February 16, 2004
Robert Pitt, Alex
Maestre, and Renee Morquecho
Dept. of Civil and
Environmental Engineering
University
of Alabama
Tuscaloosa,
AL 35487
Abstract 2
Project
Description and Background. 2
Data
Collection and Analysis Efforts to Date. 4
Preliminary Summary of U.S. NPDES
Phase 1 Stormwater Data. 5
Simple Data Relationships. 14
Example
Statistical Analyses of Data Comparing First Flush and Composite Sample
Concentrations. 23
Modeling
Building using the NSQD.. 25
Factors Potentially Affecting
Stormwater Pollutant Concentrations. 25
Power Calculations as a Function of
Numbers of Data Observations. 27
Multivariate Analyses of Factors. 28
Sampling
Guidance for Stormwater Monitoring. 30
Numbers of Samples Needed. 30
Typical
Numbers of Samples Needed for a Basic Stormwater Monitoring Program.. 31
Detection Limits of Analytical
Methods. 31
Sampling Methods. 32
Conclusions. 32
Suggested Role for Continued
Stormwater Monitoring. 33
Acknowledgements. 34
References. 34
The University of Alabama
and the Center for Watershed Protection were awarded an EPA Office of Water
104(b)3 grant in 2001 to collect and evaluate stormwater data from a
representative number of NPDES (National Pollutant Discharge Elimination
System) MS4 (municipal separate storm sewer system) stormwater permit holders.
The initial version of this database, the National Stormwater Quality Database
(NSQD, version 1.1) is currently being completed. These stormwater quality data
and site descriptions are being collected and reviewed to describe the
characteristics of national stormwater quality, to provide guidance for future
sampling needs, and to enhance local stormwater management activities in areas
having limited data.
The monitoring data collected over nearly a ten-year period
from more than 200 municipalities throughout the country have a great potential
in characterizing the quality of stormwater runoff and comparing it against
historical benchmarks. This project is creating a national database of
stormwater monitoring data collected as part of the existing stormwater permit
program, providing a scientific analysis of the data, and providing
recommendations for improving the quality and management value of future NPDES
monitoring efforts.
Each data set is receiving a quality assurance/quality
control review based on reasonableness of data, extreme values, relationships
among parameters, sampling methods, and a review of the analytical methods. The
statistical analyses are being conducted at several levels. Probability plots
are used to identify range, randomness and normality. Clustering and principal
component analyses are utilized to characterize significant factors affecting
the data patterns. The master data set is also being evaluated to develop
descriptive statistics, such as measures of central tendency and standard
errors. Regional and climatic differences are being tested, including the
influences of land use, and the effects of storm size and season, among other
factors. The data will be used to develop a method to predict expected
stormwater quality for a variety of significant factors and will be used to
examine a number of preconceptions concerning the characteristics of
stormwater, sampling design decisions, and some basic data analysis issues.
Some of the issues that are being examined with this data include: the
occurrence and magnitude of first-flushes, the effects of different sampling
methods (the use of grab sampling vs. automatic samplers, for example) on
stormwater quality data, trends in stormwater quality with time, the effects of
infrequent wrong data in large data bases, appropriate methods to handle values
that are below detection limits, the necessary sampling effort needed to
characterize stormwater quality, for example. This paper describes the data
collected to date and presents some preliminary data findings.
When this National Stormwater Quality Database (NSQD) is
completed (populated with most of the NPDES stormwater monitoring data), the
continued routine collection of outfall stormwater quality data in the U.S.
for basic characterization purposes may have limited use. Some communities may
have obviously unusual conditions, or adequate data may not be available in
their region. In these conditions, outfall monitoring may be needed. However,
stormwater monitoring will continue to be needed for other purposes in many
areas having, or anticipating, active stormwater management programs
(especially when supplemented with other biological, physical, and hydrologic
monitoring components). These new monitoring programs should be designed specifically
for additional objectives, beyond basic characterization. These objectives may
include receiving water assessments to understand local problems, source area
monitoring to identify critical sources of stormwater pollutants, treatability
tests to verify the performance of stormwater controls for local conditions,
and assessment monitoring to verify the success of the local stormwater
management approach (including model calibration and verification). In many
cases, the resources being spent for outfall monitoring could be more
effectively spent to better understand many of these other aspects of an
effective stormwater management program.
The importance of this project is based on the scarcity of
nationally summarized and accessible data from the existing U.S. EPA’s NPDES
stormwater permit program. There have been some local and regional data
summaries, but little has been done with nationwide data. A notable exception
is the Camp, Dresser, and McGee (CDM) national stormwater database (Smullen and
Cave 2002) that combined historical Nationwide Urban Runoff Program (NURP) (EPA
1983), available urban U.S. Geological survey (USGS), and selected NPDES data.
Their main effort has been to describe the probability distributions of these
data (and corresponding EMCs, the event mean concentrations). They concluded
that concentrations for different land uses were not significantly different,
so all their data were pooled.
Between 1978 and 1983, the EPA conducted the NURP that
examined stormwater quality from separate storm sewers in different land uses
(EPA 1983). This project studied 81 outfalls in 28 communities throughout the U.S.
and included the monitoring of approximately 2300 storm events. The data was
presented for several land use categories, although most of the information was
obtained from residential lands. Since NURP, other important studies have been
conducted that characterize stormwater. The USGS created a database with more
than 1100 storms from 98 monitoring sites in 20 metropolitan areas. The Federal
Highway Administration (FHWA) analyzed stormwater runoff from 31 highways in 11
states during the 1970s and 1980s (Cave 1995). Strecker (personal
communication) is also collecting information from highway monitoring as part
of a current NCHRP-funded project. The city of Austin
also developed a database having more than 1200 events (Smullen 2003).
Other regional databases also exist, mostly using local
NPDES data. These include the Los Angeles area database,
the Santa Clara and Alameda
County (California)
databases, the Oregon Association of Clean Water Agencies Database, and the Dallas,
Texas, area stormwater database. These
regional data are (or will be) included in the NSQD national database. However,
the USGS or historical NURP data will not be included in the NSQD database due
to lack of consistent descriptive information for the older drainage areas and
because of the age of the data from those prior studies. Much of the NURP data
is available in electronic form at the University
of Alabama student American Water
Resources Association web page at: http://www.eng.ua.edu/~awra/download.htm.
The results (especially the stormwater characteristic prediction procedures)
from these other databases will be compared to similar findings from the final
analyses using this expanded database to indicate any important differences.
Outside the U.S.,
there have been important efforts to characterize stormwater. In Toronto,
Canada, the Toronto Area
Watershed Management Strategy Study (TAWMS) was conducted during 1983 and 1984
and extensively monitored industrial stormwater, along with snowmelt in the
urban area (Pitt and McLean 1986), for example. Numerous other investigations
in South Africa, the South Pacific,
Europe and Latin America have also been conducted over
the past 30 years, but no large-scale summaries of that data have been
prepared. About 3,500 international references on stormwater have been reviewed
and compiled since 1996 by the Urban Wet Weather Flows literature review team
for publication in Water Environment
Research (Field, et al. 1997,
1998; O’Connor, et al. 1999; Fan, et al. 2000; Clark, et al. 2001, 2001, 2003). An overall compilation of these literature
reviews is available at:
http://www.eng.ua.edu/~rpitt/Publications/Publications.shtml
The reviews include short summaries of the papers and are
organized by major topics. Besides journal articles, many published conference
proceedings are also represented (including the extensive conference
proceedings from the 8th International Conference on Urban Storm
Drainage held in Sydney, Australia, in 1999, the 9th International
Conference on Urban Storm Drainage held in Portland, OR, in 2002, and the
Toronto Stormwater and Urban Water Systems Modeling conference series, amongst
many other specialty conferences).
