graph in Figure 13
    compares light or calm onshore wind conditions with moderate onshore winds. At Belle Isle, E.
    coll levels were actually somewhat lower, but not significantly so, when winds were onshore in the
    morning, but virtually zero difference with wind vector is seen in the afternoon.
                                               40
    

    -------
    Figure 13.    Geometric mean indicator density for various weather conditions.
                            Wind
        I "I"'
             EE     EE    EE    EE     EE
             -  s.     a  &.    -  -    co.     AD.
            We«t Bch.   Belle lilt  WolhUOl   Imperial   Miami Bch.
                                                                         Rain
                                                     1,000
     =  I     II     I  I    I  I    II
    Went Bch.   Belle Isle  Wollasto*  Imperial  Miami Bch.
                      Water temperature
                Air temperature
        1,000
                                                     1,000
                     I  I    I  I    I  I    It
            WcstBrfl.   Hilkl.lt   Vlollislol   Imperial  MiiniBch,
     EE     EE     EE    E  £     EE
     • 6.     •-.  —     s  s.    a  S.    no.
    
    West Bch.   Bclkltlc  Wollnitoi  Imperial  MiiniBch.
    Elevated water or air temperatures might be predicted to result in an increase in microbial counts at
    a beach. However, either may also be associated with increased sunshine.  Even when cloud cover
    at the time of sampling is accounted for, there may have been occasions on which cloud cover was
    more than 50%, at the time of sampling, yet water temperature was relatively high because of earlier
    sunshine. This is one possible explanation for both cases where either water or air temperature was
    associated with decreased densities of indicators organisms, at Miami Beach Park (water temperature)
    and Wollaston (air temperature), as indicated by Figure 13.  Note that E. coll levels were generally
    higher at Belle Isle when water temperature was  relatively high compared to those occasions when
    water temperature was on the low side, but only on the afternoon visits. The opposite was true  for
    this beach on morning visits, and this is a statistically significant difference  between morning and
    afternoon effects of temperature at this particular beach.  Whether changes in water temperature have
    an effect on growth of these organisms may also be supposed to depend on the range of temperatures
    involved; water temperatures  at Imperial Beach were generally the lowest among these study beaches,
    reaching a high of only 23° C, possibly still too low to promote bacterial growth.
                                                 41
    

    -------
    Table 18.   Comparisons of geometric mean indicator density per 100
               mL for various weather-related variables.
    Location
    West Beach
    Belle Isle
    Wo Hasten Beach
    imperial Beach
    Miami Beach Park
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach2
    Miami Beach Park
    West Beach
    Belle Isle
    Wollaslon Beach
    Imperial Beach
    Miami Beach Park
    West Bcacli
    Belle Isle
    Wollaston Beach
    Imperial Beach
    Miami Beach Park
    9:00 a.m.
    Rained in the j
    No
    20
    191
    18
    5
    95
    Strong
    offshore
    11
    261
    14
    4
    gt)
    Cold
    26
    176
    18
    4
    114
    Cold
    28
    198
    "T -s
    /.»
    5
    172
    Yes
    31
    346
    67
    7
    167
    Wind
    Strong
    onshore
    47
    161
    81
    6
    262
    Water tempera
    Warm
    31
    311
    16
    5
    72
    Air temperat
    Warm
    18
    270
    14
    5
    68
    2:00 p.m.
    past 24 hours?
    No
    21
    83
    14
    4
    37
    rector1
    Strong
    offshore
    19
    76
    11
    5
    24
    ture extremes
    Cold
    29
    154
    20
    4
    78
    ure extremes*
    	 Cold""
    29
    126
    18
    -»
    J
    95
    Yes
    16
    301
    •"» T
    Jj
    5
    354
    Strong
    onshore
    34
    77
    20
    4
    60
    Warm
    18
    63
    15
    6
    20
    Warm
    15
    68
    11
    5
    30
    Upper 50n percentiles of offshore & onshore winds. " No offshore winds; comparison is
    between lower & upper 25l!r pereentile of (onshore) winds. 3 Lower & upper 25lh
    percentile of recorded temperatures correspond to "Cold" & "Warm", respectively.
    42
    

    -------
    Bather density
    
    Bather density at a beach was recorded in order to ascertain whether the presence of bathers itself
    influences microbial contamination.  Both bathers on the beach and bathers in the water at the time
    of sample collection were categorized as fewer than 20, 20-100, 101-200, and more than 200. These
    are further condensed for our purposes into two categories, 0 to 19 and 20 or more, since occasions on
    which over 100 bathers were recorded at any of the beaches are rare. Figure 14 compares indicator
    densities at the 2:00 p.m. sample collection for each of the beaches with regard to whether there were
    fewer than 20 bathers either on the beach or in the water, data for this figure being shown in Table
    19.
    
    Bather effects were found to be significant at Wollaston and Imperial Beaches, where an increase in
    numbers of bathers was found to result in a corresponding increase in enterococci levels. At Wollaston
    Beach, few bathers were ever observed to be in the water at any given moment, but the presence of
    substantial numbers of bathers on the beach was adequate to trigger the increase in microbial levels
    in the water. People on the beach in greater numbers implies the possibility that there are more actual
    bathers in the water, whether more than 20 or not, or that there had been more bathers in the water
    prior to the time of sampling.
    
    Three studies looked at the potential for bather load to affect indicator density [Sekla, 1987; Cheung et
    al, Serannoe^a/., 1998]. Sekla indicated that in spite of bather numbers varying between 0 and 3000,
    there was no related variation in indicator densities.  Seranno et al., on the other hand,  clearly showed
    that as the number of bathers  increased the density of indicator bacteria decreased.   Cheung et al.
    were able to observe a correlation between indicator density and number of bathers at  only one beach
    out of nine beaches and only on one day.  These differences in observed results may be related to the
    location of the study sites and to the type of dispersion of contamination at any one site.  Locations
    with very little water movement may result in data where indicator densities increase with number
    of bathers. At locations where there is significant wave action, indicators shed into the water may be
    dispersed very rapidly and the  relationship between indicator and bather density cannot be developed.
    It appears that the effect of bather load on water quality measurement will have to be determined on
    a case-by-case basis.
                                                43
    

    -------
    Figure 14.    Geometric mean indicator organism density by number of bathers on
                 the beach and in the water.
    1,000
    -J
    1
    o
    ^H
    o3 100
    a
    jfr
    • PN
    09
    fl
    0
    'd
    h 10
    o
    -4-*
    69
    U
    SB
    fi
    HH
    1
    
    11 <20 on beach HI <20 in water
    >20 on beach >20 in water
    
    jjL
    
    West Beach Wollaston Miami Beach
    Belle Isle Imperial Beach
                    Table 19.  Comparisons of geometric mean indicator density per 100
                              mL. vs. number of bathers in the water and on the beach, for
                              the afternoon (2:00 p.m.) sampling visits.
    Location
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach
    Miami Beach Park
    Bathers on the beach
    0-19
    23
    114
    14
    3
    40
    20 or more
    20
    65
    18
    5
    43
    Bathers in the water
    0-19
    18
    115
    14
    3
    41
    20 or more
    22
    58
    —
    5
    39
                                            44
    

    -------
                                   Sampling Variance
    
    Sources of variation
    
    Populations, sampling, variance, and components of variation
    
    Sampling variance is a measure of the variability of a measurement, and is calculated from a sample
    as
                                             n-\
    
    in which n is the sample size, x , the sample mean, and x; is the i* sample value for the measurement
    (i=lton). The x; are assumed  to be  results that have been realized from a random sampling of
    somepopulation encompassing all of the possible individual x values that might be obtained.
    
    By the term "population," we  mean the aggregate  from which samples  are collected and to which
    the variance given by Formula 1  applies. It is the population to which any inferences derived from
    sampling apply.  Later, we will discuss in detail the Data Quality Objectives (DQO) Process [U.S. EPA
    2000b], which is a formal framework for considering study objectives (the  inferences and decisions
    to be made), population, and sampling. For now, we wish simply to emphasis the importance of the
    concept of  a population to sampling  and appropriate variance components for various population
    concepts.
    
    In the context of water sampling, the population might be, for example, daily values of the indicator
    density  in waist-deep water midway  along the  designated beachfront at 9:00 a.m.  In this case, a
    random sample of days would enable  one to make an inference such as "the geometric mean density
    was X over the last thirty days." Another example  of a population could be the entire stretch of
    waist-deep water within the "bather area" at noon on a specified date.  The inference becomes "the
    geometric mean density was Y in waist-deep water at noon on the 23rd of July."
    
    If day-to-day  variability (at the same point in the water and time-of-day)  is greater than between-
    locations variability (at the same point in time and water depth), then Equation 1 is expected to give a
    higher value for V for samples that have been collected among different days than for those collected
    among different locations.  "Among days" and "among locations" are two  sources of variation, or
    variance components, that can  be thought of as arising from two different, independent populations,
    as in this example.  Sources of variation relevant to the EMPACT Beaches Study are discussed next.
    For each source, the variance component due to that source is calculated with all other sources, spatial
    and temporal, held fixed.
    
    Sampling distribution
    
    Estimates of variance components are used to assess the accuracy of a mean derived from sampling
    based on a model of the sampling distribution, which gives the underlying probabilities of obtaining
    a sample value within a given range. We hope to be able to utilize fairly small sample sizes in beach
    monitoring  - current EPA guidelines  [1986] specify a single sample - a worthy goal of any sample
    design, and, therefore, cannot rely upon some large sample  theory.  Some  assurance is needed that
    normality applies, or else we need to determine the sampling distribution that does apply.
    
                                               45
    

    -------
    Previous research has indicated that Iog10 indicator densities from environmental samples tend to
    be approximately normally distributed  [Fleisher and McFadden,  1980;  Gannon and Busse,  1989;
    Mujeriego et al., 1980]; that is, valid inferences regarding the means of such data can be made in
    using normal theory.  Sufficient data have been collected in this study to enable an evaluation of the
    consequences of using a normal model. This evaluation indicates that this is a reasonable model of
    the Iog10 densities, resulting in probability statements that are in  error by only  +/-!% (i.e., a 95%
    confidence interval may actually be a 96% confidence interval). Given that a normality assumption
    is appropriate for indicator sampling,  this allows us to construct confidence  limits for the log mean
    density based on the sample log mean, log variance, and sample size using "z" (standard normal
    distribution) values and to perform z tests in comparing the sample log mean to some critical value.
    
