Full text of DOI:10.1002/etc.5620210211

Environmental Toxicology and Chemistry, Vol. 21, No. 2, pp. 298–308, 2002
2002 SETAC
Printed in the USA
0730-7268/02 $9.00 .00
᭧
ϩ
A PESTICIDE SURFACE WATER MOBILITY INDEX AND ITS RELATIONSHIP WITH
CONCENTRATIONS IN AGRICULTURAL DRAINAGE WATERSHEDS
W
ENLIN
CHEN,* PETER HERTL, SUNMAO CHEN, and DENNIS TIERNEY
Health Assessment and Environmental Safety, Syngenta, 410 Swing Road, Greensboro, North Carolina 27409, USA
(
Received 1 December 2000; Accepted 10 August 2001)
Abstract—An index to benchmark pesticide mobility relevant to surface water runoff and soil erosion (surface water mobility
index, or SWMI) was derived based on two key environmental fate parameters: degradation half-life and organic carbon–normalized
soil/water sorption coefficient (
trend of 13 pesticides monitored in six Lake Erie, USA, tributaries from 1983 to 1991. Regression using a power function of SWMI
fits concentration data well at various percentiles in the database for each tributary and all six tributaries combined, with
2 ranging
from 0.71 to 0.94 for the concentrations at the 95th percentile. Good agreement was also obtained between SWMI and the time-
weighted annual mean concentrations (r2
0.67–0.87). Although concentrations at or near peaks tend to be driven by rare
hydrological events (intense precipitation immediately after application), SWMI explains the peak concentration data generally well
r2
0.53–0.86). The SWMI–concentration relationship was further evaluated with two other pesticide monitoring databases: the
Koc). Values assigned with the index of each individual compound correlate well with the concentration
r
ϭ
(
ϭ
U.S. Geological Survey National Water Quality Assessment Program White River Study Unit (1991–1996) at Hazelton, Indiana,
USA, and the Syngenta (previously Novartis) Voluntary Monitoring Program with Community Water Systems at the Higginsville
City Lake, Missouri, USA (1995–1997). The ability of the proposed SWMI to discriminate pesticide runoff mobility and its correlation
with surface water monitoring data can be significant in the development of screening methodologies and data-based models for
government agencies and/or practitioners in general facing increasing pressure to assess pesticide occurrence in aquatic environments.
Keywords—Pesticide
Mobility index
Surface water runoff
Soil erosion
Regression model
INTRODUCTION
practices in a watershed further complicates the scale problem
of accurately predicting water runoff, soil erosion, and ulti-
mately pesticide off-field transport into surface water. Field-
scale models thus often fail to accurately predict pesticide
concentrations in water bodies within a watershed, with pre-
dictions often orders of magnitude different from the moni-
toring data [8].
Assessment of surface water exposure to pesticides in wa-
tersheds usually requires estimation of chemical concentra-
tions over different time frames of concern (e.g., peak, annual,
or long-term averages). However, a dilemma often encountered
in these estimations is the lack of modeling tools that are data
based and can reliably predict either high-percentile or long-
term average pesticide concentrations in rivers and lakes/res-
ervoirs receiving runoff water from agricultural fields. While
physically based numerical models, such as the Pesticide Root
Zone Model (PRZM), have been used for pesticide regulatory
assessment [1], the validity of these mechanistic models has
been established only under small-scale and edge-of-field con-
ditions, where physical features, such as soil hydrological
properties and slope, are relatively homogeneous [1,2].
Several recent studies have shown that field-scale runoff
and erosion do not extrapolate well to larger, heterogeneous
watershed scales because of spatial variability in land use and
hydrological processes [3–7]. Merz and Plate [6] reported that
the spatial variability in soil hydraulic properties affects runoff
differently among different catchment sizes (0.32–3.52 km2).
