ARTICLE IN PRESS

Water Research 39 (2005) 5250–5260

www.elsevier.com/locate/watres

Nowcast modeling of Escherichia coli concentrations at

multiple urban beaches of southern Lake Michigan
Meredith B. Nevers, Richard L. Whitman
US Geological Survey, Great Lakes Science Center, 1100 N. Mineral Springs Road, Porter, IN 46304, USA

Received 1 June 2005; received in revised form 9 September 2005; accepted 11 October 2005

Abstract

Predictive modeling for Escherichia coli concentrations at effluent-dominated beaches may be a favorable alternative
to current, routinely criticized monitoring standards. The ability to model numerous beaches simultaneously and
provide real-time data decreases cost and effort associated with beach monitoring. In 2004, five Lake Michigan beaches
and the nearby Little Calumet River outfall were monitored for E. coli 7 days a week; on nine occasions, samples were
analyzed for coliphage to indicate a sewage source. Ambient lake, river, and weather conditions were measured or
obtained from independent monitoring sources. Positive tests for coliphage analysis indicated sewage was present in the
river and on bathing beaches following heavy rainfall. Models were developed separately for days with prevailing
onshore and offshore winds due to the strong influence of wind direction in determining the river’s impact on the
beaches. Using regression modeling, it was determined that during onshore winds, E. coli could be adequately predicted
using wave height, lake chlorophyll and turbidity, and river turbidity (R2 ¼ 0:635, N ¼ 94); model performance
decreased for offshore winds using wave height, wave period, and precipitation (R2 ¼ 0:320, N ¼ 124). Variation was
better explained at individual beaches. Overall, the models only failed to predict E. coli levels above the EPA closure
limit (235 CFU/100 ml) on five of eleven occasions, indicating that the model is a more reliable alternative to the
monitoring approach employed at most recreational beaches.
Published by Elsevier Ltd.

Keywords: Indicator bacteria; Beach closures; Water quality; Monitoring; Indiana

1. Introduction conditions impact Escherichia coli concentrations in

Efforts to maximize recreational beach-monitoring
accuracy and efficiency have been increasing in recent
years. Many criticisms of the developing alternate
techniques, whether rapid testing or source tracking,
include lack of accuracy, low efficiency, and high cost
(Stoeckel et al., 2004). Nevertheless, a promising
technique is predictive modeling. Numerous ambient
Corresponding author. Tel.: +1 219 926 8336;
fax: +1 219 929 5792.
E-mail address: mnevers@usgs.gov (M.B. Nevers).
natural waters, including sunlight (Fujioka et al., 1981;
Davies-Colley et al., 1994; Whitman et al., 2004),
rainfall (Ackerman and Weisberg, 2003; Morrison et
al., 2003), wind speed and direction (Smith et al., 1999;
Olyphant and Whitman, 2004), wave height (LeFevre
and Lewis, 2003), tidal period in marine waters
(Crowther et al., 2001; Boehm et al., 2002), and turbidity
(Olyphant et al., 2003). By assessing the relationship
between such parameters and E. coli mathematically, E.
coli counts could be determined using the alternate
parameters, thereby eliminating the need for collecting a
water sample and culturing for E. coli, a time-intensive

0043-1354/$ - see front matter Published by Elsevier Ltd.

doi:10.1016/j.watres.2005.10.012
 ARTICLE IN PRESS
M.B. Nevers, R.L. Whitman / Water Research 39 (2005) 5250–5260 5251

