See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/8330885

The Influence of Urban Density and Drainage Infrastructure on the

Concentrations and Loads of Pollutants in Small Streams

Article  in  Environmental Management · August 2004

DOI: 10.1007/s00267-004-0221-8 · Source: PubMed

CITATIONS READS
426 2,114

4 authors, including:

Belinda Hatt Tim D Fletcher

Monash University (Australia) University of Melbourne
81 PUBLICATIONS   2,860 CITATIONS    268 PUBLICATIONS   10,643 CITATIONS   

SEE PROFILE SEE PROFILE

Christopher J Walsh
University of Melbourne
88 PUBLICATIONS   6,397 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Urban stream restoration View project

New monitoring technologies serving a better understanding of land and river degradation in urban areas View project

All content following this page was uploaded by Tim D Fletcher on 10 February 2015.

The user has requested enhancement of the downloaded file.

 DOI: 10.1007/s00267-004-0221-8

The Influence of Urban Density and Drainage

Infrastructure on the Concentrations and Loads of
Pollutants in Small Streams
BELINDA E. HATT* perviousness. Fifteen small streams draining independent
Cooperative Research Centre for Freshwater Ecology subbasins east of Melbourne, Australia, were sampled for a
Water Studies Centre suite of water quality variables. Geometric mean concentra-
Monash University tions of all variables were calculated separately for baseflow
Victoria 3800, Australia and storm events, and these, together with estimates of
runoff derived from a rainfall-runoff model, were used to
TIM D. FLETCHER
estimate mean annual loads. Patterns of concentrations
Cooperative Research Centre for Catchment Hydrology
among the streams were assessed against patterns of im-
Institute for Sustainable Water Resources
perviousness, drainage connection, unsealed (unpaved)
Department of Civil Engineering
road density, elevation, longitude (all of which were inter-
Monash University
correlated), septic tank density, and basin area. Baseflow
Victoria 3800, Australia
and storm event concentrations of dissolved organic car-
CHRISTOPHER J. WALSH bon (DOC), filterable reactive phosphorus (FRP), total phos-
SALLY L. TAYLOR phorus (TP) and ammonium, along with electrical conduc-
Cooperative Research Centre for Freshwater Ecology tivity (EC), all increased with imperviousness and its correlates.
Water Studies Centre Hierarchical partitioning showed that DOC, EC, FRP, and
Monash University storm event TP were independently correlated with drainage
Victoria 3800, Australia connection more strongly than could be explained by chance.
Neither pH nor total suspended solids concentrations were
ABSTRACT / Effective water quality management of streams strongly correlated with any basin variable. Oxidized and total
in urbanized basins requires identification of the elements of nitrogen concentrations were most strongly explained by sep-
urbanization that contribute most to pollutant concentrations tic tank density. Loads of all variables were strongly correlated
and loads. Drainage connection (the proportion of impervious with imperviousness and connection. Priority should be given
area directly connected to streams by pipes or lined drains) is to low-impact urban design, which primarily involves reducing
proposed as a variable explaining variance in the generally drainage connection, to minimize urbanization-related pollut-
weak relationships between pollutant concentrations and im- ant impacts on streams.

Runoff from urban areas is one of the leading sewerage systems, stormwater runoff is the primary de-
sources of water quality degradation in surface waters grader of streams (Walsh 2000). Increased impervious-
(U.S. Environmental Protection Agency 2000). Urban- ness (the proportion of a basin covered by surfaces
ization degrades stream ecosystems in a variety of ways impermeable to water, such as roofs and roads) and
that are not easy to separate: increased frequency and hydraulically efficient drainage systems typical of urban
intensity of flood flows, decreased groundwater levels, areas, result in increased frequency and intensity of
increased stream bank erosion, and increased loads of flood flows, decreased groundwater levels, increased
pollutants (Novotny and Olem 1994), with resulting stream channel and bank erosion, and increased loads
multiple impacts on aquatic ecosystems (Paul and and concentrations of pollutants (Novotny and Olem
Meyer 2001). In cities with separate storm and sanitary 1994).
Pollutants running off urban areas are difficult to
KEY WORDS: Urbanization; Stormwater runoff; Impervious area; Drain- measure and regulate because they ultimately arise
age connection; Catchment; Water quality
from a multitude of activities and are variable in time
Published online May 28, 2004.
because of the effects of weather. Because of these
*Author to whom correspondence should be addressed, email: difficulties, models that predict runoff and pollutant
Belinda.Hatt@eng.monash.edu.au loads are required to estimate the impacts of urbaniza-

Environmental Management Vol. 34, No. 1, pp. 112–124 © 2004 Springer Science⫹Business Media, Inc.
 Stream Pollutants and Urbanization 113