In their study, deterministic modeling based on a mean pa-
rameter input did not result in a mean output for a heteroge-
Many well-controlled small plot studies have investigated
the edge-of-field processes of runoff and erosion [9]. However,
watershed-scale information has been generally lacking to war-
rant a detailed, accurate, and mechanistic quantification of pes-
ticide transport throughout regional drainage networks. Several
recent watershed monitoring studies have documented some
general patterns of pesticide occurrence in tributaries of Lake
Erie and the Mississippi River, USA [10,11]. These studies
generally indicated that the variation of riverine pesticide con-
centrations is dependent on the quantity of chemical used and
the size of the watershed. Using statistical regression, Battaglin
and Goolsby [12] were able to show that the chemical occur-
rence in U.S. midwestern rivers was in relation with a range
of watershed factors, including soil properties, chemical use,
stream flow, and basin topographic characteristics. Very few
studies, however, have shown a quantitative relation between
watershed monitoring data and individual pesticide physical/
chemical properties. Such empirical relations would allow a
better evaluation of the relative runoff and erosion potential
for different compounds and provide quick estimates of po-
tential concentration levels in watersheds. Comparison of the
empirical models based on watershed data to a more detailed
and field-scale process-oriented model would also provide in-
sight into the scale effect on pesticide transport behavior.
The objective of this paper was to evaluate the potential
neous field. Instead, the results could differ from
Ϫ50% to
180% compared to the homogeneous cases, indicating complex
propagation of input parameter variability to model outputs on
watershed scales. That is, average values of input parameters
do not necessarily produce model outputs that represent the
mean of the simulated system such as a watershed. The var-
iability in rainfall, topography, vegetative cover, and farming
* To whom correspondence may be addressed
(wenlin.chen@syngenta.com).
298

A pesticide surface water mobility index
Environ. Toxicol. Chem. 21, 2002
299
(2)
effect of two key environmental fate properties (soil half-life
and soil/water sorption coefficient) on pesticide occurrence in
surface water and to assess if a correlation could be established
to allow predictions of surface water concentrations of dif-
ferent pesticides in agricultural drainage networks. Three
working steps were involved. First, based on a standard ex-
posure scenario and the basic equations of field runoff and
erosion, the two environmental fate parameters were integrated
into the surface water mobility index (SWMI) to benchmark
individual pesticide mobility potential specifically relevant to
surface water runoff and soil erosion. Second, historical sur-
face water monitoring data sets (two from river systems and
one with a reservoir) were selected and analyzed for time-
weighted percentile concentrations. The two river data sets
included the monitoring program for six tributaries of Lake
Erie as reported by Richards and Baker [10] and the U.S.
Geological Survey NAWQA White River Study Unit (1991–
1996) at Hazelton (http://www-dinind.er.usgs.gov/nawqa/)
[13]. The reservoir data set was from the Syngenta (previously
Novartis) voluntary monitoring program (1995–1997) con-
ducted on the Higginsville City Lake [14]. The third step in-
volves regression and model calibration using the Lake Erie
Basin monitoring data together with the corresponding pesti-
cide usage and watershed sizes to establish a relationship with
SWMI. The established relationship was then evaluated as a
screening tool to predict surface water pesticide concentrations
and compare them with the monitored concentrations noted in
the other two data sets.
Ca
⑀
Cw
Cs
ϭ
ϭ
and
1
ϩ
Kd
⑀
⑀
Ca Kd
⑀
(3)
1
ϩ
Kd
where
⑀ is mass of soil on which pesticide can be extracted
by runoff water, or runoff extraction coefficient (kg/L); Ca is
total pesticide concentration available for transport due to run-
off and soil erosion (mg/kg); and Kd is soil or sediment/water
sorption coefficient (or sorption coefficient; L/kg). The Kd is
often normalized by soil organic carbon content (OC, %), that
is, Koc
malized soil/water sorption coefficient with a unit same as Kd
The Ca is further calculated by the first-order degradation
equation:
ϭ Kd/OC. Thus, Koc refers to an organic carbon–nor-
.