process that makes current monitoring protocols inade- more clearly defined. To this point, modeling exercises
quate. Further, modeling may be used more frequently, have typically included a single study beach, but with
thereby making it more suitable for use in systems with effluent dominating along several kilometers of shore-
high E. coli variation over short periods of time or space line, it is believed that numerous beaches can be modeled
(Boehm et al., 2002; Whitman and Nevers, 2004). The simultaneously. Finally, in this study, point source
advantages of a predictive model include the ability to influences were characterized by examining samples for
use readily available hydrometeorological data (e.g., coliphage, a more certain indicator of sewage contam-
airport weather stations, stream gauges), shorter time ination.
for results, and cost-effectiveness. A sufficient database
for a given beach area would be required before
predictive modeling could be used regularly, but with
2. Materials and methods
this information, it is a promising alternative for beach
monitoring.
2.1. Study location
Fecal indicator bacteria standards were developed to
alert beach managers of sewage contamination in
The beaches in Porter and Lake Counties, Indiana,
recreational waters, which would prompt them to post
west of the Burns Ditch outfall are periodically subject
warnings or close the beaches, depending on the state’s
to beach closures due to high counts of the indicator
determination (US EPA, 1986). Recent evidence com-
bacteria E. coli. Mean E. coli levels at these beaches have
plicates the use of indicators because numerous potential
been comparable year to year, with periodic elevated
non-point sources of E. coli in natural environments
counts.
have been found, including soils (Byappanahalli et al.,
Burns Ditch is the outfall of the Little Calumet River
2003), sands (Obiri-Danso and Jones, 2000; Alm et al.,
into southern Lake Michigan. The creation of Burns
2003; Whitman and Nevers, 2003), and beach algae
Ditch between 1924 and 1926 reversed the flow of the
(Whitman et al., 2003), that may interfere with an
Little Calumet River west of the ditch and facilitated
assessment of sewage presence. It is therefore important
flow in the east branch so both branches flowed out to
also to determine the source of contamination and the
the lake. The river dominates the Lake Michigan
associated health risks.
watershed of Northwest Indiana, draining three counties
Beaches prone to sewage contamination are typically
that include five waste water treatment plants with
situated near an outfall (natural or storm drains) and are
periodic combined sewer overflows and a wide variety of
subject to flushes containing high counts of E. coli and
non-point sources from agricultural and forested areas
other bacteria. Storm water and storm drains (Boehm et
(Olyphant et al., 2003). The ditch flows into southern
al., 2003; McLellan and Salmore, 2003) and river or
Lake Michigan just east of five swimming beaches that
creek outfalls (Olyphant et al., 2003) transport water
extend 10 km: Ogden Dunes, West, Wells Street, Lake
similarly high in E. coli concentration. The originating
Street, and Marquette Beaches (Fig. 1). Research has
sources are numerous, and may include urban runoff,
shown that it contributes significant amounts of E. coli
agricultural runoff, watershed flushing, or combined
and associated bacteria to southern Lake Michigan,
sewer overflows. These types of beaches, which may be
which may increase to more than 10,000 CFU/100 ml for
termed effluent dominated, may be more suited to
several hours during a storm event (Olyphant et al.,
modeling because the characteristics of the outfall often
2003).
drive the nearshore beach-water quality. Past modeling
attempts in southern Lake Michigan have included
modeling single beaches and modeling point source 2.2. Sampling
outfalls, but studies linking the two have been limited.
In this study, we examined the relationship between Water samples were collected 7 days a week between
water quality of a river outfall and five beaches June 21 and August 15, 2004. There were 13 sampling
immediately to the west and the potential for modeling. locations, spaced evenly along each of the five contig-
These beaches are currently monitored by three different uous beaches: three sites at Ogden Dunes, three sites at
management groups and each is individually closed to West, one site at Wells Street, four sites at Marquette,
swimming if a daily sample (or mean of multiple and two sites at Lake Street (Fig. 1). In addition, there
replicates) has an E. coli count above 235 CFU/ml, was a sampling location on Burns Ditch, at a marina just
according to Indiana’s water quality standards. By upstream of the outfall. Two water samples were
sampling both the monitored beaches and the outfall, collected at each location by immersing a sterile bottle
it could be established how periodic pulses of water (for E. coli) or a polyethylene bag (for water chemistry)
containing high concentrations of E. coli influence the below the surface in approximately 45-cm deep water.
subject beaches. Further, by monitoring several beaches Samples were placed in a closed cooler and held at 4 1C
7 days a week, temporal and spatial patterns could be until analysis. Field observations at each sampling site
 ARTICLE IN PRESS
5252 M.B. Nevers, R.L. Whitman / Water Research 39 (2005) 5250–5260
Fig. 1. Study area with all locations noted from which data were collected and parameters listed.
included air and water temperature, wind speed and
direction, and wave height.
A YSI 6920 multiprobe sonde was deployed in Burns
Ditch for the duration of the study. The instrument
recorded water temperature, specific conductivity, tur-
bidity, pH, dissolved oxygen, and water level every
15 min. Data from the instrument were downloaded
every 2 weeks when the probes were cleaned and
recalibrated.
Weather data were collected from a station located at
Washington Park on the lakefront in Michigan City,
Indiana, approximately 23 km northeast of Burns Ditch
(Fig. 1). Weather conditions collected at Indiana Dunes
National Lakeshore headquarters, located approxi-
mately 6 km southeast of Burns Ditch were also used.
Onsite wave conditions were recorded with an acoustic
Doppler current profiler (ADCP) operated by NOAA,
which was located near the outfall of Burns Ditch at a
depth of 9 m. Additionally, wave and weather conditions
generated by the NOAA Great Lakes Environmental
Research Laboratory were used, which incorporated
hydrometeorological data from Gary Regional Airport.
2.3. Laboratory analyses
In the laboratory, water samples were analyzed for E.
coli using the Colilert-18 method (American Public
Health Association, 1998), which provides results as
most probable number (MPN)/100 ml. The additional
water samples were analyzed in the laboratory for
specific conductivity (Fisher Scientific Acumet meter),
color (Hach spectrophotometer), and chlorophyll and
turbidity (Turner Instruments Aquafluor).
2.4. Statistical analysis
Statistics were conducted using the software programs
SPSS (SPSS, 2003) and Systat (Systat, 2004). SPSS was
used for analysis of variance and Pearson correlation
analyses. Euclidean distance was used to measure
similarity among beach sites, and those results were
displayed using multi-dimensional scaling. Because
multi-dimensional scaling is a visual interpretation of
the distance measures, the axes do not have units,
therefore, they are identified only as dimensions 1 and 2.
Empirical regression modeling using least-squares ana-
lysis obtained the parameters that best explain the
dependent variable (E. coli concentration). Stepwise
regression that retains independent variables at 90
percent confidence was also employed. Only coefficients
with Po0:05 were retained in the model. Separate
models were developed for days with onshore and
offshore prevailing wind components.
 ARTICLE IN PRESS
M.B. Nevers, R.L. Whitman / Water Research 39 (2005) 5250–5260 5253

3. Results Table 1
Correlation coefficients and significance levels for parameters
3.1. Burns Ditch and beaches compared to beach log E. coli and Burns Ditch log E. coli