The NSQD is unique in that detailed descriptions of the test
areas and sampling conditions are also being collected, including aerial
photographs and topographic maps that are being obtained from public domain
Internet sources. Land use information used is as supplied by the communities
submitting the data, although aerial photographs and maps are also used to help
clarify questions concerning specific development characteristics. Most of the
sites have homogeneous land uses, although many are mixed. These
characteristics are all fully noted in the database.
Stormwater runoff data from existing NPDES permit
applications and annual monitoring reports are being collected during this
project. This project also includes extensive QA/QC (quality assurance/quality
control) evaluations of these data; and performing statistical analyses and
summaries of these data. The final information will be published on the
Internet (such as on an EPA OW-OWM, Office of Water and Office of Wastewater
Management, site and on the Center for Watershed Protection’s SMRC, Stormwater
Manager’s Resources Center, site at: http://www.stormwatercenter.net/).
Some of the information is currently located at Pitt’s teaching and research
web site at:
http://www.eng.ua.edu/~rpitt/Research/ms4/mainms4.shtml
The Phase
I NPDES communities included areas with:
· A stormwater discharge from a MS4 serving a
population of 250,000 or more (large system), or
· A stormwater discharge from a MS4 serving a
population of 100,000 or more, but less than 250,000 (medium system).
More than 200 municipalities, plus numerous
additional special districts and governmental agencies were included in this
program. Part 2 of the NPDES discharge permit application specified that
sampling was needed and that the following items were to be included in the
application:
· Proposed monitoring
program for representative data collection during the term of the permit;
· Quantitative data from 5
to 10 representative locations;
· Estimates of the annual
pollutant load and event mean concentration (EMC) of system discharges; and
· Proposed schedule to
provide estimates of seasonal pollutant loads and the EMC for certain detected
constituents during the term of the permit.
The permit applications were due in 1992 and 1993. For Part
2 of the application, municipalities were to submit grab (for certain
pollutants having severe holding time restrictions, such as bacteria) and
flow-weighted sampling data from selected sites (5 to 10 outfalls) for three
representative storm events at least one month apart. In addition, the
municipalities must have also developed programs for future sampling activities
that specified sampling locations, frequency, pollutants to be analyzed, and
sampling equipment.
Numerous constituents were to be analyzed, including typical
conventional pollutants (TSS, TDS, COD, BOD5, oil and grease, fecal
coliforms, fecal strep., pH, Cl, TKN, NO3, TP, and PO4),
plus many heavy metals (including total forms of arsenic, chromium, copper,
lead, mercury, and zinc, plus others), and numerous listed organic toxicants
(including PAHs, pesticides, and PCBs). Many communities also analyzed samples
for filtered forms of the heavy metals. This database currently includes information
for about 125 different stormwater quality constituents, although the current
database is mostly populated with data from 35 of the commonly analyzed
pollutants (as summarized later in Table 1). Therefore, there has been a
substantial amount of stormwater quality data collected during the past 10
years throughout the U.S.,
although most of these data are not readily available, nor have detailed
statistical analyses been conducted and presented.
As of mid-summer 2003, 3,770 separate events from 66
agencies and municipalities from 17 states have been collected and the data
entered into NSQD. Figure 1 shows the locations of these municipalities on a
national map, along with EPA Rain Zones. Excellent national coverage is
anticipated, although there will be few municipalities from the northern,
west-central states of Montana, Wyoming,
and North and South Dakota (where
cities are generally small, and few were included in the Phase 1 NPDES
program). This current database (NSQD, Version 1.1) covers areas mostly in the
southern, Atlantic, central, and western parts of the US.
Anticipated future project phases will help extend the national coverage.
Some of the municipalities that have been contacted (and
some in which data was received) have information that could not be used for
various reasons. One of the most common reasons was that the samples had been
collected from receiving waters (such as Washington
state, Nashville, and Chattanooga).
Only data from well-described stormwater outfall locations are being used for
the database. These can be open channel outfalls in completely developed areas,
but are more commonly conventional outfall pipes. The other major problem is
that the sampling locations and/or the drainage areas were not described. Data
with some missing information is being used for now, with the intention of
obtaining the needed information later. However, there will likely still be
some minor data gaps that will not be able to be filled. In addition, the list
of constituents being monitored has varied for different locations. Most areas
evaluated the common stormwater constituents, but few have included organic
toxicants. The most serious gap is the frequent lack of runoff volume data,
although all sites have included rain data. Finally, if all the data were
collected that was requested, the current project resources will not permit
their full utilization, as it requires a great deal of time to enter and review
this information. About 10% of the collected data needed verification during
the QA/QC process. If that potentially faulty data remained in the database,
spurious statistical analyses would have resulted. The collection and review of
the data is a necessary first step to facilitate later analyses.
The assembled data was entered into NSQD, including site
descriptions (state, municipality, land use components, and EPA rain zone),
sampling information (date, season, rain depth, runoff depth, sampling method,
sample type, etc.), and constituent measurements (concentrations, grouped in
categories). In addition, more detailed site, sampling, and analysis
information has been collected for most sampling sites and is also included as
supplemental information. The reported land use information supplied by the
communities is being used, with verification of some areas with aerial
photographs and maps. In many cases, the sampled watersheds have multiple land
uses and those designations are included in the database (the database lists
the percentages of the drainage as residential, commercial, industrial,
freeway, institutional, and open space). The final data analyses will consider
these mixed sites also, especially for verification for the model development
activities, although the following preliminary results are only for the
homogeneous land use sites.
Additional site information is being acquired to complete
most of the missing records before the final data analyses. The following data
and analysis descriptions should therefore be considered preliminary and will
change with this additional data and analyses. However, this presentation only
uses the most basic and robust analyses for preliminary consideration. The
final report and data presentations will obviously be much more comprehensive.
Table 1 is a summary of the Phase 1 data collected and
entered into the database as of mid-summer 2003. The data are separated into 11
land use categories: residential, commercial, industrial, institutional, freeways,
and open space, plus mixtures of these land uses. Summaries are also shown for
mixed land use areas (indicating the most prominent land use), and for the
total data set combined. Only data having at least 50 total detected
observations and at least 10 detected observations per land use category are
shown on this table. The full database includes all of the data. In most cases,
many more than these minimum numbers are available. The total number of
observations and the percentage of observations above the detection limits are also
shown on this summary table. However, some constituents were not monitored by
very many stormwater permit holders, and some constituents were mostly all in
the “not detected” category, and those data are not shown. As an example,
filtered heavy metal observations, and especially organic analyses, have many
fewer detected values than other constituents.
The total number of individual events included in the
database is 3,770, with most in the residential category (1,069 events). For
most common constituents, detectable values are available for almost all
monitored events. The median and coefficient of variation (COV) values are only
for those data having detectable concentrations. If the non-detected results
were used in these calculations, extreme biases would invalidate many of the
calculations. The final analyses will further examine issues associated with
different detection limits, multiple laboratories, and varying analytical
methods on the reported results and statistical analyses. Burton
and Pitt (2002), and the many included references in that book, contains
further discussions on these important issues.
Figure 1. Communities from which data has been obtained and entered in the
NSQD, along with EPA Rain Zones.