    Components of spatial variation
    
    On eight days at each of the five beaches that participated in this  study, replicate samples were
    collected at each location in the water during both the 9:00 a.m. and 2:00 p.m. sampling visits.  On
    these occasions, ten samples were collected from the central location (Location  5, knee-deep  water
    at Transect 2) and two samples, at the remaining eight locations.  These samples yield information
    on the small-scale variability inherent in measuring recreational water quality. Only if the water of
    the bathing area were perfectly mixed, by definition, would the variability of water quality between
    different locations in the water equal this small-scale value. Otherwise, between-locations variability
    will always be higher.
    
    The study design used explicitly accounted for all three physical dimensions.  Locations in the water
    at which samples were collected varied in directions parallel to the beachfront, at right angles to the
    beachfront, and vertically from different depths within the water column.
    
    We have seen that the zone from which the sample was collected has the greatest predictable impact
    on microbial  indicator densities of any factor considered in this study. In each of the study beaches,
    bacterial densities become progressively lower as one moves from ankle-deep to knee-deep to chest-
    deep water.
    
    On the other hand, as long as the sample came  from the  same depth zone, there  were only minor
    consistent differences among transects at two of the beaches. Therefore,  given a sample of water, it
    makes a great deal of difference in our ability to assess overall water quality to know how deep the
    water was where that sample was collected, but makes little or no difference to know what where
    along the beachfront it was collected.
    
    Likewise, we found no consistent differences among the various sampling depths.  Recall that, on four
    days, samples were taken from two depths in knee-deep water (0.3 and 0.425 M from the surface) and
    from three depths in chest-deep water (0.3, 0.75,  and 1.425 M from the surface), whereas a standard
    sampling depth of 0.3 M was used on all other days.
    
    In evaluating their respective variances, the different zones, transects, and sampling depths from which
    sample are taken are regarded as having been randomly selected. In reality, however, these dimensions
    were fixed at the start of the study. Thus, the sampling plan is similar to a systematic sample (with the
    exception that a random starting point is not determined each time).  Given the observed differences
    among zones, it is apparent that the value for the variance among zones will be dependent on exactly
    how these zones are defined. Had waist-deep been used instead of knee-deep, the resulting variance
    may have been much different.  Likewise, for different transect spacing or sampling depths. We shall
                                                46
    

    -------
    see, however, that this is not so great a concern - zone and sampling depth variance are not relevant,
    and a way of accounting for transect separation is shown in the next section.
    How is the population denned spatially?
    
        •   If we consider the population to consist of water 0.3 M below the surface within one zone, then
           variation in sampling depth or between zones are not considerations in our sample design.
        •   Variance between zones is important when we expand the population to include all of a given
           area of water. This may be beyond limits of practicality, and is possibly not even desirable
           since health studies themselves were based on approximately waist-deep water results.
        •   We may not have to consider them in sample design, but knowledge of the magnitude of inter-
           zone and depth variances is useful in describing the indicator levels that may be experienced
           along these dimensions.
    Components of temporal variation
    
    We estimate the variance of indicator density among days for samples collected at the same location
    in the water and at the same time of day. This estimates the variance among, for example, samples
    collected in knee-deep water along the leftmost transect at 9:00 in the morning, where a single sample
    is taken on each of several days. Because the duration of this study was 62 days, from July 1 to August
    31, the variance thus estimated is appropriate for a set of random dates chosen out of a sixty-two day
    period of time.
    
    Another temporal component is the intra-day variability, the variation in indicator levels throughout
    a day. The eight to fourteen days of sampling at each beach during which samples were collected on
    the hour from 9:00 a.m. through 6:00 p.m. will enable us to determine an hourly variance.
    
    How is the population denned temporally?
    
        •   EPA criteria refer to a geometric mean over 30 days.  The population implied by this consists
           of microbial water quality during those 30 days. Are samples collected at the same time in the
           morning? Then the population is further limited to water quality at that time  in the morning
           over 30 days.
        •   We have shown that a mean over 30 days does not reflect water quality at any given point
           in time except by chance. An estimate of water quality at a point-in-time is needed.  Our
           population is, then, the water quality at that point in time, and temporal variation ceases to be
           a concern.
        •   Unfortunately, given the state of the art of recreational water assessment, even though we may
           base a sample design on a population that consists of the water quality indicator at 9:00 a.m.
           on Friday, the population of concern is the water quality at noon on the following day.
               o   Temporal variation is again a concern, but now affects our ability to predict.
               o   No amount of sampling will reduce this type of uncertainty. This takes us into the realm
                   of modeling in an attempt to reduce the unexplained part of temporal variability.
                                                47
    

    -------
    Estimates of variance components for the EMPACT beaches
    
    The various components of spatial and temporal variance, and standard deviations (simply the square
    root of the variance), as estimated for each of the five study beaches are presented in Table 20.  Two
    kinds of variance are given for each component. One is a "pure" factor variance for each component,
    that is, the variance that would be seen if we had "perfect" knowledge of the indicator density at each
    sampling location. This is added to the variance within sampling location (that is, the variance among
    the replicate samples taken on occasion at each location), to derive the variance that would actually
    be observed when a single sample is collected at each location and many such samples are taken from
    different transects, zones, days, etc. The pure and observed variances for a given spatial or temporal
    factor apply to the situation in which all of the other factors are held fixed; for example, the variance
    among zones is applicable to samples that are taken along the same transect at the same time.
    
    Statistical tests would reveal significant differences among the different variance components for any
    individual beach, and, likewise, significant differences among the different beaches for any individual
    variance component for most comparisons. There is, however, no reason to suspect that they should
    be the same.
    
    Note that the incremental factor variance components, except for replicates variance, are the lowest, or
    among the lowest, at Imperial Beach and among the highest at Belle Isle. Variance among replicates,
    the "pure error" component, however, follows the opposite trend, being lowest at Belle Isle and highest
    at Imperial Beach.  This implies that, on a large scale, the  distribution of indicator organisms is the
    most uniform at Imperial Beach, the least so at Belle Isle.
    
    Spatial components of variation
    
    Small-scale (replicate) variance
    
    This variance is a result of the variation among different samples taken at the same time from the same
    location in the water from a depth of 0.3  M from the surface.  There  are two components that may
    contribute to the small-scale  variation:
    
        1.  Method variance is  likely negligible for both methods. Precision  (standard deviation) for
           method  1600 is on the order of+/-0.02 [U.S. EPA, 1997b], implying a Iog10 standard deviation
           of about 0.01, affecting only the least significant digit of small-scale standard deviation
           displayed in Table 20.
       2.  This leaves true variation among replicates as the only real contributor to small-scale variance.
           Although we describe replicate samples as being collected from the same location at the same
           time, they, of course, are not. They are as close to one another, temporally and spatially, as
           we can get.
                                               48
    

    -------
              Table 20.   Standard deviation and variance component estimates for the
                         EMPACT beaches.
    Location
    Source of variation
    Replicate
    Samples1
    Sampling
    Depth2
    Depth
    Zones
    Location
    Within
    Zone3
    Hourly
    9:00am -
    6:00pm)
    Among
    Days
    Pure factor variances4
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach
    Miami Beach Pk.
    0.055
    0.037
    0.096
    0.138
    0.039
    5
    0.021
    0.052
    0.010
    0.047
    0.078
    1.480
    0.203
    0.017
    0.911
    0.031
    0.103
    0.036
    0.032
    0.169
    0.091
    0.182
    0.190
    0.118
    0.329
    0.234
    0.168
    0.339
    0.114
    0.432
    Total variances6
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach
    Miami Beach Pk.
    0.055
    0.037
    0.096
    0.138
    0.039
    0.045
    0.058
    0.147
    0.147
    0.086
    0.133
    1.517
    0.299
    0.155
    0.950
    0.086
    0.140
    0.132
    0.169
    0.208
    0.147
    0.219
    0.285
    0.256
    0.368
    0.289
    0.205
    0.434
    0.252
    0.471
    Standard deviations
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach
    Miami Beach Pk.
    0.235
    0.193
    0.309
    0.371
    0.198
    0.212
    0.241
    0.384
    0.384
    0.293
    0.365
    1.232
    0.547
    0.394
    0.975
    0.294
    0.375
    0.363
    0.412
    0.456
    0.383
    0.468
    0.534
    0.506
    0.607
    0.538
    0.453
    0.659
    0.502
    0.686
     1  Samples collected at the same point in the water 0.3 M from the surface. 2 Samples collected at
     the same position in the water, but at different depths below the surface. 3 Samples collected from
     water of the same depth & from the same depth below the surface, but at different points in water.
     4 Variance attributable solely to its respective factor. 5 Estimate of this variance component was
     negative, so is assumed to be zero. 6 Total variance = pure factor variance + replicate sample
     variance (except in the case  of replicate samples themselves, where it is identical to pure factor
     variance).
    Replicate variance represents the least variability that we may expect among any set of water samples,
    even if the water comprising the bathing area were "well-mixed" prior to our collection of the samples.
    This small-scale variability is a component of every other variance shown in Table 20.
    
    Variance among sampling depths
    
    In addition to samples that were collected at a depth of 0.3 M from the surface, on occasion samples
    also were collected at 0.075 M from the bottom in both knee- and chest-deep water, and at 0.75 M from
    surface (mid-depth) in chest-deep water.  Variance among sampling depths represents the variation
    among these samples (including the 0.3 M sample). In most cases, we note that the contribution to
    replicate variance is relatively small, being at most on the same order as the magnitude of the replicate
    variance itself.
                                                49
    

    -------
    In one case (West Beach), the estimate of pure variance due to different sample depths is negative,
    leading to a total variance for this factor that is smaller than the corresponding replicate variance.
    That this should be true is not believable, but that this might occur for one of these estimates is not
    particularly surprising. This simply implies that the variance due to sampling depth is sufficiently
    small that such a result is likely to occur.
    