Parameter Beach Burns Ditch

Over the course of the study period, E. coli densities E. coli E. coli
(mean log MPN/100 ml) were significantly higher for
Burns Ditch than the mean for the five study beaches Turbidity (NTU)-Lake 0.389** 0.203**
(F 5 ;438 ¼ 43:7, Po0:0001) and were positively correlated Color (PtCo units)-Lake 0.352** 0.233**
(Po0:001). E. coli counts in Burns Ditch were also Specific conductance (mg/l)-Burns 0.164** 0.039
positively correlated with each individual beach except Ditch
for Wells Street. E. coli counts at Lake Street were Dissolved oxygen (mg/l)-Burns Ditch
0.169*
0.229**
significantly higher than the other beaches. Multi- Water depth (m)-Burns Ditch 0.350** 0.345**
pH (pH units)-Burns Ditch
0.193**
0.191**
dimensional scaling revealed a pattern with Burns Ditch
Wave height (m)-near Burns Ditch 0.419** 0.317**
most dissimilar from the beaches (Fig. 2). Among Gauge height (m)-Burns Ditch 0.291** 0.462**
sampled beaches, Ogden Dunes was least like the other 4 h precipitation (cm)-Washington 0.167* 0.090
sampled beaches but grouped most closely with West Park
Beach, its nearest neighbor to the west. Wind speed (m/s)-Washington Park 0.308** 0.157**
Wind gust (m/s)-Washington Park 0.288** 0.148**
Relative humidity (%)-Washington 0.050 0.223*
3.2. E. coli and ambient conditions Park
Air temperature (1C)-Indiana Dunes
0.074 0.125*
Mean log E. coli densities for the five study beaches Wind speed (m/s)-Indiana Dunes 0.282** 0.252**
over time were significantly positively correlated with Relative humidity (%)-Indiana Dunes
0.077 0.126*
color, turbidity, wave height, Burns Ditch gauge height, Air temp (1C)-Portage GLERL 0.056 0.243**
Cloud cover (%)-Portage GLERL 0.075 0.115*
previous 4 h precipitation, wind speed, wind gust, and
Current speed (m/s)-Portage GLERL 0.202** 0.238**
specific conductance and significantly negatively corre- Dew point (1C)-Portage GLERL 0.102* 0.259**
lated with pH and dissolved oxygen. The highest Wave height (m)-Portage GLERL 0.271** 0.271**
correlation was with wave height (Pearson R ¼ 0:419) Wind speed (m/s)-Portage GLERL 0.304** 0.190**
(Table 1).
A comparison of E. coli counts in Burns Ditch with Only parameters with at least one significant correlation are
ambient conditions resulted in significant positive relation- included.  Po0:05,  Po0:01.

2 ships with lake color and turbidity, wind speed and gust,
wave height, Burns Ditch gauge height, air temperature,
and relative humidity. Significant negative correlations
were with dissolved oxygen and pH, as was the case at the
1 beach locations. The highest correlation was with Burns
Ditch gauge height (Pearson R ¼ 0:462) (Table 1).
OD
The relationship between rainfall and E. coli counts
on the beach was explored more closely, given past
Dimension-2

BD
research findings of high E. coli counts associated with
0 WB
LS close proximity to a point source outfall. There were
MQ only 7 days with greater than 0.5 cm rainfall in 24 h
WS during the study period, and there was typically a spike
in E. coli in both Burns Ditch and on the beaches
-1 following such an event. There were too few cases to
associate rainfall and E. coli levels at the beaches
statistically, although the bacteria levels for Burns Ditch
have previously been associated (Olyphant et al., 2003).
-2 The sources and means of transport, however, have not
-2 -1 0 1 2
been completely described.
Dimension-1

Fig. 2. Multi-dimensional scaling of similarity measurements 3.3. Models

for mean log E. coli counts at each of the five study beaches
(OD ¼ Ogden Dunes, WB ¼ West Beach, WS ¼ Wells Street, Due to the strong influence of wind direction, best-fit
MQ ¼ Marquette, LS ¼ Lake Street) and Burns Ditch (BD). models were developed for days with a predominant
 ARTICLE IN PRESS
5254 M.B. Nevers, R.L. Whitman / Water Research 39 (2005) 5250–5260

north wind vector, predominant south wind vector, and
all winds. It has been previously established (aerial
photography) that north winds force the outfall from
Burns Ditch toward the shoreline, presumably increas-
ing the impact Burns Ditch water has on these beaches.
Similarly, south winds force the outfall water northward
and away from the beaches. Because each beach is
managed separately in terms of leaving it open or closing
it to swimming, each was considered separately in
developing the model.
The model developed for prevailing north winds
incorporated the variables lake chlorophyll, lake turbid-
ity, wave height, and Burns Ditch turbidity (Table 2).
The variables with the greatest impact were wave height
and Burns Ditch turbidity, indicating that E. coli counts
were closely related to the Burns Ditch outfall (Fig. 3).
The multiple regression model used was
y ¼ B0 þ B1 X 1 þ B2 X 2 þ B3 X 3 þ;    ; þBx X x þ e,
where y is the concentration of E. coli as log CFU/
100 ml; B0 is a constant, B1, B2, B3,y,Bx are the
regression coefficients for the predictors wave height,
Burns Ditch turbidity, chlorophyll, and lake turbidity;
and e is residual error, and is written as
Log E: coli ¼ 0:441 þ 0:652ðwave heightÞ
þ 0:003ðBD turbidityÞ
þ 0:010ðchlorophyllÞ
þ 0:011ðturbidityÞ þ e.
Using the model, E. coli counts could be predicted
with 64% of the variance explained. Predicting E. coli
count relative to a beach closure included only three
type II errors, meaning that on three occasions the
model predicted that E. coli counts would be below the
235 single-sample limit when it was, in fact, higher than
that count. The model accurately predicted six closures
for the beaches during north winds, five of which
occurred on the same date, following a significant rain
event.
The model developed for prevailing south winds was
less effective at predicting E. coli counts. The regression
coefficients included were for the predictors wave height,
wave period, and total precipitation in the 4 h preceding
sample collection (Table 2). Wave height had the
greatest effect, but overall there was high variation, in
part due to the abundance of 0 readings for precipitation
(Fig. 3). The model for south winds is written as
Log E: coli ¼ 1:575 þ 0:765ðwave heightÞ
þ
0:131ðwave periodÞ
þ 2:944ðprecip_4 hÞ þ e:
This model could predict E. coli count with 32% of
the variance explained. It should be considered that only
two closure level events occurred during the study
period, however, neither of these was predicted using the
model. All of the times the E. coli was below 235 were
correctly predicted.