tion in urban waterways and receiving waters (Duncan of total imperviousness, a comparison of their relative
1999). Furthermore, effectively managing the impacts importance is best achieved by comparing their inde-
of urbanization requires identification of the elements pendent parts. Effective imperviousness can be consid-
of urbanization that contribute most to pollutant loads. ered the product of total imperviousness and drainage
Although basin-scale features are recognized as impor- connection (defined as the proportion of all impervi-
tant drivers of instream condition (Johnson and others ous areas directly connected to streams). We aim to
1997, Gergel and others 2002), the mechanisms of this assess the relative importance of these two elements of
behavior have not yet been clearly described (Duncan effective imperviousness in explaining concentrations
1999, Brezonik and Stadelmann 2002). of pollutants.
Land-use zoning has commonly been used as a pre- We use geometric means of baseflow and storm
dictor of water quality (Soranno and others 1996, Car- event concentrations and a rainfall-runoff model to
penter and others 1998). However, assigning a pattern estimate loads of each water quality variable. We then
of pollutant output to a particular land use provides assess whether the effects best explaining variation in
little insight into the processes impacting on aquatic concentrations among the 15 streams are the same as
systems, or how to manage that land use to reduce its those best explaining loads. We propose that effective
pollutant output. The U.S. EPA’s Nationwide Urban imperviousness is likely to be a good predictor variable
Runoff Program found land use to have little power in of loads and concentrations for predictive models of
explaining site-to-site differences in pollutant loads the effects of urban land use.
(Novotny and Olem 1994).
Two elements of urban land use— basin impervious- Methods
ness and drainage infrastructure—are major contribu-
tors to changes in hydrology arising from stormwater Study Area
runoff (Leopold 1968). Relationships between imper- Sites on 15 first- or second-order streams with similar
viousness and water quality have emerged in recent riparian cover, draining independent subbasins of sim-
years (Arnold and Gibbons 1996, May and others 1997, ilar area, spanning the eastern edge of the Melbourne
McMahon and Cuffney 2000), and a predictive relation- metropolitan area, Victoria, Australia, were chosen for
ship between pollutant loads and basin imperviousness the study (Figure 1). These sites represent a rural-to-
has been promoted (Center for Watershed Protection urban gradient and were selected to encompass as wide
2003). However, correlations between pollutant loads a range as possible of both imperviousness and drain-
and imperviousness tend to be weak. We postulate that age connection (Figure 2, Table 1). In selecting sites,
the hydraulic efficiency of drainage infrastructure can we also aimed to minimize variation in physiographic
explain a large proportion of the unexplained variation and climatic conditions, and to minimize spatial con-
in the relationship between pollutant loads and imper- founding of subbasin characteristics. As a result, most
viousness. streams were located within the Dandenong Ranges,
Drainage efficiency has long been recognized as an east of Melbourne. One of these streams had near zero
important determinant of hydrological change subbasin imperviousness. A second near-zero impervi-
(Leopold 1968), accounting for up to 80% of the in- ous subbasin further to the east of the Dandenong
crease in peak discharge in urbanized basins (Wong Ranges was also chosen. Eleven other Dandenong
and others 2000). However, no published studies have Ranges streams ranging in imperviousness from 2.2%
assessed the importance of stormwater drainage con- to 12% were selected. Two streams with heavily urban-
nection to pollutant loads or concentrations in receiv- ized subbasins were selected in the suburbs to the west
ing waters. This lack of basin-scale study is surprising, of the ranges (Figure 1).
considering that most stormwater abatement tech- To ensure that urban stormwater runoff was the
niques reduce drainage efficiency through infiltration major anthropogenic impact in the study streams, only
or retention for treatment (Novotny and Olem 1994). subbasins with primary land-uses of urban or forest
Our hypothesis of the importance of hydraulic effi- were selected. Subbasins with intensive agriculture were
ciency of stormwater drainage is consistent with the excluded. A second possible impact across the study
suggestion of “effective imperviousness” as a better pre- area that we have assessed is polluted subsurface flow
dictor of stream degradation than total imperviousness from poorly maintained septic tanks.
(Booth and Jackson 1997). Effective impervious areas
are defined as those impervious areas directly con- Determination of Subbasin Characteristics
nected to streams or receiving waters by pipes or lined Seven environmental variables were considered as
channels. Because effective imperviousness is a subset potential correlates with water quality variables (Table
114 B. E. Hatt and others

Figure 1. Map of the study area.

The shaded area indicates the
Melbourne metropolitan area.
The fifteen subbasins chosen are
outlined with dashed lines.

1). Four were indicators of potential impacts of urban both draining a settlement on the main ridge of the
land use: imperviousness (the proportion of basin area Dandenong Ranges. Because we are primarily inter-
covered by impervious surfaces), drainage connection ested in assessing the impacts of stormwater pipes di-
(the proportion of impervious surfaces directly con- rectly delivering stormwater to streams, this settlement
nected to streams by pipes or lined channels), septic was considered unconnected.
tank density, and unsealed (unpaved) road density
(area of unsealed roads to basin area). Three variables, Data Collection
basin area, elevation, and longitude were potentially Water quality was sampled at the 15 sites during
correlated variables that may more strongly indepen- baseflow and storm events from September 2001 to
dently explain observed patterns than anthropogenic March 2003 (Table 1). Samples were collected every 2
impacts. weeks, along with additional storm event samples from
Elevation and longitude (measured as kilometers September 2001 to November 2002. After November
east of the most westerly site) were estimated from the 2002, only storm event samples were collected. To avoid
Victorian 1:25 000 topographic digital map series (URL biasing results with respect to time of day, samples
http://www.land.vic.gov.au). Basin areas were delin- taken every 2 weeks were collected from the 15 sites in
eated using 10-m contours from the topographic map a random order on each sampling trip. Event sampling
data and local government digital data of stormwater occurred on a response basis with the aim of capturing
drainage lines. Septic tank density was determined us- the variation in water quality experienced between and
ing a local government database of septic tank loca- within storm events. Event sampling took two forms: (1)
tions. Impervious areas (and unsealed road areas) were single grab samples taken at all sites; and (2) multiple
mapped using digital road data, local government samples taken manually at Em, Fe, Ga, Ly, Ol, and Sc
building area data, and aerial orthophotography. The (Table 1). In addition to the manual sampling, au-
connection status of impervious areas was estimated tosamplers connected to a float switch (Sigma Model
from proximity to stormwater drains, allowing for local 900 Standard Portable Sampler) were installed at Br
topography, and was checked by ground truthing. The and Do in September 2002 to sample storm events.
methods of determination were described in more de- Samples collected by the autosamplers were processed
tail by Walsh and others (2004). In that study, areas within 36 hours of collection.
drained by stormwater pipes, but in turn draining to Water quality variables measured were water temper-
dry earthen or grassed channels, were treated as am- ature, pH, electrical conductivity (EC), total suspended
biguously connected, and two sets of analyses were run: solids (TSS), total nitrogen (TN), total phosphorus
one with ambiguous areas treated as connected and (TP), filterable reactive phosphorus (FRP), ammonium
another with them treated as unconnected. In our (NH4⫹), nitrate/nitrite (NOx) and dissolved organic
study, only two sites were affected by this distinction, carbon (DOC). Analyses were undertaken by the NATA
 Stream Pollutants and Urbanization 115

calculated separately for baseflow and storm-event data.