    We had previously shown that there were no consistent differences among the various sampling depths.
    Additionally, the human health studies that are the basis for recreational water criteria (Dufour, 1984;
    Cabelli, 1983) are based on water samples taken at a depth of 0.3 M from the surface.  For these
    reasons, there is little reason for collecting samples at any other than the standard depth of 0.3 M.
    Thus, the variation in sampling depth is not a relevant consideration in recreational water sampling.
    
    Variance among zones
    
    We have divided the remaining spatial variation into two components, variation among zones (outward
    from  the beach) and variance among locations within depth zone (parallel to the beachfront) for
    reasons mentioned earlier, mainly, that variation among zones was shown to be substantial and, to a
    large degree, predictable.  The consistent differences in indicator levels among zones can be exploited
    by the technique of stratification - sampling within each zone and then combining the individual zone
    means using a weighting scheme to account for their relative importance.
    
    Variance among depth zones as given in  Table 20, however, represents  variation that would be
    encountered if one were to randomly select sampling locations (i.e., ankle-, knee-, chest-deep and
    locations) along a fixed transect.  Variance among zones is seen to have the largest component of
    spatial variation at all but Imperial Beach, a fact that is particularly evident when unique, pure factor
    variances are examined.  At Imperial Beach, all components of spatial variation seem to result in
    virtually the same total variance, whether we look down through the water column, across  the beach,
    or outwards from the beach.
    
    This variance, though, is avoidable because of the consistent ordering of indicator levels, decreasing as
    one goes out from the shoreline from ankle-deep to chest-deep water. Besides the technique of sample
    stratification, it seems reasonable that water of a particular depth be  a separate population of study
    in itself, given that, as previously mentioned, epidemiological models for swimmers' illnesses have
    relied on indicator values from  waist- to chest-deep water. Additionally, contamination in shallow
    water is likely to affect different bather sub-populations.
    
    Variance among locations within zone (among transects)
    
    This leaves the variation within depth zone, or "transect" variance, to contend with. Variance among
    locations within the same depth zone in Table 20 represents an inescapable source of sampling variation.
    This source of variation must be considered in any sampling plan because  rarely, if ever, would we
    be interested in the indicator level at a single point in the water.  Even if we limit our sampling to one
    such location, the question remains, "How 'representative' is this of the water quality in general?"
    
    We can use between locations variance to estimate how much we might expect a single sample taken
    in, say, knee-deep water to differ from the true mean for knee-deep water along that stretch of beach.
    Using a normal  distribution model for log indicator density the half-width of the 95% confidence
    interval is calculated as two times the standard deviation among locations by 2.  At West Beach, for
    
                                                 50
    

    -------
    example, this gives a 95% confidence interval of 2 x 0.294=0.588 logs. The interpretation is that 95%
    of the time the value from a single sample will be within 0.588  logs of the actual geometric mean
    density in knee-deep water. This is equivalent to a factor of nearly 4 (100588=3.9), so that our 95%
    confidence interval implies that the density of a single sample may range from about % of the true
    mean to 4 times the true mean value.
      It may be more interesting to consider what kind of difference might be expected between single
      samples, such as two samples taken in knee-deep water but at somewhat different locations.  At
      West Beach, we would expect the two  samples to differ from one another by more than 0.28
      logs 50% of the time, based on the normal distribution. This means that if one relies on a single
      sample, half of the time a different sample would have given a result more than about twice
      (100-28=1.9) or less than half (1/1.9) the value.
    Replication, sampling from several different locations, is the means of reducing the impact of this
    variance on the uncertainty of the sample estimate.  If V represents the variance of single samples
    among locations, or transects, the variance of the mean of n samples taken from different locations
    within the same zone is V/n.
    
    A valid concern is whether variance among sampling locations itself varies from one zone to another.
    In fact, this question may be asked of all sources of variation.  On theoretical grounds, if these data
    are log-normally distributed, and this does appear to be at least a reasonable model, this implies
    proportionality between mean and standard deviation. Their logs (base 10 or any other base), then,
    will follow a normal distribution with constant variance. In particular, the variance, or, equivalently,
    standard deviation, within the zone will be the same value for each zone. When separate variances
    are computed for ankle, knee, and chest depth zones, as they were for Table 21, there is no evidence
    to refute that this is the case.  This is found to be true for other factors of variation as well, even in
    the case of day-to-day variation where one may suppose that whatever environmental factors may
    affect this sort of variation, the magnitude of their effects close to shore may differ from that in deeper
    water.
    
                               Table 21.  Standard  deviation  among
                                          locations (transects)  within
                                          each separate depth zone.
    Location
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach
    Miami Beach Pk.
    Depth zone
    Ankle
    0.329
    0.389
    0.360
    0.410
    0.458
    Knee
    0.275
    0.361
    0.336
    0.423
    0.428
    Chest
    0.270
    0.369
    0.393
    0.402
    0.470
                                                51
    

    -------
    Temporal components of variation
    
    Hourly variation
    
    At least for the period from 9:00 a.m. to 6:00 p.m., we can determine the variance that takes place
    from hour to hour within a day. Variation on this time scale might be of interest if one were to collect
    samples at random times during the day in order to estimate some daily average.
    
    In our case, however, it is interesting to compare the variance hour-by-hour with that on the larger
    day-to-day time scale. Despite this difference in time scale, the hourly variation  is equal to the daily
    variation, or approximately so, at three of these beaches. At two of these beaches, Miami and Belle
    Isle, we had previously noted a pronounced tendency for indicator levels to drop between the early
    morning and mid-afternoon visits (specifically, between the normal visit times of 9:00 a.m. and 2:00
    p.m.).  This may explain some of the apparent "excess variance" among hourly  results at these  two
    beaches, just as the drop off in indicator levels from shallow to deep water accounts for much of the
    inter-zone variance we saw earlier. While we could not discern a substantial decline between morning
    and afternoon at Imperial Beach, its hourly graph in Figure 11 does show a generally U-shaped curve,
    which may account  for its relatively high hourly variance as well. We do note, also, that all of the
    variance components there, spatial and temporal, appear to be much more like one another than at any
    of the other beaches studied.
    
    Variance among days
    
    This source of variation refers to single samples that are collected at the same time of day at the same
    relative location in the water (e.g., along transect 1 in knee-deep water) at a depth of 0.3 M from the
    surface.  This, in fact, is the variance referred to in the EPA criteria document [Dufour and Ballantine,
    1986] in relation to single sample limits in recommending "a one-sided confidence limit... based on a
    site specific standard deviation " (emphasis added). In the absence of such information, recommended
    log standard deviations are given - 0.4 for E. coli in freshwater and 0.7 for enterococci in marine water
    - which appear to be entirely reasonable based on the standard deviation results among days given in
    Table 20.
    
    The total variance among days would also be used in assessing  the precision of the log mean of 5
    samples over a thirty-day period, as recommended in the criteria document. Letting V represent the
    variance among days, the corresponding variance of the log mean is V/5, and  we could construct
    confidence intervals or upper bounds for the true mean. Here, the "true mean" refers to the actual log
    mean value at the same (relative) location in the water at the same time of day over the entire 30-day
    period.
    
    Variance of the change over a 24-hour period
    
    Current technology  for monitoring recreational water quality results in a one-day lag between the
    time that the samples are collected and the time that results from these samples are known.  Therefore,
    another source of temporal variability that we must be concerned about is that of the change in indicator
    levels that occur over approximately a 24-hour period. This variance component reflects how good a
    prediction of today's water quality based on yesterday's samples will be.
                                                52
    

    -------
    That a prediction is being made implies that a model exists for making that prediction. The simplest,
    and probably most often used, such model is that of status quo - that there will be no change over
    the 24 hours. Lacking any other information, this would be as good a model as any. In this case, the
    variance of the error in our predictions is that of the raw change in log mean density, tomorrow's log
    mean density minus today's.
    
    Earlier, in examining  environmental and bather effects on indicator levels, we developed a linear
    regression for each EMPACT study beach that predicted what the log mean density will be 24 hours
    later based on various weather, tide, and bather density information along with  the current day's
    sample value (see Table 16).  If these models  are used in place of the status quo, the error variance for
    our predictions becomes the difference between the true log mean density and our predicted value.
    
    These two variances are given in Table 22 for each of the EMPACT Study beaches. With remarkable
    consistency, the variance of predictions based on the regression models of Table 16 is about one-half
    that of the raw change (the "status quo" model).  Variances for predictions based on the regression
    models are likely understated in Table 22, simply because of the  amount of "data mining" that went
    into the development of these models, leaving plenty of opportunity for over-fitting the regressions.
    Nevertheless, the benefit of basing the assessment on information in addition to yesterday's samples
    is apparent.
                   Table 22.  Variance and standard deviation of 24-hour change
                              in Iog10 indicator density and of 24-hour change in
                              Iog10 indicator density when modeled as in Table 16.
    Location
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach
    Miami Beach Park
    Raw change in mean
    Iog10 density
    Variance
    0.397
    0.205
    0.423
    0.195
    0.540
    Standard
    Deviation
    0.631
    0.452
    0.651
    0.442
    0.735
    Modeled change in mean
    Iog10 density
    Variance
    0.201
    0.100
    0.213
    0.108
    0.225
    Standard
    Deviation
    0.448
    0.316
    0.462
    0.329
    0.474
                                                53
    

    -------
                     Designing a Beach Monitoring Program
    
    The preceding sections have led us through an examination of the spatial, temporal, and environmental
    factors that influence recreational water quality to the description of potential sources of spatial and
    temporal variance and estimation of their respective magnitudes. The concept of a population as it
    relates to sampling, how it affects which variance components come into play, and the  importance of
    the sampling distribution have been emphasized.  These, together with what we know about human
    health effects, are all factors to be considered in developing an optimal sampling strategy.
    
    A standard for systematically planning for the collection of environmental data have been established
    by the U.S. Environmental Protection Agency by its Data Quality Objectives (DQO)  Process [U.S.
    EPA 2000b]. Implications from the EMPACT Beaches study  with respect to monitoring recreational
    waters will be given in the context of this standard.
    