Table 2 is a summary of methylene chloride and
bis(2-ethylhexyl) phthalate, the most commonly reported and detected organic
constituents. There were up to several hundred events that included PAH
and pesticide data. The percentage of samples that had observable
concentrations of these constituents ranged from 15 to 35%, about the same
detection rate as in previous stormwater investigations, such as Pitt, et
al. 1995.
Statistical analyses are being conducted in stages. Probability
plots were used to identify range, randomness, and normality. Figure 2 is an
example of log-normal probability plots for some of the constituents and for
all data pooled. Probability plots shown as straight lines indicate that the
concentrations can be represented by log-normal distributions. This is
important as it indicates that data transformations, or the use of
nonparametric statistical analyses, will be needed. Plots with obvious
discontinuities imply that multiple data populations may be included. The
future analyses will identify the significance of these different data
categories (such as land use, region, and season).
Figure 2. Log-normal probability plots of stormwater quality data for
selected constituents.
Table 2. Summary of Selected Organic
Information in NSQD, version 1.0
|
|
Methylene-chloride (mg/L)
|
Bis(2-ethylhexyl)
phthalate (mg/L)
|
|
All
Data Combined
|
|
|
|
Number of
observations
|
251
|
250
|
|
% of
samples above detection
|
36
|
30
|
|
Median of
detected values
|
11.2
|
9.5
|
|
Coefficient
of variation
|
0.77
|
1.13
|
The master data set will also be evaluated to develop
descriptive statistics, such as measures of central tendency and standard
errors. The runoff data will then be evaluated to determine which factors have
a strong influence on event mean concentrations, including sampling methods.
Tests for regional and climatic differences will be conducted, including the
influences of land use and the effects of storm size, among other factors.
Figure 3 includes example scatter plots of COD vs. BOD5, ammonia vs.
TKN, filtered copper vs. total copper, and filtered zinc vs. total zinc,
illustrating close relationships between these pairings, as expected.
Figure 4 shows scatter plots of suspended solids,
phosphorus, fecal coliforms, and total zinc concentrations for different rain
depths. Little variation of these concentrations with rain depth are seen when
all of the data are combined, implying little likelihood of important
“first-flush” effects at stormwater outfall locations. If a first-flush was
evident, one would expect higher concentrations associated with smaller rain
depths (see Maestre, et al. 2003 for more detailed analyses of
first-flush effects using the NSQD database information). A simple plot of COD
concentrations vs. percentage imperviousness of the drainage area (Figure 5) doesn’t
indicate any obvious trends. Each vertical set of observations represent a
single monitoring location (all of the events at a single location have the
same percent imperviousness). The variation of COD at any one monitoring
location is seen to vary greatly, typically by about an order of magnitude.
These large variations will make trends difficult to identify. All of the
lowest percentage imperviousness sites are open space land uses, while all of
the highest percentage imperiousness sites are freeway and commercial land
uses. As indicated below in Figure 6, many of the constituents have significant
concentration differences by land uses. Therefore, it is expected that these
other constituents will show an obvious trend because of the strong correlation
between percentage imperviousness and land use. In addition, currently there is
little data in the NSQD showing how the impervious areas are connected to the
drainage systems. Some historical data shows much smaller concentrations (and
especially yields) for areas that are drained by grass swales compared to
concrete curbs and gutters. With this additional information, the
imperviousness data can be adjusted (“effective” imperviousness is commonly
used to designate directly connected paved areas) to potentially identify more
obvious data trends.
Figure 6 contains examples of grouped box and whisker plots
for several constituents for different major land use categories. The TKN, plus
copper, lead, and zinc observations are lowest for open space areas, while the
freeway locations generally had the highest median values, except for
phosphorus, nitrates, fecal coliforms, and zinc. The industrial sites had the
highest reported zinc concentrations. Preliminary statistical ANOVA analyses
for all land use categories (using SYSTAT) found significant differences for
land use categories for all pollutants. The final analyses will further
investigate this important finding and will also examine possible confounding
factors.
The seasonal variations for the example residential data
shown in Figure 7 are not as obvious, except that the bacteria values appear to
be lowest during the winter season and highest during the summer and fall (a
similar conclusion was obtained during the NURP, EPA 1983, data evaluations).
The database does not contain any snowmelt data, so all of the data corresponds
to rain-related runoff.
Figure 8 presents example plots for selected residential
area data for different EPA rain zones for the country. Zones 3 and 7 (the
wettest areas of the country) had the lowest concentrations for most of the
constituents.
Figure 3. Example scatter plots of stormwater data (line of equivalent
concentration shown).
Figure 4. Example scatter plots of concentrations vs. rain depth.
Figure 5. Plot COD concentrations against watershed area percent
imperviousness values for different land uses (CO: commercial; FW: freeway; ID:
industrial; OP: open space; and RE: residential)
Figure 6. Example stormwater data sorted by land use (no mixed land use
data included in plots).
Figure 6. Example stormwater data sorted by land use (no mixed land use
data included in plots) (continued).
Figure 7. Example residential area stormwater pollutant concentrations
sorted by season.
Figure 8. Example residential area
stormwater pollutant concentrations sorted by geographical area.
We are also examining trends of concentrations with time. A
classical example would be for lead, which is expected to decrease over time
with the increased use of unleaded gasoline. Older stormwater samples from the
1970s typically have had lead concentrations of about 100 mg/L,
or higher, while most current data indicate concentrations in the range of 1 to
10 mg/L.
Figure 9 shows a plot of lead concentrations for residential areas only (in
rain zone 2), for the time period from 1991 to 2002. This preliminary plot
shows likely decreasing lead concentrations with time. Statistically however, the
trend line is not significant due to the large variation in observed
concentrations (p=0.41; there is insufficient data to show that the slope term
is significantly different from zero). The similar COD concentrations in Figure
9 also have an apparent downward trend with time, but again, the slope term is
not significant (p=0.12).
|
|
|
Figure 9. Residential lead and COD concentrations with time (EPA Rain Zone
2 data only).
As part of their MS4 phase 1 applications, Denver and Milwaukee both
returned to some of their earlier sampled monitoring stations used during the
local NURP projects. In the time between the early 1980s (NURP) and the early
1990s (MS4), they did not detect any significant differences, except for large
decreases in lead concentrations. Figure 10 compares suspended solids, copper,
lead, and zinc concentrations at the Wood Center NURP monitoring site in Milwaukee. The
average site concentrations remained the same, except for lead, which decreased
from about 450 down to about 110 µg/L.
Figure
10. Comparison of pollutant concentrations collected during NURP (1981) to MS4
application data (1990) at the same location (personal communication, Roger
Bannerman, WI DNR).
Similar comparisons were made in the Denver
Metropolitan area by the Urban Drainage and Flood Control District. Table 3
compares stormwater quality for commercial and residential areas for 1980/91
(NURP) and 1992/93 (MS4 application). Although there was an apparent difference
in the averages of the event concentrations between the sampling dates, they
concluded that the differences were all within the normal range of stormwater quality
variations, except for lead, which decreased by about a factor of four.