Table 2
Results for best model regressions developed for north, south, and combined north and south winds

B SEB t-ratio Significance Part Partial

North winds R2 ¼ 0:635

Constant 0.441 0.127 3.486 0.001
Wave height 0.652 0.107 6.067 o0.001 0.386 0.539
Burns Ditch turbidity 0.003 0.001 5.201 o0.001 0.331 0.481
Chlorophyll 0.010 0.002 4.865 o0.001 0.310 0.456
Lake turbidity 0.011 0.003 3.520 0.001 0.224 0.348

South winds R2 ¼ 0:320

Constant 1.575 0.147 10.715 o0.001
Wave height 0.765 0.133 5.769 o0.001 0.433 0.464
Wave period
0.131 0.032
4.113 o0.001
0.308
0.350
Precip_4 h 2.944 0.777 3.791 o0.001 0.284 0.326

North and south R2 ¼ 0:465

Constant 1.124 0.131 8.602 o0.001
Wave height 0.789 0.102 7.738 o0.001 0.397 0.397
Burns Ditch turbidity 0.003 0.001 4.038 o0.001 0.207 0.273
Chlorophyll 0.004 0.001 3.500 0.001 0.180 0.239
Lake turbidity 0.013 0.003 3.774 o0.001 0.194 0.256
Wave period
0.114 0.027
4.162 o0.001
0.214
0.280
Precip_4 h 3.220 0.732 4.400 o0.001 0.226 0.295

B (beta) ¼ regression coefficient; SEB ¼ standard error for B; t-ratio ¼ t-value for B; Significance ¼ two-tailed significance level of t;
Part ¼ part correlation; Partial ¼ partial correlation.

M.B. Nevers, R.L. Whitman / Water Research 39 (2005) 5250–5260
North
3.00
Log E. coli
Log E. coli
2.00
1.00
R Sq Linear = 0.635
0.00
0.50 1.00 1.50 2.00 2.50 3.00 3.50
Predicted Value
Fig. 3. Actual log E. coli count and predicted E. coli based on developed models for north and south winds. Lines on X and Y-axes
indicate 2.38, the EPA recommended single-sample limit for recreational waters.
The variables that best predicted E. coli counts
without subdividing by predominant wind direction
were also determined. The model included regression
coefficients for all variables included in the separate
wind models (Table 2). Wave height had the greatest
effect on the model.
Log E: coli ¼ 1:124 þ 0:789ðwave heightÞ
þ 0:003ðBD turbidityÞ
þ 0:004ðchlorophyllÞ
þ 0:013ðturbidityÞ
þ
0:114ðwave periodÞ
þ 3:220ðprecip_4 hÞ þ e:
Using the model, E. coli count could be accurately
predicted with 47% of the variance explained. Six type II
errors and no type I errors resulted.
3.4. Beach-specific results
The regression models were subdivided to compare
model ability between beaches. Site designations, it
should be noted, were solely jurisdictional and did not
take environmental divisions into account. An analysis
of counties determined that there was no difference
between the two counties represented, and no other
alternative dividing principles presented themselves.
Because of this type of site division, different numbers
of samples were collected at each beach, and Wells Street
in particular was set apart because only one sample was
collected daily.
In the north winds model, the total R2 was 0.64, but
higher R2 results were achieved for most of the beaches,
ranging from 0.610 to 0.731, with Wells Street having
the lowest value (Fig. 4). In south winds, however, the
R2 for all beaches was lower: 0.320. Better values could
be obtained for individual beaches, with the highest of
ARTICLE IN PRESS
5255
South All winds
wind direction
North
South
3.00 3.00
Log E. coli
2.00 2.00
1.00 1.00
R Sq Linear = 0.32 R Sq Linear = 0.465
0.00 0.00
0.50 1.00 1.50 2.00 2.50 3.00 3.50 0.50 1.00 1.50 2.00 2.50 3.00 3.50
Predicted Value Predicted Value
0.372 at Wells Street, but a much lower R2 at Lake
Street (0.113) impacted the results for the overall model.
4. Discussion
Point source outfalls can impact beach-water quality,
and point sources containing sewage input can be of
particular concern. Many criticisms of the current
beach-monitoring strategies using E. coli specifically
refer to the lack of source determination, that is,
whether the E. coli are from human sewage or natural
sources (Hardina and Fujioka, 1991; Byappanahalli et
al., 2003). E. coli was selected as an indicator of sewage
contamination based on correlative epidemiological
studies with swimmer illnesses (US EPA, 1986). If the
originating source is not human, however, the health
risks are possibly different. At beaches subject to point
source outfalls that are known sources of human
sewage, a reliable indicator is of great importance.
Several municipalities along the Little Calumet River
have long been allowed to bypass sewage treatment and
discharge combined sewage and stormwater when heavy
rainfall overwhelms their combined sewer systems.
These combined sewer overflows deliver high counts of
E. coli, along with sewage, to the southern Lake
Michigan shoreline. During the course of this study,
on 10 occasions (eight dry and two rainfall-associated)
samples from the study beaches and Burns Ditch were
analyzed for male-specific (FRNA) and somatic coliph-
age (double agar overlay method with duplicates);
somatic coliphage was detected in Burns Ditch on eight
occasions, and FRNA coliphage during the two rainfall
events. Also, somatic coliphage was found at four
beaches during dry events and four beaches during
rainfall events. FRNA was detected at Lake Street
Beach during a rainfall event. The presence of FRNA
and somatic coliphage indicates a high likelihood of
 ARTICLE IN PRESS
5256 M.B. Nevers, R.L. Whitman / Water Research 39 (2005) 5250–5260