    The DQO approach
    
    The DQO Process [U.S. EPA 2000b] addresses the requirements for systematic planning to "ensure
    that data collected for the characterization of environmental processes and conditions are of the
    appropriate type and quality for their intended use ..." enumerating seven  steps to be taken in planning
    for data collection projects that support environmental decision-making. These seven steps are shown
    in Figure 15.  The reader is encouraged to refer to the  U.S. EPA DQO  Guidance document, which
    presents in a readable fashion a formal framework for planning any environmental sampling program,
    and can be downloaded from the Internet at http://www.epa.gov/qualityl/qs-docs/g4-final.pdf.
    
               Figure 15. The Data Quality Objectives Process (from "Guidance for the
                          Data Quality Objectives Process", U.S.EPA [2000b])
           Step 1. State the Problem
                  Define the problem; identify the planning team; examine budget,
                  schedule.
           Step 2. Identify the Decision
                  State decision; identify study question; define alternative actions.
           Step 3. Identify the Inputs to the Decision
                  Identify information needed for the decision (information sources,
                  basis for Action Level, sampling/analysis method).
           Step 4. Define the Boundaries of the Study
                  Specify sample characteristics;  define spatial/temporal limits and
                  units of decision-making.
           Step 5. Develop a Decision Rule
                  Define statistical parameters (mean, median); specify Action Level;
                  develop logic for action
           Step 6. Specify Tolerable Limits on Decision Errors
                  Set acceptable limits for decision errors relative to consequences
                  (health effects, costs).
           Step 7. Optimize the Design for Obtaining Data
                  Select resource-effective sampling and analysis plan that meets the
                  performance criteria.
                                               54
    

    -------
    Results from the EMPACT Beaches Study will be related primarily to steps 4 and 7 of the DQO Process.
    Step 6 will be discussed in some detail in order to present concepts that are useful in understanding the
    rationale behind sample size determination. Some brief comments on the remaining steps are:
    
        •   State the problem. Fecal contamination at a recreational beach poses a threat to human health
            and because of this, water quality must be monitored.  Questions to be answered at a local
            level are: Who is to be involved in the monitoring program? What is our budget? What is the
            timeframe for developing the program?
        •   Identify the decision.  At a minimum, results from monitoring will lead to the decision to
            close the beach or not, depending on whether contamination is found to be at an unacceptable
            level. Additional actions may be identified, for example, public posting of microbial counts
            or posting a warning when contamination levels are in a questionable range.
        •   Identify the inputs to the decision.  The very purpose for conducting the EMPACT Beaches
            Study itself was to provide this input.  Other studies on sampling recreational beaches have
            been published [Brenniman et al, 1981; Cheung et al, 1990;  Churchland and Kan, 1982;
            Noble et al., 2000; Pande et al.,  1983; Van Ess and Harding,  1999]. The World Health
            Organization (WHO) has published a series of guidelines for beach monitoring [1975, 1977,
            1998]. A good starting point for Internet resources is the U.S. EPA's Beach Watch website at
            http://www.epa.gov/OST/beaches/. Other resources may be available locally.
        •   Develop a decision rule.  Examples of decision rules are given in current EPA guidelines
            [Dufour and Ballentine,  1986; U.S. EPA, 1986] and are based on limits of acceptable illness
            rates among swimmers of 8 per 1000 in freshwater and 19 per 1000 in marine waters. Beach
            monitoring programs using federal grants authorized by the BEACH Act must use the water
            quality standards (decision rules) adopted by the state [U.S. EPA, 2004].
    
    Defining the boundaries of the study: the population
    
    The concept of a population in its statistical sense was discussed earlier in the section on sources of
    variation. We defined the term "population" as "the aggregate from which samples are collected and
    to which inferences and, ultimately, decisions are made." The phrase "define the boundaries of the
    study" tells us to identify this population in detail. What should the target population, the body of
    water, be in terms of sampling depth, distance from shore, distance along the beachfront, and time?
    
    Sampling depth
    
    We have used 0.3 meters from the surface  as the standard depth at which to collect water samples in
    water that is of swimming depth [APHA, 1998].  Sampling at this depth, at which a swimmer is likely
    to be exposed to pollution,  is recommended in the WHO  approach to monitoring  [1998] and was
    the depth used for microbial sampling in the U.S. EPA epidemiological studies [Cabelli et al., 1982;
    Dufour, 1984], upon which United States recreational water quality standards have been based.
    
    While this study found little in the way of systematic differences among samples collected at different
    depths, this does not mean that there may not be meaningful differences at other beaches.  Regardless,
    sampling from different depths below the surface would seem to introduce an additional component
    of variance, unnecessary in light of the lack of interpretability in regard to human health for sample
    results that come from depths other than 0.3 meters from the surface.
                                                55
    

    -------
    Distance from shore (Depth zone)
    
    This study has demonstrated that the depth zone from which samples are to be collected will likely be
    a critical decision to be made in a beach sampling plan.  At all of the study beaches, indicator levels
    were seen to fall off substantially as one goes from ankle-deep to chest-deep water, to such a degree
    that the likelihood that a beach met current U.S. EPA guidelines (1986) at any point in time was highly
    dependent upon the depth zone from which samples were taken.
    
    Depth  zones in a bathing area can be associated with particular usage patterns and segments of the
    human population that are likely to be affected by the  water quality in that zone. Infants and toddlers,
    given their activities at the beach, are most likely to ingest water in the "ankle zone" as denned by this
    study.  For adults, exposure to contamination from immersing one's head in water seems most likely
    to occur in waist-deep or deeper water, given that substantial swimming activity probably does not
    occur until one reaches this depth. It is advisable that a sampling plan account for these differences in
    the affected population, and be capable of yielding separate estimates of indicator density for shallow
    water and water of swimming depth.
    
    Epidemiological studies have concentrated on adults, and have measured indicator densities in water
    of swimming depth. U.S. EPA guidelines for beach water quality (1986) are based on these studies,
    and, thus, on fecal contamination levels in deeper waters.  Unfortunately, epidemiological data that
    would  enable quantitative evaluations of the risks inherent to infants and toddlers with respect to
    exposure to sewage-contaminated water are not available.
    
    Distance along the beachfront
    
    We define the beach itself in setting  the boundaries for this final spatial dimension, the shoreline
    to be encompassed by a monitoring  program.  This expanse comprises the beach with regard to
    which decisions will be made. One will need to determine areas that may be exposed to pollution, as
    determined by the locations of point sources on the shore or in the water. Conversely, other areas may
    be shielded from contamination. Usage is another consideration. Where do bathers tend to congregate
    and when?  What areas are used primarily by surfers or by swimmers? Answers to these questions
    can help to define unique recreational areas of the beach. In some cases, political boundaries or the
    authority responsible for managing the beach may come into play in determining what constitutes
    a beach.  Public  perception of the beach area will need to be considered, regardless of the officially
    recognized boundaries of the beach. Those responsible for communicating the quality of recreational
    waters  to the public, along with reporting the results  of the day's sample, must unambiguously state
    the area of the beach to which the results apply, that area decided upon in this phase of the design
    process.
    
    The World Health Organization, in its recommendations for beach monitoring [1998],  addresses
    the issue of defining a beachfront for sampling purposes through  its  "primary microbiological
    categorization protocol". This protocol entails a sanitary survey to determine sources of contamination
    and a preliminary sampling phase to determine if "significant variation" along the beach is indicated.
    The term "significant variation", as used in the WHO report, connotes systematic variation in which
    a section of the  beach is consistently more, or less, contaminated than other sections of the  same
    beach.  We have found "significant variation" to exist at each of the five study beaches in the statistical
    meaning of this term, mainly that there is a between-locations variance component, over and above
    the variance we see from replicate sampling at the same location in the water.  This variation, however,
    
    
                                                56
    

    -------
    is not systematic, as seen from the lack of any apparent consistent differences among transects (with
    possible exceptions at Belle Isle and Miami Beach Park).
    
    The length of shoreline that one decides upon to define the monitoring area may itself affect the
    magnitude of the variance encountered in the process of sampling.  Sampling variances shown in
    Table 20 apply to the forty-meter, centrally located expanses of beach that comprised the sampling
    area at each of the study beaches. If sampling had encompassed larger sections of the beaches, these
    variances would likely be higher.  This is so  because values obtained from two points separated by
    a relatively large distance are likely to be less correlated than values obtained from two points that
    are closer together.  The effect can be illustrated by considering only those points  along transects 1
    and 3, the two extreme transects, in our variance calculation, yielding a variance between sampling
    locations that are separated by 40 meters. Variances computed from all the data, as they are  in Table
    20, reflect an average separation of 26.7 meters (40 m separation for one pair of points, and 20 m for
    each of the two other pairs). In addition, "pure error" variance is available, reflecting zero separation
    between sampling points (replicate samples from the same location). Respective variances calculated
    for average separations of 0, 26.7,  and 40 meters are plotted in Figure 16, and joined by a smooth
    curve to show hypothetical relationships
    
    Figure  16.    Variance of loglO indicator density per 100 mLvs. average separation
                  among sampling locations.
              0.25
              0.00
                                                    20       26.7 30
                                     Average separation (m)
    40
                                               57
    

    -------
    between average distance separating the sampling locations and the corresponding sampling variance.
    Stratifying the beach along shoreline segments is a potential counter to this trend, as discussed later,
    in the section "Optimize the design for obtaining data."
    
    Time
    
    A geometric mean calculated from five values over a four-week period did not accurately represent
    actual conditions on any given day at any of our five study beaches.  Figure 8 in the section on
    temporal variation clearly illustrates this fact.  A point-in-time estimate of indicator levels is needed.
    
    The water quality measurement from which decisions will be made should be the contaminant level in
    the swimming area at a time when beachgoers will be exposed, and, unfortunately, current membrane
    filtration and culture methods do not allow us to obtain the sample results any earlier than about 24
    hours after the samples have been collected.  A scenario was introduced earlier, in which the beach
    manager sampled on a Tuesday in order to make decisions  for the next weekend.  Results from a
    Tuesday  sample were shown to have little bearing on conditions that exist the following  Saturday.
    Correlations between indicator levels on one day and those four days later (see Table 10) were seen to
    be negligible, while  serial correlations for a one-day lag were, if not impressive, at least positive.
    
    Collecting samples the day before one must make a decision with regard to public safety at a beach
    may not be convenient, but it appears that data obtained earlier than this may have little or no relevance
    to such a decision.
    