Samples taken every 2 weeks were classified as being
taken in either baseflow or storm-flow conditions, using
a two-step process. First, frequency analysis (using the
data density distributions) was used to identify the
threshold between baseflow and storm flow. Seasonal
variation was then taken into account by examining the
flow data over time, and identifying whether a variable
threshold should be applied across seasons. In most
cases, there was very little or no seasonal influence.
For each date on which multiple samples from a
storm event were collected, a flow-weighted mean con-
centration (EMC in mg/L) was calculated, as follows
(Eq. 1):

冘共共t
n

i⫹1 ⫺ t i 兲 ⫻ 0.5共Q i ⫹ Q i⫹1 兲共c i ⫹ c i⫹1 兲兲

1
EMC ⫽

冘共共t
n

i⫹1 ⫺ t i 兲 ⫻ 0.5共Q i ⫹ Q i⫹1 兲兲

1

(1)
where: n ⫽ number of samples collected during the
event, ti ⫽ time sample i was taken, ci ⫽ concentration
for sample i, and Qi ⫽ flow at time ti.
Because EMCs were not calculated for all streams,
geometric mean and median storm-flow concentrations
were calculated in two ways for analysis. First, means
and medians were calculated using only storm-flow data
Figure 2. Relationships between percentage imperviousness from the two-weekly sampling. Second, the mean and
and (a) percentage drainage connection, (b) septic tank den- median for each site were calculated using the set of
sity, (c) unsealed road density, (d) subbasin area, (e) eleva- concentrations from all single storm-flow samples from
tion and f) longitude, in the 15 study sites. Consistent with
the two-weekly data, together with all EMCs calculated
analyses, axes have been transformed.
for that site. The number of baseflow and storm event
accredited (URL http://www.nata.asn.au/) Water samples collected, and EMCs calculated for each site
Studies Centre analytical laboratory, using standard are shown in Table 1.
methods and quality control/assurance procedures The relationships between concentrations (baseflow
(Table 2) (Water Studies Centre 2001). and storm event, median and geometric mean) of all
Staff gauges were installed at each site, and flow water quality variables and the seven subbasin variables
heights were recorded when water quality samples were were assessed using multiple linear regression. Analyses
taken. Flow was estimated at each site using an appro- were undertaken on transformed data in order to min-
priate hydraulic model, constructed with HEC-RAS 3.0 imize the influence of outliers and to ensure that re-
(Hydrologic Engineering Centre 2002), or Manning’s sidual distributions approximated normality. Impervi-
equation if a suitable culvert was near the sampling ousness and subbasin area were fourth-root
location. The hydraulic models were calibrated with transformed, while drainage connection, septic tank
limited flow measurement data (velocity measurements density, elevation, median NOx and temperature, and
undertaken using a flow meter: Hydrological Services, the geometric means of DOC and temperature were
Model C.M.C 20). square-root transformed. Longitude, median TN, and
pH were untransformed. Median DOC, EC, FRP, NH4⫹,
Data Analysis TP and TSS, and all other geometric means were all
Factors explaining patterns in concentrations. Median log10-transformed. All reported correlation strengths
and geometric mean concentrations for each site were are based on the transformed data.
116 B. E. Hatt and others

Table 1. Summary of subbasin characteristics: subbasin area (Area), longitude (distance east of the most westerly
site), elevation, imperviousness (Imp), drainage connection (Conn), septic tank density (Septics), percentage of the
subbasin covered by unsealed roads (U Roads) and the number of baseflow, storm event grab samples collected
and event mean concentrations (EMC) calculated
No. samples
Area Longitude Elevation Imp Conn Septics U Roads
Subbasin name (ha) (km) (m) (%) (%) (N/km2) (%) Base flow Event flow EMC
Bungalook Ck (Bg) 579 16.2 133 6.8 57 52.8 0.58 21 13 —
Brushy Ck (Br) 1479 14.9 82 22 89 37.8 0.03 19 11 11
Dandenong Ck (Da) 424 16.8 154 2.5 17 61.5 1.33 20 11 —
Dobsons Ck (Do) 365 16.7 168 7.6 47 43.8 1.23 21 17 7
Hughes Ck (Hu) 275 16.9 173 3.5 11 62.5 1.76 15 15 —
Emerald Ck (Em) 188 23.0 290 2.2 0 58.6 1.00 20 9 1
Ferny Ck (Fe) 642 16.9 163 12 79 80.4 0.44 29 9 1
Gardiners Ck (Ga) 982 1.19 81 47 98 0 0 19 12 1
Lyrebird Ck (Ly) 724 23.1 222 0.1 0 0 1.74 18 15 1
McRae Ck (Mc) 1310 40.1 200 0.1 0 1.4 0.47 16 18 —
Olinda Ck (Ol) 907 23.1 222 3.9 3.0 88.2 1.51 18 13 1
Perrins Ck (Pe) 218 20.8 327 5.3 20 74.9 0.73 19 13 —
Sassafras Ck (Sa) 189 19.9 370 8.0 8.3 141 1.30 20 11 —
Scotchmans Ck (Sc) 812 0.00 78 39 99 0 0 17 14 1
Little Stringy bark Ck (St) 451 24.2 137 10 58 47.9 0.67 19 18 —

Table 2. Water quality variable collection and processing methods

Variable Method of collection and processing Detection limit
TSS, EC Unfiltered, clean polyethylene bottle, refrigeration prior to laboratory TSS: 0.5 mg/L
analysis (AMHA-AWWA-WPCF 1998)
Total nutrients (TN,TP) Unfiltered, clean polyethylene bottle, acidified, laboratory analysis 0.02 mg/L
(Hosomi and Sudo 1986)
Nutrient species Filtered (0.2 ␮m) into clean polyethylene bottle (in situ for manual 0.001 mg/L
(NH4⫹, NOx, FRP) samples; in laboratory for samples collected by autosamplers),
frozen prior to laboratory analysis (AMHA-AWWA-WPCF 1998)
DOC Filtered (0.2 ␮m) into clean polyethylene bottle, acidified, 1 mg/L
laboratory analysis (AMHA-AWWA-WPCF 1998)
pH, temperature Measured in situ using Horiba U-10 Water Quality Checker —
DOC, dissolved organic carbon; EC, electrical conductivity; TN, total nitrogen; TP, total phosphorus; FRP, filterable reactive phosphorus; NH4⫹,
ammonium; NOx, nitrate/nitrite; TSS, total suspended solids.