Wells Street. Interestingly, Ogden Dunes had the lowest

North wind E. coli counts of all beaches despite being closest to the
outfall, and Lake Street had the highest counts despite
3.00 being furthest from the outfall. A similar pattern was
apparent in coliphage comparisons, with Lake Street
samples being more commonly positive for FRNA
coliphage than all other beaches. This pattern may
Log E. coli

2.00 result from the Burns Ditch plume initially moving from
the outfall past Ogden Dunes before being pushed
onshore and delivering any bacteria it may contain.
Possible explanations for higher counts at Lake Street
1.00 include the presence of a breakwall at the beach’s
BEACH R2
Ogden Dunes 0.719
western end that may produce an eddy that recirculates
× West Beach 0.722 contaminated water. Alternately, the weekly clearing of
Wells Street 0.610
Marquette 0.726
sand from the boat launch area by bulldozing may
0.00 Lake Street 0.731 disturb and resusupend E. coli-laden sands; sand has
been documented as a source of E. coli nearby (Whit-
0.50 1.00 1.50 2.00 2.50 3.00 3.50 man and Nevers, 2003). These areas were purposely
Unstandardized Predicted Value avoided for this study, but suspended material may have
contaminated the area. ANOVA showed no significant
3.00 South wind difference between sampling days, suggesting that the
plowing was not a dominant factor.
2.50 Several Burns Ditch water quality factors were related
to beach E. coli counts, including Burns Ditch water
stage, turbidity, and E. coli load, all of which may
2.00 indicate instances of flushing. During flushing/rainfall,
Log E. coli

high lake stage results from the influx of high volumes of

1.50 water, much of which carries with it wash containing
bacteria and E. coli from numerous sources. This influx
leads to high turbidity as a result of suspended sediment
1.00
and particles that also can change the apparent color.
2
BEACH R Consequently, streams may potentially increase bacter-
Ogden Dunes 0.271
0.50 × West Beach 0.342 ial loading into the lake during rain events from runoff
Wells Street 0.372
Marquette 0.333
and domestic wastes (Kim et al., 2004) and also from
0.00 Lake Street 0.113 natural areas (Byappanahalli et al., 2003), in addition to
direct combined sewage overflows. Sediment and wash
1.00 1.20 1.40 1.60 1.80 2.00 2.20 2.40 from impervious surfaces can inundate the receiving
Unstandardized Predicted Value waters, causing a significant increase in bacteria or E.
coli counts (Bannerman et al., 1993). This was seen in
Fig. 4. Actual E. coli count and predicted E. coli count based
on developed models for north and south winds and
Burns Ditch in previous studies when E. coli counts,
distinguished by beach. Lines on X and Y axes indicate 2.38, turbidity, and stream discharge peaked nearly simulta-
the EPA recommended single-sample limit for recreational neously (Olyphant et al., 2003), suggesting that runoff
waters. contributes to the increased indicator bacteria.
During heavy rainfall events, there is the possibility of
combined sewer overflows along this waterway. In the
sewage contamination. Data were collected periodically, summer of 2004, a heavy rainfall on July 21–22 resulted
but the presence of FRNA coliphage in Burns Ditch in sewage overflows at several treatment plants in three
even during a dry sampling event highlights the need to municipalities that empty into Lake Michigan via Burns
determine accurately the influence of the ditch on nearby Ditch, with a reported total of 1.25  108 gallons of
beaches. stormwater and untreated sewage. This event caused a
spike in E. coli counts in Burns Ditch with a concentra-
4.1. Relationships between Burns Ditch and beaches tion of 2398 CFU/100 and a stream load of
2.28  108 CFU/h. This was followed by a dramatic
E. coli counts in Burns Ditch greatly exceeded beach increase in the mean beach E. coli counts (Fig. 5). For
water E. coli counts and Burns Ditch E. coli was the summer, these were the highest results recorded for
correlated with E. coli at each individual beach except any of the individual beaches. Additionally, somatic
 ARTICLE IN PRESS
M.B. Nevers, R.L. Whitman / Water Research 39 (2005) 5250–5260 5257

4 3
BD E. coli
Beach E. coli
3.5 24 hr precipitation
2.5

3
Mean Log E. coli (MPN/100 ml)

Precipitation (cm)
2.5

2 1.5

1.5
1

0.5
0.5

0 0
2/ 4
7/ 004
7/ 004

11 4
14 4
17 4
20 4
23 4
26 4
29 4

1/ 4
8/ 004
8/ 004

10 4
13 4

4
7/ 00

7/ 00
7/ 200
7/ 200
7/ 200
7/ 200
7/ 200
7/ 200

8/ 00

8/ 00
8/ 200
00
/2
2
2
2

/2
2
2
2

/2
5/
8/

/
/
/
/
/
/

4/
7/

/
29
6/

Fig. 5. Mean log E. coli for the five study beaches, mean log E. coli at Burns Ditch (BD), and precipitation in the previous 24 h for the
entire study period.