    The time of day when sampling is performed is also likely to be governed by practical considerations
    - one would prefer to collect samples in the morning in order to assure ample time for transporting the
    samples to the lab and for the labs to process the samples and start incubation.  In addition, the earlier
    the sampling time, the more timely the results from the analysis. This study  has demonstrated that,
    for the five beaches studied,  contamination levels in the afternoon were, if anything, likely to be lower
    than in the morning. Thus, morning sampling would tend to err on the safe side.
      To summarize, this study indicates appropriate spatial and temporal boundaries for beach
      sampling to be:
            •     0.3 M from the surface
            •     in water of swimming depth
            •     along a pre-determined beachfront
            •     in the morning
            •     of the day before
    
      Monitoring in shallow (ankle-deep) water may be  performed as an adjunct to, and  viewed
      separately from, monitoring in deeper waters, although results  with respect to children's health
      effects may be difficult to interpret without further health studies
      Specifying tolerable limits on decision errors
    
    The U.S. EPA's DQO guidance document [EPA, 2000b] uses the concept of a power curve to define
    tolerable limits on decision errors.  In our case, a power curve gives the probability of rejecting a
    beach as unfit for bathing as a function of the true (log) indicator level at that beach.  This probability
    will be the complement of what is commonly known as the "Type II" or "/?" (beta) error in those cases
    in which the true indicator level exceeds a given standard but a sample fails to detect this exceedence,
                                                58
    

    -------
    and equal to the "Type I" or "a" (alpha) error when the true indicator level falls below this standard.
    We may also view this in terms of acceptance sampling(see box), in which the beach is considered a
    product that we wish to accept or reject as suitable for use.
    
    For either the power function or the operating characteristics curve, it is necessary and sufficient to
    specify two points on that curve, the shape of the curve being given by the normal (usually) probability
    density function.  The DQO guidance document illustrates this by showing a true value equal to the
    "action level" for the decision variable and another true value equal to an upper limit specified by
    the user.  Figure 18 emphasizes two true density levels that determine the producer and consumer
    points.
    
    Meaningful producer and consumer points or action levels can be determined based on the relationship
    between implied risk and indicator level. Health studies conducted by the EPA give these relationships
    in terms of extra cases of "highly credible gastrointestinal illness" (HCGI) per 1000 swimmers as
    
                                -11.9 + 9.5-log10(£co// per 100 mL)                             (2)
    
    in freshwater [Dufour, 1984], and
    
                                0.1+ 12.3'log10(enterococci per 100 mL)                         (3)
    
    in marine water [Cabelli,  1983]. At EPA recommended limits for the 30-day geometric mean, 126 per
    100 mL for E.  coli and 35 per 100 mL for enterococci in marine water, the implied illness  rates are
    about 8 and  19 per thousand swimmers, respectively.
    
    As an example to be used for sample size estimation in the following section, "Optimizing the design
    for obtaining data", consider the freshwater case and assume that we would desire a low probability
    of accepting a beach as safe for swimming if the swimming-related illness (HCGI) rate is 10 per 1000
    or higher. Somewhat arbitrarily, based on custom more than anything else,  we set this probability
    at 0.05 (5%), and this represents our consumer's point, as in Figure 18.  Conversely, should the
    swimming-related illness rate be 6 per 1000 or less, we  consider this  to be an acceptable risk, and
    want a high probability of accepting this beach, i.e., declaring it to be suitable for swimming. We set
    this probability at 0.95 (again, based on a commonly accepted value for what we mean by a "high
    probability"), to obtain the producer's point in Figure 18.
    
    Equation 2 can be used to determine the values of the Iog10 E. coli density per 100 mLthat correspond
    to each of these decision points. An incidence of 6 cases of HCGI per 1000 is seen to correspond to
    about 1.9 logs, and of 10 cases per 100, to 2.3 logs. Whenever the beta risk at the consumer's point
    is equal to the complement of the alpha risk at the producer's point, the action level will lie exactly
    halfway between the two corresponding illness rates, or log densities forthe indicator organisms. These
    decision points are centered about an illness rate of 8 per 1000, corresponding to 2.1 log density per
    100 mL, the current EPA-recommended action level for a geometric mean over time (Iog10126=2.1).
    
    This is not to be construed to advocate the retention of current EPA recommendations, which, after
    all, apply to a geometric  mean over 30 days (or to single samples). The  EPA recommended action
    level simply offers a convenience reference point for this presentation of a rationale for determining
    an appropriate action level based on the acceptance sampling concepts of consumer and producer's
    points and an operating characteristics curve. Indeed, as far as sample size requirements are concerned,
    
    
                                                 59
    

    -------
    what is important is the width of the interval between the Iog10 indicator values at the producer and
    consumer's points (the tolerance interval).  In our case, this width is 0.4 logs, or +/- 0.2 logs with
    respect to the action level (the half-width of the tolerance interval). A different action level may be
    selected,  but as long as our decision points remain +/- 0.2 logs from this value (corresponding to +/-
    2 illnesses per 1000 swimmers in fresh water), the operating characteristics of the sampling plan, as
    exemplified by the curve in Figure 18,  do not change.
    
      Figure 17. Correspondence between health effects criteria and indicator density.
      o
      o
      ^H
      -
      o>
      OH
      QB
      u
    18
    
    16H
    
    14
    
    12
    
    10
    
     8
    
     6-
    
     4-
    
     2-
                                                    HCGI= -11.9 + 9.5 Iog1ft(£. coll
                                1.5                 2                  2.5
                                L,og(E.coli density per 100 mL)
    A rationale for setting tolerance intervals
    
    We may choose a tolerance interval to be as narrow as we wish. The narrower this interval is, that
    is, the closer the producer and consumer points are to each other, the more precise our sampling plan
    becomes, and the better we are able to "pinpoint" the condition of the beach.
    
    Is there a point at which increased precision,  and the associated expense, is wasted?  The ultimate
    purpose of sampling is to determine if a beach is safe for recreational use.  Measurements of indicator
    organisms help to make this determination through what we know about their relationship to human
    health, as given in Equations 2 and 3. Thus, uncertainty in water quality assessment should be viewed
    not only in terms of sampling precision, but also in terms of the precision of our knowledge of health
    effects.
    
    The +/-0.2 log tolerance interval that we have described corresponds to a 90% confidence interval for
    the estimated mean log density, because of the 5% tail regions at each end of the interval. This spread
    for log densities was shown to be equivalent to +/-2  HCGI cases per 1000 swimmers. Available data
    from the freshwater health studies [Dufour, 1984] indicate a 90% confidence interval for the expected
    health effect of about +/-4 HCGI cases per 1000 swimmers at 126 E. coli per 100 mL (this will be
                                                60
    

    -------
    dependent to a degree on the actual indicator density). Thus, the uncertainty due the estimated density
    is substantially less than the uncertainty in its health effect, at least based on the EPA studies alone.
    In fact, the square of the total uncertainty will be equal to the sum of the squares of the individual
    components, leading to the conclusion that the total uncertainty will increase by only 12%, from
    4 to 4.5, when the sampling error alone is equivalent to +1-2 illnesses. For marine waters [Cabelli,
    1983], the 90% confidence interval for the expected illness rate is about +/- 3 HCGI cases per 1000
    at  35  enterococci per  100 mL.  In this case, a sampling error equivalent to +1-2 illnesses  per 1000
    contributes about 20% over the uncertainty in the health effect itself.
    
    Another component of uncertainty in recreational water quality assessment unfortunately comes into
    play when we consider the 24  hour lag in information from a membrane filtration assay.  Using
    ancillary data collected at the study beaches, we were able to achieve a meaningful  reduction in the
    standard deviation of the change in indicator density (Table 22).   However, the residual standard
    deviation, which represents the uncertainty in our ability to project today's values based on what they
    were yesterday, is still on the order 0.4 logs. This is equivalent to +/- 6 illnesses per 1000 swimmers for
    a 90% confidence interval, larger than even the uncertainty in the health effect itself, and underscores
    the need for refined predictive models and rapid methods for assessing water quality.
    
    Optimizing the design for obtaining data
    
    We have described appropriate spatial and temporal boundaries to encompass sampling from within
    a single depth zone (say knee- to chest-deep water), at a point in time (the morning before), from the
    same  depth below the surface (0.3 M).  The only difference among samples, then, is the  locations
    (transects)  along the beachfront from which they have been collected.   The appropriate  sampling
    variance for such a design is given by  "total variance between  locations" in Table 20 as experienced
    in  each of the individual EMPACT study beaches.
    
    One possible specification for tolerance requirements of a sampling was given in the previous section,
    mainly, that the sampling plan should be capable of correctly classifying the beach 95% of the time
    when the true  mean indicator density at the beach is 0.2 logs  removed in either direction  from  our
    action level. This was derived from a consideration of the implied health effects at these  indicator
    densities specifically for a freshwater beach, where +/-0.2 logs  is equivalent to +1-2 illnesses (HCGI)
    per 1000 swimmers.  In the case of marine beaches,  a 0.2 log increase in indicator (enterococci)
    density has a somewhat greater predicted effect  on illness due to exposure, being equivalent to an
    increase of 2.5 illnesses per 1000 swimmers per Equation 3.  However, rather than basing freshwater
    and marine designs on equivalent illness rates, we will use the  +/-0.2 log range for each so that each
    has the same precision with respect to the decision variable  itself.  In practice, one might choose to
    consider the same marginal illness rates for each type of beach, or, perhaps, choose  a wider interval
    for marine beaches, given that their expected illness rate is more than twice that of freshwater beaches
    at existing recommended indicator levels.
    
    Sample sizes in Table 23, the number of samples to be collected in a single visit, are based on the
    respective sampling variances within depth zone at each of the study beaches using the tolerance
    requirements of a 95% probability for correctly identifying +/-0.2 log  deviations from the action
    level. In addition, we show sample size requirements for +/-0.3 log deviations (equivalent to a health
    effect of about 3 illnesses per 1000 swimmers in freshwater and 3.7 illnesses per 1000 swimmers at
    marine beaches).  Note that, because the tolerance interval is expressed in terms of a log scale, the
    effect is that, in terms of the raw indicator densities, the lower and upper limits of this interval will
    
                                                 61
    

    -------
    be a constant factor of the action level.  For instance, 0.2 logs over the action level is about 1.6 times
    the action level and the log action level -0.2 logs is about 1/1.6, or about 0.6, times the action level.
    Likewise, +0.3 logs and -0.3 logs from the log of the action level are equivalent to twice and one-half
    the action level, respectively.
    