Table
3. Comparison of Commercial and Residential Stormwater Runoff Quality from
1980/81 to 1992/93 (Urban Drainage District, Flood Hazard News, Dec 1993.)
|
|
Commercial
|
Residential
|
|
Constituent
|
1980/81
|
1992/93
|
1980/82
|
1992/93
|
|
Total suspended solids (mg/L)
|
251
|
165
|
226
|
325
|
|
Total nitrogen (mg/L)
|
3.0
|
3.9
|
3.2
|
4.7
|
|
Nitrate plus nitrite (mg/L)
|
0.80
|
1.4
|
0.61
|
0.92
|
|
Total phosphorus (mg/L)
|
0.46
|
0.34
|
0.61
|
0.87
|
|
Dissolved phosphorus (mg/L)
|
0.15
|
0.15
|
0.22
|
0.24
|
|
Copper, total recoverable (µg/L)
|
27
|
81
|
28
|
31
|
|
Lead, total recoverable (µg/L)
|
200
|
59
|
190
|
53
|
|
Zinc, total recoverable (µg/L)
|
220
|
290
|
180
|
180
|
As part of their NPDES stormwater permit, some communities
collected grab samples during the first 30 minutes of the event to evaluate a
“first flush” in contrast to the flow-weighted composite data. More than 400
paired samples representing the first flush and composite samples from eight
communities (mostly located in the southeast U.S.)
from NSQD were reviewed. Box and probability plots were prepared for 22 major
constituents. Nonparametric statistical analyses were then used to measure the
differences between the sample sets. This discussion summarizes the results of
this preliminary analysis, including the effects of storm size and land use on
the presence and importance of first flushes. Only concentration data were
available for these analyses, so traditional accumulative mass curves could not
be developed.
First flush refers to an assumed elevated load of pollutants
discharged in the first part of a runoff event. First flush has been observed
more in small catchments than in large catchments (Thompson, et al. 1995; WEF and ASCE 1998). In
large catchments (>162 ha, or >400 acres) the highest concentrations have
been observed at the times of flow peak (Soeur, et al. 1994; Brown, et al.
1995). The presence of a first flush has been reported to be associated with
runoff duration by the City of Austin, TX (Swietlik, et al. 1995). An observed first flush may be present for some
pollutants, but not others (Ellis 1986; Adams 2000). Adams
(2000) and Deletic (1998) both concluded that the presence of a first flush
depends on numerous site and rainfall characteristics.
It is expected that peak concentrations generally occur
during periods of peak flow (and highest rain energy). On relatively small
paved areas, however, it is likely that there will always be a short period of
relatively high concentrations associated with washing off of the most
available material near the beginning of the runoff event (Pitt 1987). This
peak period of high concentrations may be overwhelmed by periods of high rain
intensity that may occur later in the event. In addition, in more complex
drainage areas, the routing of these short periods of peak concentrations may
blend with larger flows and may not be noticeable. A first flush in a separate
storm drainage system is therefore most likely to be seen if a rain occurs at
relatively constant intensity over a paved area having a simple and small drainage
system.
A total of 417 storm events with paired first flush and
composite storm samples were available from the NSQD. The majority of the
events were located in North Carolina (76.2%), but some
events were also from Alabama (3.1%), Kentucky
(13.9%) and Kansas (6.7%). All of
the data were from end-of pipe samples in separate storm drainage systems.
The initial analyses were used to select the constituents
and land uses that meet the requirements of the statistical comparison tests.
Probability plots, box plots, concentration vs. precipitation, and standard
descriptive statistics, were performed for 22 constituents for each land use,
and for all land uses combined. Nonparametric statistical analyses were
performed after the initial analyses. Mann Whitney and Fligner Policello tests
were most commonly used. Minitab and Systat statistical programs, along with
Word and Excel macros, were used for the analyses.
The Mann-Whitney and Fligner-Policello non-parametric tests
were selected to determine if there were statistically significant differences
between the first flush and composite data sets for each land use and
constituent. These tests are very useful because they require only data
symmetry, not normality, to evaluate the hypothesis. The null hypothesis during
the analysis was that the median concentrations of the first flush and
composite data sets were the same. The alternative hypothesis was that the
medians were different, with a confidence of at least 95%.
A complete description of these analyses is presented in
Maestre, et al. (2004). Table 4 summarizes the results of the analysis.
The “>” sign indicates that the median of the first flush data set is higher
than for the composite storm data set. The “=” sign indicates that the there is
not enough information to reject the null hypothesis. Events without enough
data for the analyses are represented with an “X”. Also shown on this table are
the ratios of the medians of the first flush and the composite data sets for
each constituent and land use. The first flush samples were larger than for the
composite samples if the ratio is great than one. Generally, a statistically
significant first flush is associated with a median concentration ratio of
about 1.4, or greater (the exceptions occurred when the number of samples in a
specific category is small). The largest significant ratios are about 2.5,
indicating that the first flush concentrations may be about 2.5 times greater
than the composite concentrations. More of the larger ratios are found in the commercial
and institutional land use categories, areas where larger paved areas are
likely to be found. The smallest ratios are associated with the residential,
industrial, and open space land uses, locations where there may be larger areas
of unpaved surfaces.
Results indicate that for 55% of the evaluated cases, the
medians of the first flush data sets were significantly larger than for the
composite sample sets. In the remaining 45% of the cases, both medians were
expected to be the same, or the concentrations were possibly greater later in
the events. About 70% of the constituents in the commercial land use category
had first-flushes, while about 60% of the constituents in the residential,
institutional and the mixed (mostly commercial and residential) land use
categories had first flushes, and about 45% of the constituents in the
industrial land use category had first-flushes. In contrast, no constituents
were found to have first-flushes in the open space category.
COD, BOD5, TDS, TKN, and Zn all had first flushes
in all areas (except for the open space category). In contrast, turbidity, pH,
fecal coliforms, fecal strep., total N, dissolved and ortho-P never showed a
statistically significant first flush in any category. The conflict with TKN
and total N implies that there may be some other factors involved in the
identification of first flushes besides land use. If additional paired data
become available during later project periods, it may be possible to extend
these analyses to consider rain effects, drainage area, and geographical
location.
Table 4. Presence of Significant First Flushes (ratio of first flush to
composite median concentrations)
|
Parameter
|
Commercial
|
Industrial
|
Institutional
|
Open Space
|
Residential
|
All Combined
|
|
Turbidity
|
= (1.32)
|
X
|
X
|
X
|
= (1.24)
|
= (1.26)
|
|
pH
|
= (1.03)
|
= (1.00)
|
X
|
X
|
= (1.01)
|
= (1.01)
|
|
COD
|
> (2.29)
|
> (1.43)
|
> (2.73)
|
= (0.67)
|
> (1.63)
|
> (1.71)
|
|
TSS
|
> (1.85)
|
= (0.97)
|
> (2.12)
|
= (0.95)
|
> (1.84)
|
> (1.60)
|
|
BOD5
|
> (1.77)
|
> (1.58)
|
> (1.67)
|
= (1.07)
|
> (1.67)
|
> (1.67)
|
|
TDS
|
> (1.82)
|
> (1.32)
|
> (2.66)
|
= (1.07)
|
> (1.52)
|
> (1.55)
|
|
O&G
|
> (1.54)
|
X
|
X
|
X
|
= (2.05)
|
> (1.60)
|
|
Fecal
Coliform
|
= (0.87)
|
X
|
X
|
X
|
= (0.98)
|
= (1.21)
|
|
Fecal
Strep.