coliphage was found at all study beaches immediately
following this event, and FRNA coliphage was found in
Burns Ditch and at Lake Street. The presence of phages
suggests that sewage from Burns Ditch reaches the
beaches.
4.2. Parameters for the model
Parameters independent of Burns Ditch but closely
associated with beach E. coli counts were also con-
sidered for inclusion in the model. Because the impact of
Burns Ditch is at times more pronounced than others, it
was important to determine what processes occurring on
the beach were impacting E. coli counts and therefore
could be included in the model. The impacts of many of
the parameters correlated with E. coli along the beaches
have been previously indicated in other studies. Solar
radiation will inactivate E. coli, making them non-
culturable, and counts will decrease throughout the day
as radiation increases (Davies-Colley et al., 1994; Boehm
et al., 2002; Whitman et al., 2004). The effect of wave
height and wave period on E. coli have been widely
recognized, but the interaction has been described in
numerous ways. High waves may resuspend E. coli that
have been previously deposited or are naturally occur-
ring in the beach sand (Whitman and Nevers, 2003);
alternately, high waves may move E. coli away from the
shoreline, given a convergence of currents and wind
direction (Kim et al., 2004). Along southern Lake
Michigan, high waves occur periodically during the
summer sometimes associated with and sometimes
independent of storms. The high levels of E. coli in
local beach sands (Twinning et al., 1993) may be
resuspended during periods of high waves, leading to
elevated E. coli counts in the water.
Turbidity and apparent color are interrelated in many
cases, and they often indicate high volumes of suspended
sediment. This may be the result of sediment resuspen-
sion or alternately a biological response to water
conditions. Along the shoreline, turbidity can also
decrease light penetration, thereby counteracting the
effects of solar insolation on E. coli and other
photosensitive bacteria. These factors often link turbid-
ity to high E. coli counts as a result. High concentrations
of chlorophyll can also been associated with high E. coli
counts, which are related to the release of nutrients
along with fecal indicator bacteria and subsequent
increase in primary producers.
4.3. Modeling
Wind direction has long been suspected as a major
factor in determining the effects of Burns Ditch on
beach-water quality. Outflow from Burns Ditch is
directed to the west by a protective breakwall, and if
wind direction is from the north, this plume can be
forced onshore to nearby beaches. Southern or south-
west winds have an opposite effect, typically pushing the
 ARTICLE IN PRESS
5258 M.B. Nevers, R.L. Whitman / Water Research 39 (2005) 5250–5260
plume out past the breakwall where it may be
intercepted by swimming beaches further west. By
dividing data into two groups—north or south
winds—predictions could take into account this strong
influence.
Due to the stronger influence of Burns Ditch water on
beaches during north winds, the model was more reliable
for these days and included Burns Ditch-driven factors.
However, the number of days that had north winds was
limited during the study season; during summer, the
prevailing wind direction is from the south, which is
fortunately associated with lower E. coli concentrations
and fewer closures (P ¼ 0:05). For south winds, factors
included in the model were not driven by Burns Ditch.
South winds tend to moderate or negate the impact of
Burns Ditch water on the beaches, which makes the
modeling exercise similar to attempts at modeling a
beach without a point source outfall, a more difficult
task. In these cases, non-point sources and local
influences are more influential, which has been well
described (Whitman and Nevers, 2004).
The relationship between E. coli for the day of
swimming and E. coli results from the previous day
has been relied upon for current monitoring protocols.
In this study, the examination of this relationship has
been artificially improved relative to actual monitoring
practices because all beaches were sampled daily, all but
one was sampled in replicate, and numbers were log-
transformed to achieve normal distribution. However,
the regression analysis still shows essentially no relation-
ship between the two variables, with only 3% of the
variation explained, with 19 type I errors and 19 type II
errors, compared to 64% for onshore and 32% for
offshore models developed here. Other models devel-
oped in the Great Lakes have had varying results, with a
maximum R2 of 0.4 for a group of Ohio beaches (based
on turbidity and wave height) (Francy et al., 2003) and a
maximum R2 of 0.7 for a Chicago beach (based on a
group of six variables and an interaction term)
(Olyphant and Whitman, 2004). A modeling exercise
at Milwaukee beaches had higher sensitivity than
traditional monitoring techniques using such parameters
wind direction, turbidity, water temperature, amount of
stranded algae (Bruesch and Biedrzycki, 2002).
4.4. Individual beaches
The explanatory capabilities of the models were not
noticeably different by beach, with Wells Street having a
lower R2 for north winds and Lake Street a lower R2 for
south winds. Interestingly, Wells Street was also the only
beach whose E. coli counts were not significantly
correlated with E. coli counts in Burns Ditch, indicating
that perhaps Burns Ditch is not exerting as strong an
influence on Wells Street as it may be on the other study