    Any given beach, after a routine monitoring program has been in place for some time, will have data
    approaching the  quality  we have from this study upon which to base a sampling program. At the
    outset, one necessarily will have to make assumptions about sampling variances. Results from the
    EMPACT studies may serve as a guideline to  some provisional estimate  of the appropriate sampling
    variances, but these provisional estimates will soon be superseded by experience. Because monitoring
    will be an ongoing process, once data is obtained this serves as a feedback to further data obtainment.
    The feedback addresses earlier steps in the DQO process as well, for example, denning the boundaries
    of the monitoring program, which may change as heretofore-unknown "hot spots" are discovered.
                       Table 23.  Sample   size   requirements   based  on
                                  estimated overall sampling variance for the
                                  geometric mean within a single depth zone
                                  in the water.
    Location
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach
    Miami Beach Park
    Sample size required for a
    95% tolerance limits of:
    +/-0.2 logs1
    6
    9
    9
    11
    14
    +/-0.3 logs1
    3
    4
    4
    5
    6
                       1 Specifies a range centered on the action level such that
                       there is a 95% probability of accepting the beach as safe for
                       swimming when the true log mean indicator density per 100
                       mL is at the lower end of the range and a 95% probability of
                       rejecting the beach when the true log mean indicator density
                       is at the upper end of the range.
    Because estimates of spatial variance are critical to evaluating precision and, subsequently, determining
    sample size requirements, samples should be analyzed individually during the initial phases of a
    monitoring program so that microbial density can be established for each location in the water from
    which a sample was collected. This procedure should be followed at least until enough samples have
    been collected over a number of days to enable a reliable estimate of the underlying variances. After
    that, one may consider compositing samples prior to analysis (see the box on composite  sampling),
    with the benefit of potential cost savings. Even  so, whole  samples should be assayed periodically in
                                                62
    

    -------
    order to monitor and update sampling variances.
    Sample size requirements, when individual samples are assayed, may be calculated using the same
    formula as was used to create Table 23. The sample size, n, required is given by
                                                   Ax
                                                                                          (4)
    Here, V is the variance within the zone, Ax is the half-width of the desired tolerance interval for Iog10
    indicator density per 100 mL, and zQ05 represents the upper 5th percentile of the standard normal
    distribution (=1.645). Again, the value of V will be a rough estimate at first, perhaps based on these
    EMPACT beach study results.  As more data are collected, a more appropriate value for V will be
    obtained.
      Given the common practice of collecting a single water sample at the beach, we are interested in
      knowing the sampling properties of a sample size of one. In this case, the only estimate of spatial
      variance would necessarily come from prior knowledge, since at least two samples are required
      in order to derive a variance estimate.  Corresponding tolerance intervals for a sample size of
      one are  shown for each of the study beaches are shown in Table 24, based on their respective
      variances within depth zone. The effects of more intensive sampling are illustrated by considering
      the width of this tolerance interval  as the sample size increases. Log indicator density tolerance
      intervals are translated into equivalent health effect tolerances in the second part of Table 24
      based on the relationships between incidence of HCGI among swimmers and indicator organism
      density given in Equations 2 and 3.
    
                          Table 24.  Width of tolerance intervals for various
                                    sample sizes based on sample variances
                                    within depth zone at the study beaches.
    Location
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach
    Miami Beach Park
    Location
    West Beach
    Belle Isle
    Wollaston Beach
    Imperial Beach
    Miami Beach Park
    Width of tolerance interval in Iog10 indicator
    density per 100 mL for sample size =
    1
    +/- 0.45
    0.60
    0.60
    0.68
    0.74
    2
    +/- 0.32
    0.42
    0.42
    0.48
    0.52
    4
    +/- 0.22
    0.30
    0.30
    0.34
    0.37
    8
    +/-0.16
    0.21
    0.21
    0.24
    0.26
    16
    +/-0.11
    0.15
    0.15
    0.17
    0.18
    Width of tolerance interval in terms of equivalent
    health effect (cases of HCGI per 1000 swimmers)
    for sample size =
    1
    +/- 4.3
    5.7
    7.4
    8.3
    9.1
    2
    +/-3.0
    4.0
    5.2
    5.9
    6.4
    4
    +1-2.1
    2.9
    3.7
    4.2
    4.5
    8
    +/- 1.5
    2.0
    2.6
    3.0
    3.2
    16
    +/-1.1
    1.4
    1.8
    2.1
    2.3
                                               63
    

    -------
    Composite sampling, while reducing analysis costs, will necessitate collecting additional samples
    beyond the requirements of Equation 4 in order to achieve the same precision in terms of a fixed
    tolerance interval.  There also must be consideration of the fact that a composite sample yields an
    estimate of an arithmetic mean rather than a geometric mean (the geometric mean being the antilog of
    the log mean which we have used thus far in describing sampling strategy).  See the boxed discussion
    on composite sampling and its effects on sample size requirements and equivalent action levels.
    
    Stratification along the shoreline
    
    Study areas limited to a beachfront of forty meters resulted in variances among locations within depth
    zone as shown in Table 20. In practice, the shoreline comprising a beach may be much longer than
    this, and division of the beach into "shoreline strata" may be indicated in order to maintain variance
    within each such stratum at levels comparable to those given in Table 20.  Stratification along the
    shoreline may be based on a priori considerations, for example,  the locations  of point sources of
    pollution into the beach area  or of physical barriers that may serve to  isolate an area of the beach
    from contamination.  Stratification also may be decided upon as a result of continued sampling that
    reveals systematic differences between different sections of beachfront.  In any event, strata need not
    be contiguous - an area affected by drainage may comprise one stratum while the expanses of beach
    on each side together comprise another single stratum.
    
    Table 21 shows there to be no substantial differences among standard deviations within different depth
    zones.  Because these standard deviations pertain to log densities,  this tells us that the coefficient of
    variation, or "percent standard deviation," is roughly the same for each zone. Assuming that the same
    holds true for shoreline strata, one would substitute a common (log) variance for V in Equation 4 in
    calculating the total sample size requirement.  The total sample size would then be allocated among
    the various shoreline strata in proportion to the total length of each respective shoreline. If composite
    sampling is to be used,  only  samples collected  from the  same stratum (/'. e., segment of shoreline)
    should be composited, and the results from different strata combined via a weighted geometric mean
    (antilog of the weighted mean of log densities), the weights being the proportion of beach represented
    by each stratum.
      Large systematic differences may warrant special consideration.  Depending on the magnitude of
      the difference, one may want to limit sampling to known "problem areas." For example, in the
      knee-deep zone at Belle Isle, densities between transects 2 and 3 were observed to be over twice
      that of transects 1. If this holds true for waist-deep water as well, we would limit sampling to
      the area within transects 2 and 3.  Cut this difference by one-half, and we would be inclined to
      combine all three transects.
                                                64
    

    -------
                                  Acceptance sampling
    
    EPA's guidance  document on the  DQO process discusses the decision process, and
    decision errors in particular, in terms of a "decision performance curve", which describes
    the probability of deciding that the decision parameter exceeds some allowable level based
    on a sample given the true value of that parameter.  Similar in concept is the operating
    characteristics (OC) curve, an example of which is shown in Figure 18.
    
                        Figure 18.  Operating characteristics curves
    1 ill}
    ^^sT^Sv
    n 0*1 ^ m. i
    v.ys \ rtyDiluct
    n on -\ \
    u.yu j \
    a> \ \
    0 n on \ .
    M W.BU
    S« \ \
    — j \ \
    ji n 7n \ .
    g^ u. /u
    *> \\
    u \
    ?j n an V
    W lu
    *S n in '
    O U.SU
    &
    j n .in
    2
    «n in
    U.JU
    £
    i_ n ?n
    CM
    Oin
    .IU
    On^.
    Ofift
    
    . »F«'»l —^=6 -• N=20
    
    
    
    
    \\
    •\
    s
    \\
    \ ^x>'*"-«*. 1^ 1 • i
    
    
    •uu 1 1 1 1 1 1 1 1
    10 20 30 35 40 50 60 70 80 90 100
    Enterococci per 100 mL
    For every possible value that the true geometric mean indicator density at a beach might be (on the
    horizontal axis), the OC curve indicates the probability of not declaring beach to be unacceptable
    (the vertical axis) based  on a sample result. The producer's point and the consumer's point,
    as labeled on the OC curve, are concepts of acceptance sampling, a  statistical quality control
    technique used in industry to accept or reject a lot (batch) of some product based on a sample. The
    consumer's point specifies the probability of accepting a "bad" lot, and the producer's point, the
    probability of accepting a "good" lot.  Usually, "good" and "bad" are defined in terms of percent
    defective. When this percent is truly some specific, acceptably low value, we want the probability
    of acceptance to be high (commonly, 95%), thus specifying the producer's point. Conversely, for
    some unacceptably high value for percent defectives, we want the probability of acceptance to
    be low (e.g., 5%), giving the consumer's point. The producer's risk, corresponding  to the Type I
    (a) error, is  100% minus the probability of acceptance at the producer's  point, and the consumer's
    risk, the Type II or /3 error, is the probability of acceptance at the consumer's  point.
                                             65
    

    -------
                                 Composite Sampling, Part I
    
    Composite sampling is a commonly used technique in environmental monitoring [Lancaster and Keller-
    McNulty,  1998] and may be a cost-effective practice for monitoring recreational water quality.  For our
    purposes, we will define composite sampling as follows.
    
      Composite sampling involves thoroughly mixing together equal volumes from various samples collected
      from different locations in the water and subsequently assaying a subsample of the resulting mixture.
      For example, ten 100 mL samples are combined and mixed, and a single 100 mL aliquot of this 1 L
      mixture is selected for analysis.  Two or more composites may be formed from a like number of distinct
      sets of samples, such as those drawn from knee-deep and those drawn from chest-deep waters and
      combined to create knee-deep and chest-deep composites.
    