Hierarchical partitioning of R2 values was used to upper 95% confidence limit (Z-score ⱖ 1.65: Mac Nally
determine the proportion of variance explained inde- 2002, Walsh and Mac Nally 2003).
pendently and jointly by each variable (Chevan and One of the subbasins (Bg) was found to be an ex-
Sutherland 1991, Mac Nally 2000). This method allows treme outlier in terms of phosphorus concentrations.
identification of variables whose independent correla- Analyses for phosphorus were therefore conducted
tion with the dependent variable is strong, in contrast with Bg included and again with it omitted. We postu-
to variables that have little independent effect but have late that the phosphorus loads in Bg originated from a
a high correlation with the dependent variable result- nonstormwater origin, because FRP and TP were the
ing from joint correlation with other independent vari- only variables for which Bg was an outlier.
ables. Variables that independently explained a larger Calculation of standardized pollutant loads. Annual
proportion of variance than could be explained by loads of each water quality variable were estimated
chance were identified by comparison of the observed using geometric mean concentrations for base flow and
value of independent contribution to explained vari- storm flow at each site, together with a modeled esti-
ance (I) to a population of Is from 500 randomizations mate of total base flow and storm flow volumes. A
of the data matrix. Significance was accepted at the subdaily rainfall-runoff model (Model for Urban
 Stream Pollutants and Urbanization 117

Stormwater Improvement Conceptualisation: Wong enced by subbasins Sc, Ga, and Br (Table 1) with high
and others 2002) was used to generate base flow and imperviousness and drainage connection and subbasins
storm flow volumes for each basin. The rainfall-runoff Ly and Mc (Table 1) with near zero imperviousness and
model, based on the algorithms of Chiew and McMa- no drainage connection. However, drainage connec-
hon (1999), was run using a 12-minute timestep, based tion varied widely across a small range of impervious-
on a standardized rainfall record (Melbourne rainfall ness among the Dandenong Ranges sites. Impervious-
for the period January 1, 1990 to December 31, 2000: ness was significantly negatively correlated with
mean annual rainfall ⫽ 648 mm, mean annual areal unsealed road density (␳ ⫽ ⫺0.65; Figure 2c) and
potential evapotranspiration ⫽ 1051 mm). The exact longitude (␳ ⫽ ⫺0.84; Figure 2f). The latter correlation
quantum of, and geographic variation in, annual rain- was strongly influenced by two highly urbanized subba-
fall, is not important to this study. Our aim is to quan- sins in the west (Sc and Ga; Table 1) and a single
tify the pattern between effective imperviousness and subbasin in the east (Mc; Table 1); among the sites in
pollutant loads, for a given (constant) level of rainfall. the Dandenong Ranges, imperviousness was not
Actual loads measured at any location will of course be strongly correlated with longitude. Elevation was mod-
influenced by rainfall volume and distribution at that erately correlated with subbasin imperviousness (␳ ⫽
location. Although total annual rainfall varies across ⫺0.58; Figure 2e) and more strongly correlated with
the 15 basins, our objective was to determine the rela- drainage connection (␳ ⫽ ⫺0.75). Septic tank density
tive differences in loads between basins of differing showed a unimodal relationship with imperviousness:
effective imperviousness, rather than to quantify abso- septic tanks were sparse both in basins of low and high
lute differences in loads, as a function of climatic vari- subbasin imperviousness, but more abundant in the
ation. The rainfall-runoff model was run using an im- moderately urbanized basins (Figure 2b). Subbasin
pervious store (based on our estimates of effective area was not strongly correlated with imperviousness (␳
imperviousness), and a single pervious store, with store ⫽ 0.13, Figure 2d) or drainage connection (␳ ⫽ 0.28).
properties (impervious area initial loss, pervious area Therefore, of the seven variables considered, five (im-
capacity and initial storage, groundwater recharge and perviousness and drainage connection, and the in-
drainage rate) held constant. Estimated runoff volumes verses of elevation, longitude, and unsealed road den-
were therefore solely driven by the estimates of effective sity) were intercorrelated indicators of urban density.
basin imperviousness.
Loads were calculated based on two sets of mean Factors Explaining Patterns in Water Quality
storm-flow concentrations, calculated in the two ways Concentrations
described previously to ensure that exclusion of EMCs
made no difference to the results. Derived loads were Because the results for median and geometric mean
scaled to a per-hectare basis, for standardized compar- concentrations were nearly identical, only geometric
ison between subbasins. The strengths of correlations means are reported. Similarly, only storm event mean
between the elements of effective imperviousness and concentrations calculated using EMCs are reported be-
pollutant loads were compared qualitatively with the cause patterns for storm event mean and median con-
correlations for pollutant concentrations. No statistical centrations were unchanged if EMCs were included in
testing of the relationships between pollutant loads and calculations. For most water quality variables, patterns
impervious area has been undertaken, because the two were the same for baseflow and for storm event con-
variables are necessarily correlated (i.e., the modeled centrations: where there was a difference, it is noted
flow is itself a function of imperviousness). Nonethe- below.
less, the patterns of loads among the 15 sites allow us to Temperature, EC, and concentrations of DOC, FRP,
examine whether the effects of effective imperviousness TP, and NH4⫹ all increased with subbasin impervious-
on runoff volumes override other factors that may have ness (Figure 3). More than half of the explained vari-
been stronger correlates with concentrations. ation in these water quality parameters was jointly ex-
plained by imperviousness, drainage connection,
elevation, longitude, and unsealed road density: vari-
Results ables that were intercorrelated indicators of urban den-
sity. Independently of that joint correlation, drainage
Relationships Between Environmental Variables connection explained more of the variance in TP
Subbasin imperviousness and drainage connection (storm event only), DOC, EC, and FRP than could be
were correlated (Pearson correlation coefficient, ␳ ⫽ explained by chance (Table 3, Figure 3). Elevation was
0.89; Figure 2a). This correlation was primarily influ- also a strong independent correlate of variance in EC
118 B. E. Hatt and others