beaches or alternate co-factors mask these relationships.
The general difficulty of modeling E. coli during south
winds was exaggerated at Lake Street beach perhaps
because the system is strongly driven by the interaction
of the Burns Ditch plume and the breakwater that
interrupts the plume flow. Lake Street Beach had the
only two type II errors using the south winds model and
also had the highest local disturbance due to the sand
removal activities previously mentioned. It should be
noted that although the variation explained improves
when beaches are examined individually, the number of
type I and type II errors resulting is not affected, with
only one type II error recorded for Marquette, Ogden,
and Lake Street in the north wind model and two type II
errors at Lake Street for the south wind model.
4.5. Exploratory implications
The data and models resulting from the exercise
emphasize that the relationships between E. coli load
and environmental factors are far more complex than
merely sewage release followed by excess bacteria in the
stream and consequently down-current beaches. There is
reason to believe that indicator bacteria arise from
sources closer than the large sewage treatment plants,
that sewage releases occur independent of storms, and
that there are lake, ambient and Ditch interactions. The
complexity of the system increases when the Ditch’s
effluent disperses into Lake Michigan. Besides input
characteristics and antecedent conditions, hydrometeor-
ology, lake morphology, and local influences (birds,
bathers, boaters, beach grooming) are clearly interac-
tive. Deterministic modeling alone may not account for
these factors; empirical modeling, however, has the
capacity to smooth over these uncertainties, but multi-
year data sets and robust validation are needed.
4.6. Application
Using the parameters measured as a part of this study,
a predictive model for the five beaches to the west of
Burns Ditch would be useful to beach managers. During
north winds, E. coli count can be reliably predicted using
parameters that are easier to test or acquire and that can
provide results faster than currently used culturing
techniques for E. coli and enterococci. Further, because
specificity is not as important as an open/closed
determination, it should be noted that the models only
had incorrect open/closed predictions 3% (north wind)
and 2% (south wind) of the time, based on the EPA
single sample standard for recreational waters.
Given the poor record of current monitoring techni-
ques to predict accurately whether a beach should be
open or closed, a new, more reliable technique is needed.
Predictive modeling is not only reliable and cost-
effective, but it can be more accurate than the current
standards, which is not better than chance of accurately
 ARTICLE IN PRESS
M.B. Nevers, R.L. Whitman / Water Research 39 (2005) 5250–5260
opening or closing the beach. At beaches with not only
known point sources but also known sewage point
sources like Burns Ditch, modeling is more promising
because Burns Ditch directly affects beach E. coli levels.
Further, application to five beaches in this situation
significantly decreases costs and efforts associated with
beach monitoring. With readily available data collected
from existing weather and monitoring stations, predic-
tions could be made relatively easily and with far more
frequency, thereby increasing the flexibility of monitor-
ing programs and better protecting visitor health.
In summer 2005, managers implemented these models
on an experimental basis. The predictions were given to
managers as probability of closure at individual beaches.
Most managers used the information along with routine
EPA monitoring to make their daily determinations on
swimming advisory. The results are still being analyzed,
but with the addition of 2005 monitoring data to the
overall dataset used in the model, efficiency will likely
improve. The use of these tools coupled with judgment
and experience of local managers may prove superior to
traditional recommendations that evolved from regula-
tory and public health agencies.
5. Conclusions
1. The fecal indicator bacteria in the Little Calumet
River via Burns Ditch influence E. coli levels at the
five beaches extending west; E. coli of these beaches
are spatially and temporally correlated.
2. During heavy rainfall, the Burns Ditch outfall
sometimes contributes bacterial contaminants (i.e.,
coliphage) probably originating from sewage sources
to the beaches.
3. By relating water and weather parameters, E. coli
counts can be predicted fairly accurately, particularly
during north winds that force the Burns Ditch plume
onshore; however, the rate of success is much higher
for accurately determining whether a beach exceeds
the EPA standard for safe swimming.
4. Developing predictions for these five beaches has
distinct economic, logistic, and management advan-
tages over single beach models.
Acknowledgments
We thank the many people who contributed data,
time, and expertise for this project: Dawn Shively,
Muruleedhara Byappanahalli, Alyssa Bishel, Stacey
Byers, Jessica Hardesty, and Cassie Peller, USGS; Dave
Schwab, NOAA; Alex DaSilva and Rudy Osenbaugh,
IDEM; Jim Meyer and Bob Theodorou, City of Gary
and Gary Sanitary District; Tim Frame, Michigan City
Port Authority; Lefty’s Coho Landing, Portage. This
5259
article is Contribution 1348 of the USGS Great Lakes
Science Center.
References
Ackerman, D., Weisberg, S.B., 2003. Relationship between
rainfall and beach bacterial concentration on Santa Monica