|
= (1.05)
|
X
|
X
|
X
|
= (1.30)
|
= (1.11)
|
|
Ammonia
|
> (2.11)
|
= (1.08)
|
> (1.66)
|
X
|
> (1.36)
|
> (1.54)
|
|
NO2
NO3
|
> (1.73)
|
> (1.31)
|
> (1.70)
|
= (0.96)
|
> (1.66)
|
> (1.50)
|
|
Total N
|
= (1.35)
|
= (1.79)
|
X
|
= (1.53)
|
= (0.88)
|
= (1.22)
|
|
TKN
|
> (1.71)
|
> (1.35)
|
X
|
= (1.28)
|
> (1.65)
|
> (1.60)
|
|
Total P
|
> (1.44)
|
= (1.42)
|
= (1.24)
|
= (1.05)
|
> (1.46)
|
> (1.45)
|
|
P
Dissolved
|
= (1.23)
|
= (1.04)
|
= (1.05)
|
= (0.69)
|
> (1.24)
|
= (1.07)
|
|
Phosphate
Ortho
|
X
|
= (1.55)
|
X
|
X
|
= (0.95)
|
= (1.30)
|
|
Cd
|
> (2.15)
|
= (1.00)
|
X
|
= (1.30)
|
> (2.00)
|
> (1.62)
|
|
Cr
|
> (1.67)
|
= (1.36)
|
X
|
= (1.70)
|
= (1.24)
|
> (1.47)
|
|
Cu
|
> (1.62)
|
> (1.24)
|
= (0.94)
|
= (0.78)
|
> (1.33)
|
> (1.33)
|
|
Pb
|
> (1.65)
|
> (1.41)
|
> (2.28)
|
= (0.90)
|
> (1.48)
|
> (1.50)
|
|
Ni
|
> (2.40)
|
= (1.00)
|
X
|
X
|
= (1.20)
|
> (1.50)
|
|
Zn
|
> (1.92)
|
> (1.540
|
> (2.48)
|
= (1.25)
|
> (1.58)
|
> (1.59)
|
As indicated earlier, an important objective of the NSQD is
to develop a predictive tool to enable stormwater managers to determine the likely
stormwater quality for their area. In many cases, adequate data may be
available in the NSQD to fit their situation. However, it is also expected that
some will need to establish a local monitoring program to obtain reliable
estimates of their stormwater quality. The next subsection provides some
monitoring guidance for this situation, while this subsection presents an
example of the model building process that we are currently using.
The database contains information for the monitored
watersheds, along with the outfall runoff quality. Each sample is labeled with
the land use, season, geographical area, percent imperviousness, rain amount,
and many other attributes in the database. The first phase of the NSQD project
focused on the mid Atlantic and Gulf coast areas,
although additional data has been collected for other locations. About 54% of
the existing data in the database is from communities located in Maryland,
Virginia, Pennsylvania, North
Carolina, Kentucky and Tennessee.
The following factors may affect the reported stormwater pollutant
concentrations:
·
Landuse: All of the watershed areas were
separated into residential, commercial, industrial, open space and freeway land
uses. Data are also available from mixed landuse areas which will be used later
to verify the prediction methods.
·
EPA Rain Zone: As shown in Figure 1, the country
is divided in 9 rain/climatic regions representing all combinations of areas
having warm summers, cold winters, large rainfalls, and little rain.
·
Season: Four seasons were identified by the
month when the samples were collected: Winter (December to February); Spring
(March to May); Summer (June to August); and Fall (September to November).
·
Percentage Imperviousness: About 2/3 of the
monitoring sites currently have percentage imperviousness data.
·
Rainfall: Almost all of the events have the
rainfall amount associated with the monitored event.
·
Type of sample collection: Some of the events
represent special “first-flush” and composite sample pairs for the same event.
These data were evaluated previously to identify these effects on runoff water
quality. The type of sampler and sampling method has been identified for about
¼ of the sampling locations.
·
Runoff amount: About 1/3 of the events have the
runoff amounts associated with the monitored events.
·
Watershed area: All of the monitored locations
have the watershed areas identified.
·
Date of sample collection: All of the data are
associated with the date of sample collection. In addition to the seasonal
effects, this information can be used to examine any trends in concentration
that may have occurred during the 10 years of sample collection represented in the
NSQD.
·
Type of conveyance system: About 1/3 of the
sites have the conveyance system identified.
·
Aerial photographs and topographic maps have
been obtained for almost all of the monitoring areas.
Figure 11 is a probability plot for the observed COD concentrations
separated by land use. This plot is similar to the previously presented box and
whisker plots for the different constituents separated by land use. These plots
do show additional information that is useful for developing predictive models.
As typically assumed, the COD values closely follow log-normal probability
plots for much of the data range (Figure 2 illustrates log-normal probability
plots for many of the constituents available in the NSQD, but grouped for all
factors combined). Figure 11 shows significant differences by land uses. The
open space COD concentrations are the lowest, and the freeway COD
concentrations are the largest for most all of the data range. The residential,
commercial, and industrial areas are very similar for the lower half of the
distribution, while the residential areas are lower than the commercial and
industrial areas in the upper portion of the distribution. The effects of some
of the above listed factors on concentrations have been previously illustrated.
The following shows how we plan to develop the predictive tool for the main
watershed factors listed above. In this example, we will examine COD
concentrations as a function of EPA rain zones and season, for the residential
areas.
Figure 11. Probability plots of COD concentrations for different land uses.
It is possible to identify statistically significant differences
in the COD concentrations for residential land uses in different EPA zones and
seasons. Table 5 shows the total number of storm events collected which has residential
COD values for the different rain zones and seasons.
Table 5. Number of Events with Detected COD Values in Residential Land Use
Areas in the NSQD
|
EPA Rain Zone
|
Total
|
Spring
|
Summer
|
Fall
|
Winter
|
|
1
|
6
|
1
|
5
|
-
|
-
|
|
2
|
490
|
116
|
102
|
135
|
137
|
|
3
|
53
|
12
|
10
|
14
|
17
|
|
4
|
43
|
9
|
15
|
8
|
11
|
|
5
|
95
|
39
|
5
|
22
|
29
|
|
6
|
44
|
7
|
19
|
6
|
12
|
|
7
|
49
|
15
|
1
|
18
|
15
|
|
8
|
7
|
3
|
1
|
3
|
-
|
|
9
|
-
|
-
|
-
|
-
|
-
|
Table 5 shows that EPA rain zone 2 has about 62% of the
total number of COD observations in the database. This unbalance of sample
numbers can potentially lead to confusing results if the other areas do not
adequately represent the actual conditions in their areas and is a violation of
the data assumptions needed for a successful ANOVA test. It is possible to see
if there is a difference in the COD concentrations for the different seasons in
each zone during the four seasons using a one-way analysis of variance test, as
the numbers of samples in each season for each main zone are relatively even.
The analysis of variance requires that the residuals are
normally distributed and there is the same variance for each of the seasons.
After log transforming the data, it was found that the residuals can be
considered normal with a p-value of 0.8 using the Kolmogorov-Smirnov Goodness
of fit test. To test if the variances are the same for the four seasons, Barlett’s
test was used. This test is powerful when the normality assumption of the
residuals is achieved, as in this example. The results indicated that the
variance can be considered the same for each season in EPA rain zone 2, with a
p-value of 0.44. The results of the ANOVA found that there is a significant difference
in the COD concentrations during the four seasons. The COD concentration in EPA
rain zone two during winter seems to be smaller than summer and spring. The
pooled standard deviation of the observations was calculated as 0.677
Figure 12 is a set of power curves showing the difference in
the mean COD concentrations for the different subgroups that can be identified
for different numbers of samples. If the ANOVA test indicated a significant
difference with a confidence of five percent (a=0.05),
these mean differences can be detected for the noted sample sizes. Table 6
lists the sample sizes needed, for a power level of 0.8 and a confidence of
0.05, to detect the noted differences in mean concentrations. If a goal of at
least a 25% difference was desired, then about 120 samples in each season would
be needed. This is approximately the conditions for EPA rain zone 2 residential
land uses. However, if only 10 samples are available for each season, then the
“detectable” difference would be relatively large (larger than 50%).