Figure 3. Relationships between geometric means of baseflow (closed circles, solid regression lines) and storm event (open circles,
dashed regression lines) concentrations and three environmental variables. R values for baseflow concentrations (storm event
concentrations in parentheses) for each graph and for the full model (all seven environmental variables) are indicated. Consistent with
analyses, axes have been transformed. The bar charts indicate percentage of explained variance in baseflow concentrations explained
by each of the seven environmental variables (conn, connection; imp, imperviousness; septic, septic tank density; unseal, unsealed road
density; elev, elevation; area, subbasin area; long, longitude) independently (solid bars) and jointly (open bars) as determined by
hierarchical partitioning. Variables marked with an asterisk were identified as significant independent correlates. Negative joint
variance indicates that the other variables are “suppressors” of the variable in question (Chevan and Sutherland 1991). DOC, dissolved
organic carbon; EC, electrical conductivity; FRP, filterable reactive phosphorus; NH4⫹, ammonium; TP, total phosphorus.
 Stream Pollutants and Urbanization 119

Table 3. Strength of correlations between median concentrations (base flow and storm event) and the two major
indicators of subbasin urbanization and variables identified by hierarchical partitioning as being significant
independent correlates.a
Pearson correlation coefficient Significant independent correlates
Base flow Storm event
Water quality
parameter Conn Imp Conn Imp Base flow Storm event
DOC 0.76 0.50 0.77 0.44 Conn, Elev Conn, Elev
EC 0.90 0.77 0.88 0.72 Conn Conn, Elev
FRP 0.71 0.48 0.81 0.58 Conn Conn
NH4⫹ 0.71 0.67 0.80 0.71 — —
NOx ⫺0.17 ⫺0.07 ⫺0.06 0.20 Septic Septic
pH ⫺0.29 ⫺0.46 ⫺0.28 ⫺0.50 Imp Imp
Tem 0.71 0.63 0.61 0.59 Elev —
TN ⫺0.04 0.09 0.28 0.38 Septic Septic
TP 0.55 0.38 0.67 0.47 — Conn
TSS ⫺0.45 ⫺0.29 0.13 0.17 — —
a
conn, drainage connection; imp, imperviousness; Elev, elevation; Septic, septic tank density; TN, total nitrogen; TP, total phosphorus. For other
abbreviations, see Table 2.

(storm event only), temperature (baseflow only), and similar regardless of the method used for calculating
DOC (Table 3, Figure 3). mean concentrations (i.e., with and without EMCs).
For DOC, EC, FRP, and TP, the correlation with
drainage connection was stronger than with impervi-
ousness. The correlations with imperviousness were Discussion
strongly influenced by the most and least impervious
sites, with wide variation within the intermediate imper- Concentrations
viousness range (Figure 3). Omission of the outlying Bg Urbanization was the most likely primary determi-
from phosphorus analyses generally improved the fit of nant of stream water quality degradation. Both drain-
relationships but did not alter the significance of the age connection and imperviousness, as subbasin scale
independent contributions of the environmental vari- indicators of urban density, explained much of the
ables. observed variation in pollutant concentrations. In par-
For pH and TSS, the overall correlations with imper- ticular, we hypothesize that efficiency of runoff delivery
viousness, drainage connection, or other environmen- from impervious areas is a cause of increased EC and
tal variables were weak. Imperviousness was the stron- concentrations of DOC, EC, FRP, and TP. Drainage
gest independent correlate of variance in pH (Table 3). pipes bypass subsurface pathways of water flow, both in
NOx and TN were not strongly correlated with im- the riparian and nonriparian sections of basins, result-
perviousness or drainage connection, but were strongly ing in direct transportation of stormwater to receiving
correlated with septic tank density (␳ ⫽ 0.80 and 0.77, waters, with little or no terrestrial processing of nutri-
respectively), which independently accounted for more ents and pollutants. This bypassing, and the associated
than a quarter of the explained variation in both NOx reduction in water table levels, result in riparian vege-
and TN (Table 3, Figure 3). tation being “hydrologically isolated” from uplands and
streams, further reducing the opportunity for riparian
Patterns in Mean Annual Loads zones to influence the quality of stream water (Groff-
Mean annual loads of all water quality variables were man and others 2002).
strongly positively correlated with both imperviousness Several water quality indicators were more strongly
and drainage connection (and therefore effective im- correlated with drainage connection than with imper-
perviousness: Figure 4). This was the case even for viousness, suggesting that, at least at low levels of im-
those variables for which concentrations were not perviousness, stream water quality degradation is con-
strongly correlated with imperviousness or connection: trolled more by the manner in which impervious areas
NOx, TN, and TSS. In contrast to the case for concen- are connected to receiving waters than by the presence
trations, NOx and TN loads were poorly correlated with of impervious areas alone. Previous studies (e.g. Ban-
septic tank density. The strengths of correlations were nerman and others 1993, Schueler 1994, Booth and
120 B. E. Hatt and others

Figure 4. Relationships between mean

annual loads (estimated using single event
samples and event mean concentrations)
and drainage connection, imperviousness,
and effective imperviousness. FRP, filter-
able reactive phosphorus; NH4⫹, ammo-
nium; NOx, nitrate/nitrite; TN, total ni-
trogen; TP, total phosphorus; TSS, total
suspended solids.
 Stream Pollutants and Urbanization 121