    However,  an indicator density obtained from a composite sample  differs in a fundamental way from the
    geometric mean of the individual samples; namely, a composite sample result will almost always be higher.
    This is purely a mathematical artifact arising from the fact that, for any series of non-negative numbers, the
    geometric mean is always less than the arithmetic mean (unless all the values are  identical).  Consider the
    following hypothetical counts obtained from five 100 mL samples:
    
                                            11,24,68,56,16
    
    The geometric mean of this series is 27.6.  If, instead, these samples had been combined and a single 100
    mL of the resulting composite assayed, the expected count  would the arithmetic mean of this series, 35.
    Of course, because of additional sampling variation introduced by subsampling, the actual count may be
    greater or less than 35; however, there is a 90% probability that the count will be greater than the geometric
    mean, 27.6 (we know this from the Poisson distribution, which describes the distribution of possible counts
    from "well-mixed" samples).
    
    The use of the geometric mean is based on well-founded principles, as stated elsewhere in this report. There
    is, however, a correspondence between geometric and arithmetic means that results if densities among the
    original samples follow a lognormal distribution, and we have shown that our data is reasonably described
    by such a distribution. Given lognormal data, a geometric mean estimates the median of those data (actually,
    the geometric mean is biased on the high side, but this bias becomes negligible as the sample size increases).
    The median of a lognormal distribution is related to its mean by:
                                    Median =        x i o~ ' • ' s' " (4)
    
    Here, V is the variance of the Iog10 densities, as in Table 20 for example. Multiplying the count per 100 mL
    obtained from a composite sample by the factor 10 •' 15V yields an estimate that is approximately equivalent
    to that of the geometric mean of the individual samples. Alternatively, one could use the inverse relationship
    to express a mean in terms of the corresponding median, and inflate the decision level for the indicator
    density accordingly.  In that case, the equivalent decision level for a composite sample mean becomes 35 x
    10 L15V enterococci or 126 x 10 L15V E.  coli per 100 mL.
    
    Note that the preceding assumes we know the value of V. We also need to know this variance in order to
    properly evaluate precision of the composite sample estimate, as discussed in Composite Sampling Part
    II.  The appropriate value for V can be known only from sufficient historical  results based on assays of
    individual samples. We recommend that this variance estimate be based on individual analyses of at least
    50 samples (from each depth zone when multiple zones are sampled); this precludes composite sampling
    during an initial start-up phase of a monitoring program.  More samples may be required for an adequate
    estimate of V if the sampling variance is greater than about 0.3 for the Iog10 densities. Additionally, assays
    of individual samples should be performed periodically, on the order of every 2-4 weeks, thereafter in order
    to verify and update this variance estimate.
    
                                                  66
    

    -------
                              Composite Sampling, Part II
    
    Having addressed the problem of non-equivalence between geometric means and arithmetic means, we
    must face another problem inherent in composite sampling.  As briefly mentioned in Part I, subsampling
    from a composite adds another layer of estimation; a single aliquot is used to estimate the indicator
    density over all samples, which, in turn, are used to estimate the indicator level at the entire beach. The
    extra step in the estimation process, naturally, adds additional sampling error to the final estimate.  This
    may be illustrated using the example in Part I; the expected value of the composite count is 35, which is
    the average of the individual counts, but, 95% of the time, will range between 23 and 47.
    
    However, what about collecting more samples than we otherwise would have and then compositing these
    samples?  Say, in the above example, we collect ten samples instead of five and take a 100 mL composite
    from these ten samples. Certainly, ten samples are superior to five, and this may offset the additional
    uncertainty of subsampling from a composite.
    
    In order to evaluate sampling errors resulting from compositing samples compared to that of complete
    analysis, simulation studies were performed under the assumptions that (1) individual 100 mL samples
    follow a lognormal distribution with known variance of log densities, and (2) samples are composited
    and well-mixed so that the number of CPU captured in a 100 mL aliquot follows a Poisson distribution.
    Conclusions from this, in terms of sample size  requirements with and without composite sampling, are
    shown below. Note that the additional sampling required when compositing is utilized increases rapidly
    when variance exceeds about 0.20 or so - another reason for stratifying the beach front into relatively
    homogeneous sections.
    
                Table 25. Comparison of sample size requirements for equivalent
                         precision with and without compositing.
    Sampling variance1
    0.10
    0.15
    0.20
    0.25
    0.30
    0.40
    0.50
    Number of samples to collect for equivalent precision
    Individually
    analyzed
    3
    5
    6
    8
    9
    12
    15
    Composited
    5
    8
    9
    14
    17
    32
    43
                1 Variance of log (indicator density per 100 mL)
                                               67
    

    -------
                                           Summary
    Several factors have been identified in this study as correlates of microbial indicators in recreational
    water.  These include:
    
        •  Spatial factors
               o   Zone - Indicator levels decreased the farther one goes from shore into deeper water
                   (not to be confused with the depth from which a sample is taken, which was not found
                   to be an important factor in this study)
               o   Distance along the shoreline ("transect") - There were only minor systematic effects
                   at two of the five beaches that were monitored for this study.
        •  Temporal factors
               o   Morning to afternoon (9:00 a.m. to 2:00 p.m.) - Indicator levels generally decreased
                   by the afternoon at three of the beaches (2 marine and one freshwater).  Afternoon
                   levels at the other freshwater beach tended to be lower on sunny days, but higher on
                   overcast days.  Only Imperial Beach in Southern California, where indicator densities
                   were very low at all times, failed to show this effect.
               o   Fecal indicator levels were found to vary significantly from day to day.  There was a
                   limited statistical relationship observed between levels on the sampling day and the
                   following day at three out of the five beaches studied.
        •  Environmental & bather factors
               o   Rain - Resulted in increased indicator levels at the four beaches where substantial
                   rainfall occurred (there was very little rain at Imperial Beach). The relationship may
                   be complicated; at West Beach, there was no apparent effect unless rainfall in the past
                   days was considered, in contrast to the others, where rainfall in the past 24 hours was
                   sufficient.
               o   Wind - Associated with increased indicator levels when blowing onshore at three of
                   the beaches. No effects were observed at Belle Isle, and Imperial Beach experienced
                   too little variation, as its winds were constantly onshore.
               o   Cloud cover - Under mostly sunny  conditions, both freshwater beaches tended to
                   have lower indicator densities.  This effect is in addition to its association with the
                   morning to afternoon change at West  Beach, mentioned above, since levels were
                   lower on sunny mornings also.  At Miami Beach Park, the estuarine beach, sunshine
                   tended to  enhance the decrease that was observed in the afternoon.
               o   Tides - Tidal effects  on water quality were observed. Absolute water level, rather
                   than the tide stage per se, was  found to be a predictor of water quality, high water
                   being associated with increased microbial counts at the two East Coast beaches (very
                   much so at Miami Beach Park), and with lower counts at the West Coast beach.
               o   Temperature - Only  Miami Beach Park yielded a relationship with  air or water
                   temperature. The tendency here was for counts to be lower with increased water
                   temperature.  While this may  seem  counterintuitive, this relationship does  not
                   necessarily imply causation, and may indicate  the  common  influence of some
                   unmeasured factor, such as  deep-water currents,  on both water temperature and
                   indicator levels.
               o   Bather density - Lower bather  density was associated with lower indicator density
                   at two of the beaches, but higher density at another.  However, relationship of bather
                   density to microbial counts may be influenced by conditions that bring bathers to the
                   beach (or keep them away) in the first place.
                                                68
    

    -------
    Spatial and temporal components of variation have been identified and estimated. These are:
    
        •  Spatial sources of variation
               o   Replicate variance ("Pure error") - A factor at any level of analysis that represents the
                   smallest attainable variation in sampling (standard deviation=0.235).
               o   Sampling depth - Adds to sampling variability, but is not a constraining factor, since
                   all samples should be taken from the same depth below the surface.
               o   Variance between zones -  Would  add considerably to  sampling  variance  if not
                   accounted for in the design. This variability can be eliminated by restricting sampling
                   to water of the same depth.
               o   Variance within zone (variance between transects) - The key component of variance
                   for a point-in-time estimate. May be able to  reduce this variance if the beach can be
                   stratified along its length into "hot" and "clean" regions.
        •  Temporal sources of variation
               o   Variance among days
                       •   Equivalent to the standard  deviation that is referred to in the EPA criteria
                          document [Dufour and Ballentine, 1986] in its recommendations for single-
                          sample limits.
                       •   Would be the principal source  of variability in a sample estimate over time
                          for the same location at the beach each time, e.g. a 30-day geometric mean.
                       •   Not a source of sampling variability for a point-in-time estimate, but must be
                          considered when using such an estimate to project results to future days.
               o   Hourly variation - based on hourly sampling that was performed over several days at
                   each beach, hour-to-hour variability was not much less than day-to-day variability.
    
    The following points would be  included in a protocol for sampling recreational  waters for indicators
    of fecal contamination:
    
        •  Time and location of sampling must be carefully considered.
               o   Depth zones from which samples are collected are likely to  have great effect on the
                   resulting estimate of indicator density.  Sampling in knee- to  waist-deep water would
                   seem to offer a reasonable, but still conservative, approach to estimating water quality,
                   particularly given that health effects are based on quality of water at waist-depth or
                   deeper.
               o   Sampling at 0.3 M below the surface is justified based on exposure considerations.
                   This study failed to discern differences  at lower collection depths.
               o   Sampling in the morning will likely be a conservative practice, in addition to perhaps
                   being a convenient time to sample.
               o   Sampling should be performed as close as practical to the day on which a decision is
                   to be made regarding beach closure or advisement. Preferably this should be the day
                   before, given current conditions of a one day turnaround for the results.
        •  A number of samples should be collected from different points in knee-  to waist-deep water.
               o   Results from the EMPACT Beaches Study can be used as a guideline for the initial
                   determination of sample size requirements.  For the beaches in this  study, sample
                   sizes of from 3 to 6 would be adequate to allow for 95% certainty of detecting a 0.3
                   log exceedence from the action level for the geometric mean, equivalent to a health
                   risk of 3 to 3.5  cases of HCGI per 1000 swimmers.
               o   As data are gathered, the sampling plan should be refined and ultimately  based on a
                   variance estimate that is uniquely associated with the subject beach.
    