Figure 12. Power of the one way ANOVA test for COD in EPA rain zone 2.
Table 6. Samples required to detect specific differences in the COD for
different seasons
|
Percentage difference between the mean values (%)
|
Samples Required
|
|
5
|
3844
|
|
10
|
908
|
|
20
|
202
|
|
25
|
122
|
|
30
|
80
|
|
40
|
40
|
|
50
|
22
|
A two-way analysis can also be conducted to examine the
effects of both seasons and rain zones together, and their interaction. In the
following example, rain zones 3, 4, 5, 6, and 7 were evaluated for all four
seasons. Rain zone 2 was excluded from this preliminary analysis because it had
many more samples than the other regions and could have overly emphasized those
conditions. The first step in this analysis is to check the distributions and
variances of the data sets. The residuals (the differences of the observations
from the mean) can be considered normal as they had a p-value higher than 15%
(no significant difference from a normal distribution). Barlett’s test also
indicated that the variance for the different groups can be considered the same
with a p-value of 0.35. A two-way ANOVA can therefore be used to identify any
differences between the seasons and EPA rain zones, plus their interaction,
because the data were normally distributed and they have the same variance
within each group.
The 2-way ANOVA results indicated that there are no significant
differences between the different seasons (p-value = 0.091), but that there is
a difference between the EPA rain zones (p-value < 0.001). Figure 13 contains
probability plots of the residential COD values for each season, showing no
clear distinction of these concentrations for the different seasons. The ANOVA
test also found no significant interaction between rain zone and season
(p-value = 0.25).
Figure 13. Probability plots of residential COD concentrations for
different seasons.
Figure 14 shows probability plots of residential area COD concentrations
for each EPA rain zone. There are likely three distinct groupings for residential
COD values, based on their geographical location. Samples collected in zone 6
had the highest mean concentrations and were collected in Arizona.
Samples collected in zones 2, 4 and 5 were intermediate in COD concentration
and were collected in the mid Atlantic states and Texas.
Samples collected in zones 3 and 7 had the lowest COD concentrations and were
collected in Alabama, Georgia,
and in Oregon.
Figure 14. Probability plots of residential COD concentrations in different
EPA Ran Zones.
Therefore, COD
residential area concentrations can be divided into the following three groups,
based on EPA rain zone:
Zones 3 and 7: average:
44.4 mg/L, standard deviation: 41.9 (102 observations)
Zones 2, 4 and 5: average:
72.8 mg/L, standard deviation: 61.6 (628 observations)
Zone 6: average: 162.1
mg/L, standard deviation: 100.0 (44 observations)
Overall residential
COD: average: 74.1, standard deviation: 69.2 mg/L
The statistical
analyses of the available NSQD COD residential area data did not identify any
significant differences in any rain zones that can be explained by season. There
was insufficient data in zones 1, 8, and 9 to be evaluated by season and the
overall residential COD values should therefore be used for those areas until
additional data is collected and evaluated.
Clustering and principal component analyses (PCA) are also
being used to identify expected factors influencing sample variability. Figure 15
is an example dendogram from a cluster analysis of all of the preliminary data
combined. However this analysis did not include most of the site
characteristics when it was conducted; only rain depth, watershed size, and
percentage imperviousness were included for this analysis, in addition to the
runoff concentrations. This plot indicates very close relationships between
rain depth and the nutrients (total phosphorus, dissolved phosphorus, nitrite
plus nitrate, ammonia, and Total Kjeldahl Nitrogen). Some of the heavy metals
(cadmium, nickel, and chromium) are closely related to each other, but copper,
lead and zinc are much more independent. BOD5, COD, dissolved
solids, and suspended solids are poorly related to other pollutants for the
pooled data. Pearson correlation analyses did show relatively strong relationships
between suspended solids and the total forms of most of the heavy metals,
substantiating the observation that most of the stormwater metals are not in
filtered forms.
Figure 15. Cluster analysis (dendogram) showing relationships between
stormwater pollutants.
A number of sampling issues can be statistically
investigated using the information contained in the NSQD. The following
discussion is a summary of the types of monitoring guidance that can be developed
and refined using the database information.
An important aspect of any research is the assurance that
the samples collected represent the conditions to be tested and that the number
of samples to be collected are sufficient to provide statistically relevant
conclusions. An experimental design process can be used that estimates the
number of needed samples based on the allowable error, the variance of the
observations, and the degree of confidence and power needed for each parameter.
The number of samples needed is therefore dependent on the objectives of the
data (characterization, comparison, trends, etc.), the variation of the
concentrations in the category being investigated (typically described by the
coefficient of variation, or the ratio of the mean to the standard deviation),
and the allowable errors (the confidence and the power).
A basic equation that can be used to estimate the number of
samples to characterize a set of conditions (given in Burton
and Pitt 2001) is as follows:
n
= [COV(Z1-a + Z1-b)/(error)]2
where:
n = number of
samples needed
a=
false positive rate (1-a is the degree of confidence. A value of a of
0.05 is usually
considered statistically significant, corresponding to
a 1-a
degree of confidence of 0.95, or 95%.)
b=
false negative rate (1-b is the power. If used, a value of b of
0.2 is
common, but it is
frequently and improperly ignored, corresponding to a b of 0.5.)
Z1-a
= Z score (associated with area under normal curve) corresponding to
1-a. If a is
0.05 (95% degree of confidence), then the
corresponding Z1-a
score is 1.645 (from standard statistical tables).
Z1-b=
Z score corresponding to 1-b value. If b is 0.2 (power of 80%), then
the corresponding
Z1-b
score is 0.85 (from standard statistical
tables). However,
if power is ignored and b is 0.5, then the
corresponding Z1-b
score is 0.
error
= allowable error, as a fraction of the true value of the mean
COV
= coefficient of variation (sometimes noted as CV), the standard deviation
divided by the
mean (Data set assumed to be normally distributed.)
This equation assumes a normal distribution of the data,
which would require a log transformation of most stormwater quality data. If an
allowable error of about 25% is desired and the COV is estimated to be 0.4,
then about 20 samples would have to be analyzed. The use of stratified random
sampling can usually be used to advantage by significantly reducing the COV of
the sub-population in the strata, requiring fewer samples for characterization.
The COV values for many constituents shown in Table 1 for
the NPDES database range from unusually low values of about 0.1 (for pH) to
highs between 1 and 2. There are a few COV values that are larger. One
objective of a data analysis procedure is to categorize the data into separate
stratifications, each having small variations in the observed concentrations.
The only stratification in Table 1 is land use. However, Figure 6 shows many
differences by geographical area (refer to Figure 1 for the EPA Rain Zone map).
It is expected that the final data analyses for this project will identify
separate stratifications of data (possibly considering the combination of land
use, geographical area, and season factors) to significantly reduce the
variations in each category. It is expected that COV values in the range of 0.5
to 1.0 will be common for many of these data stratifications. With a reasonable
confidence of 95% (a=
0.05) and power of 80% (b= 0.20), and a commonly accepted allowable error of 25%,
the number of samples needed to characterize conditions would likely range from
about 25 to 50. If only 12 samples are obtained for each category (strata), the
allowable errors would range from about 50% to 100%. Burton
and Pitt (2001) present many additional experimental design equations and plots
for other data quality objectives, including the effects of log transforming
the data for more appropriate sampling effort approximations. In many cases,
the actual errors in presenting data are larger than expected, due to
relatively small numbers of samples. A continuing monitoring program (such as
the Phase I stormwater NPDES permit monitoring effort) will result in better
data as more samples are obtained with time.