Jackson 1997) have acknowledged the potential impor- linity in streams with increased basin urbanization to a
tance of the degree to which impervious areas are combination of urban sources such as atmospheric dep-
directly connected to streams, but the relative impor- osition, building materials, and highways, and to accel-
tance of such connection has not been well investi- erated chemical denudation of the basin. Our finding
gated. that drainage connection independently explains a
The importance of drainage connection as a corre- large proportion of the variation in EC suggests that
late with concentrations of several variables, indepen- enhanced stormwater drainage efficiency further in-
dently of the correlation with imperviousness, suggests creases the transport of ions from urban sources to
that drainage connection may be an important cause of streams. Efficient stormwater drainage also decreases
observed variation in water quality among streams with groundwater recharge (Rose and Peters 2001, Groff-
similar levels of imperviousness. Because drainage con- man and others 2002). The resultant reduced interac-
nection is independent of the amount of impervious tion with groundwater may also at least partly explain
area being drained by the stormwater pipes, it is sur- this relationship between drainage connection and EC.
prising that connection was a stronger overall correlate Suspended solids concentrations in streams around
with several variables than was imperviousness. The the world have been reported to be weakly correlated
observed strong relationship suggests that even very with density of urban land use, with higher concentra-
small proportions of impervious area are capable of tions flowing from urbanized basins than from forested
increasing pollutant concentrations, as long as there is basins (Duncan 1999). In our study, TSS concentra-
a delivery mechanism between the impervious area and tions were not strongly correlated with urban density,
the stream. The strength of the correlations with drain- and if the three geographical outliers (Sc, Ga, Mc;
age connection may have been a result of the structure Table 1) were excluded, TSS was negatively correlated
of our dataset. Most of the variation in water quality was with urban density (␳ ⫽ ⫺0.64 with imperviousness and
among sites in a relatively narrow range of impervious- ␳ ⫽ ⫺0.74 with connection). This pattern, contrasting
ness (0 –12%), with a wide range of drainage connec- with world trends (Duncan 1999), is likely to be an
tion. It is likely that, for a wider range of total impervi- artefact of the marine sedimentary geology of the
ousness, effective imperviousness would be a stronger slopes of the Dandenong Ranges, which produces high
predictor of water quality degradation than drainage silt loads. It is possible that the effects of increased
connection alone. imperviousness and piping or lining of drainage chan-
Drainage connection was important even for base- nels in these basins may have resulted in lower than
flow concentrations, suggesting that there may be natural TSS concentrations.
sources of pollutants discharging to the stream via the Septic tank density was the dominant influence on
stormwater drainage system even when there is no di- NOx concentrations. Subbasins with the highest septic
rect rainfall runoff. It is possible, however, that this tank densities also had the highest concentrations of
pattern may be caused by residence time in streams NOx and the highest proportions of nitrogen present as
exceeding time between events. No published previous NOx. These high NOx concentrations are unlikely to
studies have compared baseflow and high flow concen- have a strong, direct impact on biota in streams drain-
trations; typically, all samples have been grouped to- ing Eucalyptus regnans forests (such as those studied
gether (but see Meador and Goldstein 2003, where here) because these streams tend to contain naturally
concentrations were flow-weighted) or only baseflow high NOx concentrations (Attiwill and others 1996)
(e.g. Johnson and others 1997) or EMCs (e.g. Brezonik and uptake is likely to be limited by the very low phos-
and Stadelmann 2002) have been considered. We sug- phorus concentrations.
gest that both baseflow and storm flow should be con- Hierarchical partitioning has proved a useful
sidered separately because, although it is storms that method for identifying environmental variables that are
contribute the majority of pollutant loads in these most likely to be causal drivers of pollutant concentra-
streams, baseflow concentrations are important to the tions in streams. This method differs from the more
stream ecosystem, because these are the conditions commonly used stepwise model selection methods to
experienced by the stream biota most of the time. relate broad anthropogenic (e.g., land-use/land cover)
The strong positive relationship between urban den- and nonanthropogenic features (basin area, slope, el-
sity and EC in the streams of eastern Melbourne is evation) to water quality (Johnson and others 1997,
consistent with the proposition that urbanization alters Brezonik and Stadelmann 2002). Stepwise methods of
both the rate of ionic flux through basins and the ionic model selection are generally regarded as flawed and
composition of the total dissolved solids loads in runoff prone to produce spurious models (Mac Nally 2000).
(Prowse 1987). Prowse (1987) attributed increased sa- Other methods that have been used to identify the
122 B. E. Hatt and others