                                                69
    

    -------
    "Hot spots" within the bathing area, which may be known a priori or discovered as a result of
    sampling, should be considered as separate strata for sampling purposes.
        o  These should be sampled independently and weighted appropriately in the  final
           result.
        o  If indicator density within the problem area is very different from that elsewhere,
           sampling  should be  limited to this  area. A proposed rule-of-thumb is a two-fold
           difference.
        o  Lesser differences would warrant a combined, stratified estimate, weighting the "hot
           spot" in proportion to its extent relative to the rest of the beachfront.
    Composite sampling may be used as  a cost-efficient technique, enabling better sample
    coverage at minimally increased cost.
        o  In the initial stages of a monitoring program, composite sampling should not be used
           in order to  develop  data that is necessary  in  estimating the appropriate  sampling
           variance for a particular beach.
        o  A composite sample estimates an arithmetic mean, which would require adjustment
           in order to equate this to standards based on a geometric mean.
    If resources are available to collect the data necessary for developing a predictive model for
    the change in indicator density over a 24 hour period, this will likely result in much-improved
    assessment of water quality, given the 1-day lag in obtaining results from membrane filtration
    assays.
                                         70
    

    -------
                                         References
    
    American Public Health Association, American Water Works Association, Water     Environment
    Federation.  1998. Standard methods for the examination of water and    wastewater, 20th edition.
    American Public Health Association, Washington, DC.
    
    Bartram, J. 1999.  Health-based monitoring of recreational waters: The feasibility of a new
    approach (the 'Annapolis Protocol'), WHO/SDE/WSH/99.1, World Health Organization,Geneva.
    
    Bordner, R., Winter, J.  and Scarpino,  P.   1978.   Microbiological methods for  monitoring the
    environment:  water and wastes.   EPA-600/8-78/017, Environmental  Monitoring  and  Support
    Laboratory, Office of Research and Development, U.S. Environmental Protection Agency,  Cincinnati,
    Ohio.
    
    Brenniman, G.R., Rosenberg, S.H. and Northdrop, R.L.  1981. "Microbiological sampling variables
    and recreational water quality standards," Am. J. Public Health, 71(3):283-289.
    
    Cabelli, V.J. 1983. Health effects criteria for marine recreational waters.  EPA-600/1-80-031, U.S.
    Environmental Protection Agency, Research Triangle Park, NC.
    
    Cabelli, V.J., A.P. Dufour, L.J. McCabe, and M.A. Levin. 1982. "Swimming-associated gastroenteritis
    and water quality, "Am. J. Epidemiol, 115:606-616.
    
    Cabelli, V., Levin, M., Dufour, A.  and McCabe, L. 1974. "The development of criteria for
    recreational waters,"  Proc. International Symposium on Discharge of Sewage from Sea Outfalls,
    Gameson, A. (editor), Pergamon Press, London/New York, p. 63.
    
    Cheung, W.H.S., Chang,  K.C.K., and Hung, R.P.S.  1991.  "Variations in microbial indicator densities
    in beach waters and health-related assessment of bathing water quality," Epidemiol. Infect. 106(2): 329-
    44
    
    Churchland, L.M., Kan, G. and Ages, A.  1982. "Variation in fecal pollution indicators through tidal
    cycles in the Fraser River estuary," Can J Microbiol. 28(2):239-47
    
    Cochran,W.G.  1977. Sampling Techniques, 3rd Ed.. Wiley, NY.
    
    Code of Federal Regulations. 2000. Protection of environment. Code of Federal  Regulations, Title
    40, Part 136.3. pp. 27-29.
    
    Dufour, A.P.  1984. Health effects criteria for fresh recreational waters, EPA-600/1-84-004, Office of
    Research and Development, U.S. Environmental Protection Agency, Cincinnati, OH.
    
    Dufour,A.P andBallentine, P. 1986. Ambient water quality criteria for bacteria- 1986 (Bacteriological
    ambient water  quality  criteria for  marine and fresh recreational  waters). EPA A440/5-84-002,
    Washington, DC.
    
    Fatal, B., Vasl, R., Katzenelson, E. and Shuval, H.  1983.  "Survival of  indicator bacteria organisms
    and enteric viruses," Water Research,17:379.
    
                                                71
    

    -------
    Fleisher, J. M. and McFadden, R.T.  1980. "Obtaining precise estimates in coliform enumeration,"
    Water Research, 14:477-483.
    
    Fujioka, R., and Narikawa, O. 1982. "Effect of Sunlight on Enumeration of Indicator Bacteria
    Under Field Conditions," Appl. Environ. Microbiol., 44:395.
    
    Gannon, J.J. and Busse, M.K. 1989. "E. coli and enterococci levels in urban stormwater, river water
    and chlorinated treatment plant effluent," Water Research, 23(9): 1167-1176.
    
    Lancaster, V.A. and Keller-McNulty, S.  1998. "A review of composite sampling methods," J. Am.
    Statistical Assoc., 93(443): 1216-1230.
    
    Messer, J.W. and A.P. Dufour. 1998.  Arapid, specific membrane filtration procedure for  enumeration
    of enterococci in recreational water. Appl. Environ. Microbiol., 64:678-680.
    
    Mujeriego, R., Bravo, J. M., Pinas, M. and Sanchez Murias, B.  1980.  "Statistical variation of
    microbiological quality of coastal waters: regulatory  implications," Ves Journees Etud. Pollutions,
    1980:49-60.
    
    National Technical Advisory Committee.  1968. Water Quality Criteria. Report on the National
    Technical Advisory Committee, Washington  DC, Federal Water Pollution Control Administration,
    p. 7.
    
    Noble, R.T., Leecaster, M.K., McGee, C.D., Moore, D.F., Orozco-Borbon, V., Schiff, K., Vainik,
    P.M. and Weisberg, S.B.  2000. Southern California Bight Region Monitoring Program: Storm
    Event Shoreline Microbiology. Southern California Coastal Research Project., Westminster, CA.
    
    Olivieri, V., Kruse, C., Kazuyoshi, K. and Smith, J. 1977. Microorganisms in urban stormwater.
    EPA-600/2-77-087, U.S. Environmental Protection Agency, Cincinnati, OH.
    
    Pande, J., Das, S. M. and Misra, D. S. 1983. "Studies on bacteriological quality of Naintal Lake
    water in different seasons," Water, Air, and Soil Pollution, 19:149-154.
    
    Public  Health  Laboratory  Service  Water  Surveillance  Group.   1995.   "Preliminary  study of
    microbiological parameters in eight inland recreational waters," Letters in Appl. Microbiol.,
    21:267-271.
    
    Sekla,  L., Williamson,  D.,  Greensmith, C.,  Balacko,  G.,  Brown,  D. and Stackiw, W.  1987.
    "Bacteriological characteristics of 15 freshwater beaches in Manitoba," Canadian J. Public Health,
    78(3):181-184
    
    Serrano, E., Moreno, B., Solaun, M., Aurrekoetxea, J.J. and Ibarluzea, J.  1998.  "The influence of
    environmental factors on microbiological indicators of coastal water pollution," Water Sci. Tech.,
    38(12):195-199.
    
    Seyfried, PL.  1973.  "Sampling bacteria in  Lake Ontario and the  Toronto Harbour,"  Proc.  16th
    Conference on Great Lakes Research, Internal. Assoc. Great Lakes Res., 163-182.
                                                72
    

    -------
    Sieracki, N. 1980. "The effects of short exposures of natural sunlight on the decay rates of enteric
    bacteria and a colipahge in a simulated sewage outfall microcosm,"  Master of Science Dissertation,
    Univ. of Rhode Island.
    
    Smith, B., and Dufour, A. 1993 "Effects of swimmers on the microbiological quality of recreational
    waters: a simulation study," Abst. Annual Meeting of ASM, p. 406.
    
    Stevenson, A.H. 1953. "Studies of bathing water quality and health," Am. J. Public Health, 43:529.
    
    U.S. Environmental  Protection Agency.   1986.   Bacteriological  ambient  water quality criteria
    availability.  Federal Register, 51(45): 8012.
    
    U.S. Environmental Protection Agency. 1997a. Manual for the certification of laboratories  analyzing
    drinking water:  criteria and procedures, quality assurance, 4th edition.  EPA-815-B-97-001, Office of
    Ground Water and Drinking Water, U.S. Environmental Protection Agency, Cincinnati, OH.
    
    U.S. Environmental Protection Agency.  1997b. Method 1600:  membrane filter test methods    for
    enterococci in water.  EPA-821/R-97/004a, Office of Water, Washington, DC.
    
    U.S. Environmental Protection Agency. 2000a.  Improved enumeration methods for the   recreational
    water quality indicators: enterococci and Escherichia coli. EPA/82l/R-97/004,    Office of Science
    and Technology, Washington, DC.
    
    U.S. Environmental Protection Agency.  2000b.  Guidance for the data quality objectives process,
    EPA/600/R-96/055, Office of Environmental Information, U.S. Environmental Protection Agency,
    Washington, DC.
    
    U.S. Environmental Protection Agency. 2002. National beach guidance and required performance
    criteria  for grants, EPA-823-B-02-004, Office of Water, U.S. Environmental Protection Agency,
    Washington, DC.
    
    Van Ess, E. and Harding, A. K. 1999. "A bacteriological study of swimming areas in Western Oregon,"
    Environ. Health, 14-19.
    
    World Health Organization. 1975. Guide and criteria for recreational quality of beaches and coastal
    waters. Bilthoven.
    
    World Health Organization.  1977. Health criteria and epidemiological studies related to coastal water
    pollution,. Athens.
    
    World Health Organization.  1998. Guidelines for safe recreational-waters environments: coastal and
    fresh waters. Consultation draft. World Health  Organization, Geneva.
                                                73
    

    -------
    vvEPA
    United States
    Environmental Protection
    Agency
      Office of Water and Office of Research and Development
      National Exposure Research Laboratory
      Cincinnati, OH 45268
    
      Official Business
      Penalty for Private Use
      $300
    
      EPA/600/R-04/023
      August 2005
    

    -------