Bay beaches. J. Water Health 1, 85–89.
Alm, E.W., Burke, J., Spain, A., 2003. Fecal indicator bacteria
are abundant in wet sand at freshwater beaches. Water Res.
37, 3978–3982.
American Public Health Association, 1998. Standard Methods
for the Examination of Water and Wastewater, 20th ed.
American Public Health Association, Washington, DC.
Bannerman, R.T., Owens, D.W., Dodds, R.B., Hornewer, N.J.,
1993. Sources of pollutants in Wisconsin stormwater. Water
Sci. Technol. 28, 241–259.
Boehm, A.B., Grant, S.B., Kim, J.H., Mowbray, S.L., McGee,
C.D., Clark, C.D., et al., 2002. Decadal and shorter period
variability of surf zone water quality at Huntington Beach,
California. Environ. Sci. Technol. 36, 3885–3892.
Boehm, A.B., Fuhrman, J.A., Mrse, R.D., Grant, S.B., 2003.
Tiered approach for identification of a human fecal
pollution source at a recreational beach: case study at
Avalon Bay, Catalina Island, California. Environ. Sci.
Technol. 37, 673–680.
Bruesch, M.E., Biedrzycki, P.A., 2002. Preliminary compara-
tive analysis of two models used to predict E. coli levels in
recreational water in Milwaukee. Great Lakes Beach
Conference 2002. Chicago, Illinois.
Byappanahalli, M., Fowler, M., Shively, D., Whitman, R.,
2003. Ubiquity and persistence of Escherichia coli within a
midwestern stream. Appl. Environ. Microbiol. 69,
4549–4555.
Crowther, J., Kay, D., Wyer, M.D., 2001. Relationships
between microbial water quality and environmental condi-
tions in coastal recreational waters: the Fylde coast, UK.
Water Res. 35, 4029–4038.
Davies-Colley, R.J., Bell, R.G., Donnison, A.M., 1994. Sun-
light inactivation of enterococci and faecal coliforms in
sewage effluent diluted in seawater. Appl. Environ. Micro-
biol. 60, 2049–2058.
Francy, D.S., Gifford, A.M., Darner, R.A., 2003. Escherichia
coli at Ohio bathing beaches—distribution, sources, waste-
water indicators, and predictive modeling. US Geological
Survey, Water-Resources Investigations Report 02-4285.
Columbus, Ohio.
Fujioka, R.S., Hashimoto, H.H., Siwak, E.B., Young, R.H.F.,
1981. Effect of sunlight on survival of indicator bacteria in
seawater. Appl. Environ. Microbiol. 41, 690–696.
Hardina, C.M., Fujioka, R.S., 1991. Soil: the environmental
source of Escherichia coli and enterococci in Hawaii’s
streams. Environ. Toxicol. Water Qual. 6, 185–195.
Kim, J.H., Grant, S.B., McGee, C.D., Sanders, B.F., Largier,
J.L., 2004. Locating sources of surf zone pollution: a mass
budget analysis of fecal indicator bacteria in Huntington
Beach, California. Environ. Sci. Technol. 38, 2626–2636.
LeFevre, N.M., Lewis, G.D., 2003. The role of resuspension in
enterococci distribution in water at an urban beach. Water
Sci. Technol. 47, 205–210.
 ARTICLE IN PRESS
5260 M.B. Nevers, R.L. Whitman / Water Research 39 (2005) 5250–5260
McLellan, S.L., Salmore, A.K., 2003. Evidence for localized
bacterial loading as the cause of chronic beach closings in a
freshwater marina. Water Res. 37, 2700–2708.
Morrison, A.M., Coughlin, K., Shine, J.P., Coull, B.A., Rex,
A.C., 2003. Receiver operating characteristic curve analysis
of beach water quality indicator variables. Appl. Environ.
Microbiol. 69, 6405–6411.
Obiri-Danso, K., Jones, K., 2000. Intertidal sediments as
reservoirs for hippurate negative campylobacters, salmonel-
lae and faecal indicators in three EU recognised bathing
waters in north west England. Water Res. 34, 519–527.
Olyphant, G.A., Whitman, R.L., 2004. Elements of a predictive
model for determining beach closures on a real time basis:
the case of 63rd Street Beach Chicago. Environ. Monit.
Assess. 98, 175–190.
Olyphant, G.A., Thomas, J., Whitman, R.L., Harper, D., 2003.
Characterization and statistical modeling of bacterial
(Escherichia coli) outflows from watersheds that discharge
into southern Lake Michigan. Environ. Monit. Assess. 81,
289–300.
Smith, P., Carroll, C., Johnson, P., Gabhainn, S.N., Smith,
L.P., 1999. The effect of wind speed and direction on the
distribution of sewage-associated bacteria. Lett. Appl.
Microbiol. 28, 184–188.
SPSS, 2003. SPSS, Version 12. SPSS Inc., Chicago.
Stoeckel, D.M., Mathes, M.V., Hyer, K.E., Hagedorn, C.,
Kator, H., Lukasik, J., et al., 2004. Comparison of seven
protocols to identify fecal contamination sources using
Escherichia coli. Environ. Sci. Technol. 38, 6109–6117.
Systat, 2004. Systat, Version 11. Systat Software, Inc.,
Richmond, California.
Twinning, T.L., Whitman, R.L., Hoff, B.E., 1993. Occurrence
of E. coli in open beach groundwaters of Indiana Dunes
National Lakeshore. 109th Annual Meeting of the Indiana
Academy of Science. West Lafayette, Indiana.
US EPA, 1986. Ambient Water Quality Criteria for Bacteria
1986. US EPA, Office of Water Regulations and Standards,
Washington, DC, p. 18.
Whitman, R.L., Nevers, M.B., 2003. Foreshore sand as a
source of Escherichia coli in nearshore water of a Lake
Michigan beach. Appl. Environ. Microbiol. 69, 5555–5562.
Whitman, R.L., Nevers, M.B., 2004. Escherichia coli sampling
reliability at a frequently closed Chicago beach: monitoring
and management implications. Environ. Sci. Technol. 39,
4241–4246.
Whitman, R.L., Shively, D.A., Pawlik, H., Nevers, M.B.,
Byappanahalli, M.N., 2003. Occurrence of Escherichia coli
and enterococci in Cladophora (Chlorophyta) in nearshore
water and beach sand of Lake Michigan. Appl. Environ.
Microbiol. 69, 4714–4719.
Whitman, R.L., Nevers, M.B., Korinek, G.C., Byappanahalli,
M.N., 2004. Solar and temporal effects on Escherichia coli
concentration at a Great Lakes swimming beach. Appl.
Environ. Microbiol. 70, 4276–4285.