causes of patterns include factor analysis (Wayland and utility, because connection in particular is an attribute
others 2003), principal components analysis, and clus- of urban land that is able to be manipulated through
ter analysis (Meador and Goldstein 2003). However, alternative drainage design.
these approaches are used to combine variables into
smaller subsets rather than to identify the individual Management Implications
features that are potentially causing degradation of The importance of drainage connection as an ex-
stream water quality. planatory factor for both concentrations and loads of a
range of pollutants points to the need for alternative
Loads approaches to stormwater management. The aim must
Loads of all water quality variables were strongly be to break the direct linkage between impervious areas
correlated with effective imperviousness and its constit- and the receiving waters. Opportunities for implement-
uent elements. The estimates of storm runoff volumes ing drainage changes exist both at-source and “end-of-
(derived from effective imperviousness) were highly pipe,” but distributed, at-source measures may provide
influential in determining estimated loads of pollut- the best opportunities for maintaining natural hydrol-
ants. As a result, the loads of variables such as NOx, for ogy (Wong and others 2000). Options include retrofit-
which concentrations were strongly explained by septic ting stormwater drainage systems in existing urban ar-
tank density, were more strongly explained by effective eas, and incorporating low-impact design (LID, also
imperviousness. called water-sensitive urban design: Lloyd and others
The most commonly used indicator of urban land- 2002) into new developments to prevent degradation of
use in the past has been total imperviousness (Center receiving waters.
for Watershed Protection 2003). However, the distinc- In our study, we were unable to find any streams that
tion between total and effective imperviousness has not drain subbasins with both imperviousness greater than
been explicit in many studies (Booth and Jackson 12% and low levels of drainage connection. However,
1997): typical values of effective imperviousness have because drainage connection was an important corre-
been used rather than direct measurement. In general, late independent of imperviousness, we postulate that
total imperviousness has been used because it is quicker effective water quality control will be possible by mini-
to measure and does not require extensive knowledge mizing drainage connection in basins with higher levels
of the drainage infrastructure (Center for Watershed of imperviousness, up to a point. The scope to discon-
Protection 2003). Our finding of drainage connection nect impervious areas will diminish as total impervious-
as the dominant independent correlate with concentra- ness increases. In highly impervious basins (e.g., dense
tions of several important pollutants and as a strong inner-city areas), maintaining impervious areas as dis-
correlate of loads of all pollutants suggests that the connected will be more difficult, although stormwater
direct determination of effective imperviousness will harvesting and recycling offer some potential.
greatly increase the predictive power of models of ur- Our findings suggest that reduction of drainage con-
ban effects on water quality. nection in urbanized basins should result in reduced
Approaches by other researchers to modeling the loads of all pollutants exported by the streams. How-
effects of urbanization on pollutant loads have ranged ever, for some variables, reduction of concentrations
from the application of a single export coefficient for might be of more importance to the stream ecosystem
each urban land-use type (Soranno and others 1996) to itself, and this may point to management priorities
empirical estimation of loads in several basins that were other than the reduction of drainage connection. For
then related to land use (Charbeneau and Barrett 1998, instance, we postulate that the reduction of nitrogen
Sokolov and Black 1999, Brezonik and Stadelmann loads in urbanized streams of our study will be most
2002). Although such models have proved useful in effectively achieved through reduction of drainage con-
gross estimates of loads from multiple-use basins (Sor- nection, but reduction of nitrogen concentrations will
anno and others 1996, Sokolov and Black 1999), corre- be most effectively achieved through replacement of
lations between land use indicators and loads or EMCs septic tank systems in the Dandenong Ranges.
have generally been found to be weak. Hence, these The spatial scale dependency of landscape attributes
models are of little use in identifying the most impor- has not been well resolved to date (Soranno and others
tant elements of urban land use causing increased pol- 1996). Although other studies have demonstrated the
lutant loads and concentrations (Charbeneau and Bar- importance of considering both the whole-basin con-
rett 1998, Brezonik and Stadelmann 2002). Models text and local site attributes, few have assessed the
using total imperviousness and drainage connection as relative importance of factors at basin and riparian
predictor variables have greater potential management scales. Those that have considered scale factors have
 Stream Pollutants and Urbanization 123

focused on land use (e.g., agricultural, urban) rather the D210 team for their assistance in the field and
than particular physical attributes (e.g., impervious cov- laboratory.
er), generally with inconclusive results. For example,
Johnson and others (1997) found that riparian charac-
teristics better explained some water quality variables Literature Cited
whereas the whole basin better explained others, and
AMHA-AWWA-WPCF. 1998. Standard methods for the exam-
this was season dependent. They also noted that, in
ination of water and wastewater. American Public Health
highly modified basins, riparian areas may simply re- Association, American Water Works Association, Water Pol-
flect dominant basin land uses. For urban streams, the lution Control Federation, Washington, D.C.
relevance of the condition of riparian zones is likely to Arnold, C. L. Jr., and C. J. Gibbons. 1996. Impervious surface
be reduced, because traditional stormwater drainage coverage: The emergence of a key environmental indicator.
pipes bypass these zones, reducing their influence on Journal of the American Planning Association 62:243–258.
in-stream water quality (Groffman and others 2002). Attiwill, P. M., P. J. Polglase, C. J. Weston, and M. A. Adams.
Adopting a basin scale approach is appropriate for 1996. Nutrient cycling in forest of south-eastern Australia,
Pages 191–227. in M. A. Adams, and P. M. Attiwill. eds,
urban streams because diffuse pollution by definition Nutrition of eucalypts. CSIRO, Melbourne.
stems from broad-scale features and therefore can most
Bannerman, R. T., D. W. Owens, R. B. Dodds, and N. J.
readily be controlled at this scale. Hornewer. 1993. Sources of pollutants in Wisconsin storm-
Our hypothesis that the widespread application of water. Water Science and Technology 28:241–259.
LID across basins will result in much-reduced pollutant Booth, D. B., and C. R. Jackson. 1997. Urbanization of aquatic
concentrations and loads could be tested by experi- systems— degradation thresholds, stormwater detention,
mental assessment of basin developments with and and the limits of mitigation. Journal of the American Water
Resources Association 33:1077–1090.
without LID, monitoring water quality before and after
development. An alternative, but related approach Brezonik, P. L., and T. H. Stadelmann. 2002. Analysis and
predictive models of stormwater runoff volumes, loads, and
would be the experimental assessment of the retrofit- pollutant concentrations from watersheds in the Twin Cities
ting of already developed basins without LID. Such metropolitan area, Minnesota, USA. Water Research
experiments will require a commitment to adaptive 36:1743–1757.
management on the part of urban land managers and Carpenter, S. R., N. F. Caraco, D. L. Correll, R. W. Howarth,
developers. A. N. Sharpley, and V. H. Smith. 1998. Nonpoint pollution
The explanatory models developed in this study of surface waters with phosphorus and nitrogen. Ecological
Applications 8:559 –568.
could form the basis of predictive models (sensu Mac
Nally 2002) of loads and concentrations using larger Center for Watershed Protection 2003. Impacts of impervious
cover on aquatic ecosystems. Center for Watershed Protec-
data sets from this and other regions. Such models tion, Ellicott City, Maryland 142.
could be refined by quantification of the influence of Charbeneau, R. J., and M. E. Barrett. 1998. Evaluation of
impervious surface type (i.e., is it more important that methods for estimating stormwater pollutant loads. Water
some surfaces, such as roads, be disconnected than Environment Research 70:1295–1302.
others, such as roofs), and by more intensive storm Chevan, A., and M. Sutherland. 1991. Hierarchical partition-
event sampling. The resulting quantitative predictive ing. The American Statistician 45:90 –96.
models of the effects of drainage connection and im- Chiew, F. H. S., and T. A. McMahon. 1999. Modelling runoff
perviousness on water quality will permit the develop- and diffuse pollution loads in urban areas. Water Science and
ment of better guidelines for urban design and for Technology 39:241–248.
prioritizing restoration efforts. Duncan, H. P. 1999. Urban stormwater quality: A statistical
overview. Cooperative Research Centre for Catchment Hy-
drology, Melbourne, Australia, 80 pp.
Gergel, S. E., M. G. Turner, J. R. Miller, J. M. Melack, and
Acknowledgments E. H. Stanley. 2002. Landscape indicators of human impacts
This study was part of CRCFE project D210, partly to riverine systems. Aquatic Sciences 64:118 –128.
funded by the Melbourne Water Corporation. The fol- Groffman, P. M., N. J. Boulware, W. C. Zipperer, R. V. Pouyat,
lowing people are thanked: Peter Newall, Geoff Taylor, L. E. Band, and M. F. Colosimo. 2002. Soil nitrogen cycle
processes in urban riparian zones. Environmental Science and
David Hatt, Barry Hart, Hugh Duncan, and Mike Grace
Technology 36:4547– 4552.
for their critical comments, Andrew Barton for his work
Hosomi, M., and R. Sudo. 1986. Simultaneous determination
on the flow models, Pua Tai Sim and Jae Yong Yoo for of total nitrogen and total phosphorus in freshwater sam-
the determination of subbasin variables, Carleen Mitch- ples using persulfate digestion. International Journal of Envi-
ell for her contribution to the laboratory analysis, and ronmental Studies 27:267–275.
 124 B. E. Hatt and others