The NSQD can also be useful when selecting analytical
methods. There are many important factors that must be considered when
selecting an analytical method (availability, cost, detection limit,
repeatability, safety and disposal problems, comparisons with historical data,
etc.), but the detection limit is likely most important when ensuring the
suitability of the data. In many cases, analytical methods are used that have
detection limits that are actually larger than a criterion value, making
accurate exceedence frequencies impossible (Burton
and Pitt 2001).
Environmental
researchers need to be concerned with many attributes of numerous analytical
methods when selecting the most appropriate methods to use for analyses of
their samples. The main factors that affect the selection of an analytical
method include: cost, reliability (the “data quality objectives,” or DQO which
includes sensitivity, selectivity, repeatability), and safety. Most of these
issues are not well documented in the literature for environmental sample
analyses. Aspects of analytical reliability have received the most attention in
the literature, but most of the other aspects noted above have not been
adequately discussed for the many analytical alternatives available. It is
therefore difficult for a water quality analyst to decide which methods to
select, or even if a choice exists.
The selection of the
appropriate analysis procedure is dependent on the use of the data and how
false negatives or false positives would affect water use decisions or
regulatory questions. The QA objectives for the method detection limit (MDL)
and precision (RPD) for the compounds of interest have been shown to be a
function of the anticipated median concentrations in the samples (Pitt, et al. 1993). The MDL objectives should
generally be about 0.25, or less, of the median value for sample sets having
typical concentration variations (COV values ranging from 0.5 to 1.25), based
on many Monte Carlo evaluations to examine the rates of false negatives and
false positives. Table 7 lists the typical median stormwater runoff constituent
concentrations and the associated calculated MDL goals, for a typical
stormwater monitoring project.
Using analytical methods having these detection limits, at
least, will result in relatively few “non-detected” values. In most cases,
analytical methods are available that can easily meet these goals. However,
common problems are associated with some of the heavy metals, as most modern
laboratories use ICP (inductively-coupled plasma) instruments that are capable
of analyzing a broad range of metals simultaneously, but may not be able to
meet these detection limit goals. When dissolved forms of the heavy metals need
to be analyzed, the detection limits must be much smaller.
The NPDES stormwater database can be used to indicate the
likely concentrations of interest for conditions similar to those that will be
monitored. These expected values are a good start in determining the needed
detection limits.
Details for all monitoring locations are desired for the
database. Basic information (land use, season, geographic location, and if the
sample is a first-flush or a composite sample) is available for all events in
NSQD, and relatively complete site and monitoring descriptions are available
for about 1/3 of the events. This data includes sampling methods (automatic
samplers vs. manual samplers; manufacture and model of sampler; etc.). Investigations
of how these factors may influence the monitoring results will be made, as
illustrated in the initial evaluation of first-flush vs. composited samples.
The effects of automatic vs. manual sampling will also be examined when
sufficient information has been collected. One example of a previous
investigation on stormwater sampling methods was conducted by Roa-Espinosa and
Bannerman (1995). They collected samples from five industrial sites using
different monitoring methods. They concluded that many time-composited
subsamples combined for a single analysis can provide improved accuracy
compared to fewer samples associated with flow-weighted samplers, and
especially compared to samples only taken during a portion of an event.
A major goal of this project is to
provide guidance to stormwater managers and regulators. Especially important
will be the use of this data as an updated benchmark for comparison with
locally collected data. These comparisons will enable local monitoring data to
be compared to typical values that should be expected for similar situations.
If the local stormwater quality is significantly worse than expected, then it
may be possible to quantify a treatment goal that should be attainable. In
addition, this data may be useful for preliminary calculations when using the
“simple method” for predicting mass discharges for unmonitored areas. This data
can also be used as guidance when designing local stormwater monitoring
programs (Burton and Pitt 2002), especially when determining
the needed sampling effort based on expected variations. The final data
analyses will expand on these preliminary examples and will also investigate
other stormwater data and sampling issues.
Table 7. Example QA Objectives for a
Stormwater Characterization Project
|
Constituent
|
Units
|
Typical COV
category1
|
Typical Median
Conc.
|
Estimated MDL
Goal
|
|
Turbidity
|
NTU
|
low
|
5
|
4
|
|
COD
|
mg/L
|
medium
|
50
|
12
|
|
suspended solids
|
mg/L
|
medium
|
50
|
12
|
|
nitrates
|
mg/L
|
low
|
0.6
|
0.4
|
|
chromium
|
mg/L
|
medium
|
7
|
1.5
|
|
copper
|
mg/L
|
medium
|
15
|
3.5
|
|
lead
|
mg/L
|
medium
|
15
|
3.5
|
|
nickel
|
mg/L
|
medium
|
10
|
2.3
|
|
zinc
|
mg/L
|
medium
|
100
|
23
|
|
1,3-dichlorobenzene
|
mg/L
|
medium
|
10
|
2
|
|
benzo(a) anthracene
|
mg/L
|
medium
|
30
|
8
|
|
bis(2-ethylhexyl) phthalate
|
mg/L
|
medium
|
10
|
2.3
|
|
butyl benzyl phthalate
|
mg/L
|
medium
|
15
|
3
|
|
fluoranthene
|
mg/L
|
medium
|
6
|
1.4
|
|
pentachlorophenol
|
mg/L
|
medium
|
10
|
2
|
|
pyrene
|
mg/L
|
medium
|
5
|
1
|
|
lindane and chlordane
|
mg/L
|
medium
|
1
|
0.2
|
1 COV
value: Multiplier
for MDL
<0.5 (low) 0.8
0.5 to 1.25 (medium) 0.23
>1.25 (high) 0.12
from:
Burton and Pitt 2001
The example investigation of first-flush conditions
indicated that a first flush effect was not present in all the land uses and
certainly not for all the constituents. Commercial and residential areas were
more likely to show the phenomenon, especially if the peak rainfall occurred
near the beginning of the event. It is expected that the effect will be more
likely in watersheds with larger amounts of imperviousness. However, the
industrial category had large amounts of imperviousness, but indicated first-flushes
less than 50% of the time. All the metals evaluated show a higher concentration
at the beginning of the event in the commercial land use category.
The current data and information contained in NSQD indicates
the potential value that a completed database (containing most of the NPDES
stormwater data) can provide. The excellent U.S.
national coverage, along with the broad representation of land uses, seasons,
and other factors, makes this information highly valuable for numerous basic
stormwater management needs. Monitoring with no specific objective, except for
general characterization in an area, is not likely to provide any additional
value beyond the data and information contained in NSQD. After a sufficient
amount of data has been collected by a Phase 1 community for representative
land uses and other conditions, outfall characterization monitoring resources
should be re-directed to other specific data collection and evaluation needs. Burton
and Pitt (2001) provide much additional information on determining an adequate
outfall monitoring program. Similarly, communities that have not initiated a
stormwater monitoring program (such as the Phase II NPDES small communities)
may not require general characterization monitoring (monitoring is not
specifically required as part of the Phase II regulations), if they can
identify a regional Phase I community that has compiled extensive monitoring
data as part of their required NPDES stormwater permit. Obviously, there will
be some situations that are not well represented in NSQD and additional
characterization monitoring may be warranted. These situations will be
identified in the final data analyses.