Hydrologic Engineering Centre. 2002. HEC-RAS, version ative hydrological approach. Hydrological Processes
3.1.0.0. U.S. Army Corps of Engineers, Davis, CA. 15:1441–1457.
Johnson, L. B., C. Richards, G. E. Host, and J. W. Arthur. 1997. Schueler, T. 1994. The importance of imperviousness. Water-
Landscape influences on water chemistry in Midwestern shed Protection Techniques 1:100 –111.
stream ecosystems. Freshwater Biology 37:193–208. Sokolov, S., and K. P. Black. 1999. Long-term prediction of
Leopold, L. B. 1968. Hydrology for urban land planning: A water quality for three types of catchment. Marine and Fresh-
guidebook on the hydrological effects of urban land use. water Research 50:493–501.
Circular 554. U.S. Geological Survey, Washington D.C., 18 Soranno, P. A., S. L. Hubler, S. R. Carpenter, and R. C.
pp. Lathrop. 1996. Phosphorus loads to surface waters: A simple
Lloyd, S. D., T. H. F. Wong, and B. Porter. 2002. The planning model to account for spatial pattern of land use. Ecological
and construction of an urban stormwater management Applications 6:965– 878.
scheme. Water Science and Technology 45:1–10. U.S. Environmental Protection Agency. 2000. National Water
Mac Nally, R. 2000. Regression and model building in conser- Quality Inventory: 1998 Report to Congress. (http://www.
vation biology, biogeography and ecology: the distinction epa.gov/305b/98report)
between—and reconciliation of—⬘predictive’ and ⬘explan- Walsh, C. J. 2000. Urban impacts on the ecology of receiving
atory’ models. Biodiversity and Conservation 9:655– 671. waters: A framework for assessment, conservation and res-
Mac Nally. 2002. Multiple regression and inference in ecology toration. Hydrobiologia 431:107–114.
and conservation biology: further comments on identifying Walsh, C. J., and Mac Nally, R. 2003. The hier/part package
important predictor variables. Biodiversity and Conservation Version 0.1–1. Hierarchical Partitioning. (http://cran.
8:1397–1401. r-project.org/)
May, C. W., R. R. Horner, J. R. Karr, B. W. Mar, and E. B. Walsh, C. J.,. Papas, P. J., Crowther, D., Sim, P.T., and Yoo, J.
Welch. 1997. Effects of urbanization on small streams in the 2003. Stormwater drainage pipes as a threat to a stream-
Puget Sound Ecoregion. Watershed Protection Techniques dwelling amphipod of conservation significance, Austroga-
2:483– 494. mmarus australis, in south-eastern Australia. Biodiversity and
McMahon, G., and T. F. Cuffney. 2000. Quantifying urban Conservation 13: 781–793.
intensity in drainage basins for assessing ecological condi- Water Studies Centre. 2001. Analytical procedures manual.
tions. Journal of the American Water Resources Association Water Studies Centre, Monash University, Melbourne.
36:1247–1261. Wayland, K. G., D. T. Long, D. W. Hyndman, B. C. Pijanowski,
Meador, M. R., and R. M. Goldstein. 2003. Assessing water S. M. Woodhams, and S. K. Haack. 2003. Identifying rela-
quality at large geographic scales: Relations among land tionships between baseflow geochemistry and land use with
use, water physicochemistry, riparian condition, and fish synoptic sampling and R-mode factor analysis. Journal of
community structure. Environmental Management Environmental Quality 32:180 –190.
31:504 –517. Wong, T.H.F., Breen, P., and Lloyd, S. 2000. Water sensitive
Novotny, V., and H. Olem. 1994. Water quality. Prevention, road design—Design options for improving stormwater
identification, and management of diffuse pollution. Van quality of road runoff. Technical Report 00/1. Cooperative
Nostrand Reinhold, New York 1054. Research Centre for Catchment Hydrology, Melbourne, 74
pp.
Paul, M. J., and J. L. Meyer. 2001. Streams in the urban
landscape. Annual Review of Ecology and Systematics Wong, T. H. F., Fletcher, T.D., Duncan, H.P., Coleman, J.R.,
32:333–365. and Jenkins, G.A. 2002. A model for urban stormwater
improvement conceptualisation. in Global solutions for ur-
Prowse, C. W. 1987. The impact of urbanization on major ion ban drainage (CD Rom) Proceedings of the Ninth Interna-
flux through catchments: A case study in southern England. tional Conference on Urban Drainag, held in Portland,
Water, Air and Soil Pollution 32:277–292. Oregon, September 8 –13, 2002. (Eds E.W. Strecker and
Rose, S., and N. E. Peters. 2001. Effects of urbanization on W.C. Huber.) (American Society of Civil Engineers, Reston,
streamflow in the Atlanta area (Georgia, USA): a compar- Virginia, USA.)

View publication stats