This is not to say that stormwater quality monitoring has no
role as part of a stormwater management program. Burton
and Pitt (2001) present extensive examples and procedures showing the
importance of a balanced monitoring program. This publication is available from
CRC Press, and a version is available at:
http://civil.eng.ua.edu/~rpitt/Publications/BooksandReports/Stormwater%20Effects%20Handbook%20by%20%20Burton%20and%20Pitt%20book/MainEDFS_Book.html
Stormwater quality monitoring is a crucial component of
local programs. Specific objectives for these include:
·
Receiving water assessments to understand local problems. Receiving water
monitoring is needed to identify local problems, especially when identifying
beneficial use impairments. Assimilative capacity calculations (TMDLs) require
knowledge of local source discharges. The NSQD data and information can be used
for preliminary designs and cost estimates, but it is also important to invest
a small amount of resources to accurately determine local discharge conditions
before expensive controls are designed.
·
Source area monitoring to identify critical sources. In many cases, source area
controls may be more cost-effective than regional controls. The identification
of critical source areas is therefore needed as part of a comprehensive
stormwater management program. Monitoring within a critical drainage area
should be conducted to identify the sources of pollutants, while simultaneous
outfall monitoring is needed to verify these source area measurements.
·
Detailed monitoring at selected outfalls, with complete monitoring of rainfall
and runoff, with high-resolution data to examine time-variability
characteristics of certain problem pollutants. This would be especially
important at small, highly paved areas where “first-flush” conditions are most
likely. This information is needed to evaluate the benefits and to quantify
design approaches of critical source area controls.
·
Treatability tests to verify performance of stormwater controls for local
conditions. In areas where stormwater controls are being installed, local
measurements of performance are a good investment. Before and after monitoring,
or parallel monitoring, is usually needed to measure the performance of many
types of stormwater controls. The ASCE National Stormwater BMP database (http://www.bmpdatabase.org/) is a good
place to start in predicting the performance of controls, but site-specific
validations in an area where the controls have not been previously used should
be conducted.
·
Assessment monitoring to verify success of stormwater management approach.
Stormwater quality monitoring is a critical component of an assessment
monitoring effort. Receiving water monitoring needs to focus on beneficial use
impairments, and associated chemical, physical, and biological monitoring. In
many cases, source area or outfall controls are being used as part of
comprehensive management programs. Therefore, outfall monitoring may also be
needed.
Many people and
institutions need to be thanked for their help on this research project.
Project support and assistance from Bryan Rittenhouse, the US EPA project
officer for the Office of Water, is gratefully acknowledged. The many
municipalities who worked with us to submit data and information were obviously
crucial and the project could not be conducted without their help. Finally, the
authors would like to thank a number of graduate students at the University of Alabama (especially Veera Rao Karri, Sanju Jacob,
Sumandeep Shergill, Yukio Nara, and Soumya Chaturvedula) and employees of the
Center for Watershed Protection (Ted Brown, Chris Swann, Karen Cappiella, and
Tom Schueler) for their careful work on this project.
Auckland
Regional Council. Annual Report.
January-December 2001 Baseline Water Quality. Streams, Lake, and Saline Waters. Technical Publication 190. August 2002.
Bertrand-Krajewski,
J. “Distribution of Pollutant Mass vs. Volume in Stormwater Discharges and the
First Flush Phenomenon”. Water Resources,
Vol 32, No. 8 pp. 2341 – 2356. 1998.
Burton,
G.A. Jr., and R. Pitt, Stormwater Effects Handbook: A Tool Box for Watershed
Managers, Scientists, and Engineers. CRC Press, Inc., Boca
Raton, FL. 911 pgs. 2002.
Clark, S., R.
Pitt, and S. Burian. “Urban Wet Weather
Flows - 2002 Literature Review.” Water Environment Research. Vol. 75, No.
5, Sept./Oct. 2003 (CD-ROM).
Clark, S., R.
Pitt, and S. Burian. “Urban Wet Weather
Flows - 2001 Literature Review.” Water Environment Research. Vol. 74, No.
5, Sept./Oct. 2002 (CD-ROM).
Clark, S., R.
Rovansek, L. Wright, J. Heaney, R. Field, and R. Pitt. “Urban Wet Weather Flows - 2000 Literature Review.” Water
Environment Research. Vol. 73, No. 5, Sept./Oct. 2001 (CD-ROM).
Deletic, A.
“The First Flush Load of Urban Surface Runoff”. Water Resources, Vol 32, No 8, pp. 2462-2470, 1998.
Fan, C-Y, R. Field, J. Heaney, R. Pitt, S. Clark,
L. Wright, R. Rovansek, and S. Olivera. “Urban Wet Weather Flows 1999 Literature Review.” Water
Environment Research. Vol. 72, No. 5, Sept./Oct. 2000 (CD-ROM), 199 pgs.
Field, R., T.
O’Connor, C-Y. Fan, R. Pitt, S. Clark, J. Ludwig, and T.
Hendrix. “Urban Wet Weather Flows - 1997 Literature Review.” Water Environment Research. Vol. 70, No.
4, June 1998.
Field, R., R.
Pitt, Hsu, K., M. Borst, R. DeGuida, C-Y. Fan, J. Heaney, J. Perdek, and M.
Stinson. “Urban Wet Weather Flow - 1996 Literature Review.” Water Environment Research. Vol. 69, No.
4, pp. 426-444. June 1997.
Fligner M.
Policello, G. “Robust Rank Procedures for the Behrens-Fisher Problem”. Journal of the American Statistical
Association, Volume 76, Issue 373. pp 162-168. 1981.
Maestre, A., Pitt, R. E., and R. Morquecho.
“Nonparametric statistical tests comparing first flush with composite samples
from the NPDES Phase 1 municipal stormwater monitoring data.” Stormwater and
Urban Water Systems Modeling. In: Models and Applications to Urban Water
Systems, Vol. 12 (edited by W. James). CHI. Guelph, Ontario, forthcoming
2004.
O’Connor, R.
Field, D. Fischer, R. Rovansek, R. Pitt, S. Clark, and
M. Lama. “Urban Wet Weather Flows - 1998 Literature Review.” Water Environment Research. Vol. 71, No.
4, June 1999.
Pitt, R., R.
Field, M. Lalor, and M. Brown. “Urban Stormwater Toxic Pollutants: Assessment,
Sources and Treatability.” Water
Environment Research. Vol. 67, No. 3, pp. 260-275. May/June 1995.
Pitt, R., M.
Lalor, R. Field, D.D. Adrian, and D. Barbe’. A User’s Guide for the Assessment of Non-Stormwater Discharges into
Separate Storm Drainage Systems. U.S.
Environmental Protection Agency, Storm and Combined Sewer Program, Risk
Reduction Engineering Laboratory. EPA/600/R-92/238. PB93-131472. Cincinnati,
Ohio. 87 pgs. January 1993.
Pitt, R.
Maestre A., Morquecho R. “Evaluation of NPDES phase I Municipal Stormwater
Monitoring Data” In: National Conference
on Urban Stormwater: Enhancing the Programs at the Local Level.
EPA/625/R-03/003. February 2003
Roa-Espinosa
A. Bannerman, R. “Monitoring BMP effectiveness at Industrial Sites”. In: Stormwater NPDES Related Monitoring Needs.
Edited by Harry C. Torno. pp 467-486. 1995.
Smullen, J.T.
and K.A. Cave,
“National stormwater runoff pollution database.” In: Wet-Weather Flow in the
Urban Watershed, edited by R. Field and D. Sullivan. Lewis Publishers. Boca
Raton, pgs. 67 – 78. 2002.
U.S.
Environmental Protection Agency, Dec. Results
of the Nationwide Urban Runoff Program. Water Planning Division, PB
84-185552, Washington, D.C.
1983.