Ϫ0.693t/Th
Ca
ϭ
Ca0e
(4)
where Ca0 is the initial total concentration (adsorbed and dis-
solved) at day 0 after application (mg/kg), Th is pesticide soil
half-life (d), and
t is time when first runoff event occurs.
Substituting Equations 2 to 4 into Equation 1 gives
Ϫ
0.693t/Th)
⑀
Ca0 e(
M
ϭ
(105Qd Aha
ϩ
rom Er Kd
)
(5)
(1 ϩ ⑀Kd
)
where Qd is runoff water depth (cm). All other variables are
the same as defined previously.
The runoff water depth can be calculated by the U.S. De-
partment of Agriculture–Soil Conservation Service curve
number method [19]:
THEORY
2
[
R
Ϫ
0.2
0.8
is precipitation (cm) and CN is the curve number,
S
]
100
Qd
ϭ
and
S
ϭ
25.4
Ϫ
1
(6)
΂ ΃
R
ϩ
S
CN
Previous quantification of pesticide mobility based on
chemical environmental fate properties has focused primarily
on the pesticide propensity relevant to leaching in soil [15–
17]. Applicability of these approaches to pesticide transport
by surface water runoff and soil erosion is limited because of
the different transport mechanisms involved in surface water
hydrology compared to leaching in the subsurface environ-
ment. In the remainder of this paper, we focus on addressing
pesticide surface runoff and erosion potential.
where
R
which can be estimated on the basis of the method of ante-
cedent moisture condition II [20].
Soil erosion can be estimated by the modified universal soil
loss equation [1]:
0.56
Er
ϭ
11.8(qpQv
)
K LS C P
(7)
where qp is peak runoff rate (m3/s),
efficient, LS is length–slope factor,
is management practice factor. The peak runoff rate is further
approximated on the basis of a simple trapezoidal hydrograph
[1]:
K
is soil erodability co-
For a single rain event, pesticide loss from an agricultural
field, dissolved in runoff water and bound to soil particles, can
be expressed as
C is crop cover factor, and
P
M
ϭ
ϭ
Mw
ϩ
Ms
ϭ
1000·Qv
Aha
·Cw ϩ rom ·Er ·Cs and
0.0278Aha RQd
qp
ϭ
(8)
ln(rom
)
2
Ϫ
0.2 ln(Er
/
)
(1)
1000
TR R
Ϫ
0.508
Ϫ 10
΂
΃
]
[
CN
where
M
is total pesticide mass loss in both runoff water and
soil particles (mg), Mw is total pesticide mass in runoff water
(mg/L), Ms is total pesticide mass bound to soil particles in
erosion (mg/kg), Qv is total runoff water (m3), Cw is concen-
tration in runoff water (mg/L), Aha is area of the runoff field
where TR is peak storm duration (h).
The derivation in Equations 5 to 8 shows that pesticide off-
field transport to surface water is not only a function of pes-
ticide environmental fate properties but also dependent on the
soil and hydrological conditions. To quantify pesticide mo-
bility only by its environmental fate properties, a uniform ex-
posure environment under identical runoff conditions must be
specified to obtain a standardized basis for all pesticides. For
this purpose, a Loring silt loam in Yazoo County, Mississippi,
USA, was selected. The selection was based primarily on the
readily available hydrological properties that have been used
frequently by the U.S. Environmental Protection Agency as a
standard surface water assessment scenario [21]. Assuming
conventional till and cotton growth, this soil is highly vul-
(ha),
soil particles (mg/kg), 1,000 is a unit conversion factor (from
m3 to L), and
om is an empirical soil organic matter enrichment
Er is total soil erosion (kg), Cs is concentration in eroded
r
ratio based on Mullins et al. [1]. The enrichment ratio is used
to account for the particle-selective effect during erosion and
is an adjustment factor in eroded soil relative to field soil.
Assuming instantaneous sorption equilibrium and that the
concentration of the chemical sorbed by the soil solid is di-
rectly proportional to the concentration in liquid phase, Cw and
Cs can be calculated as [2,18]

300
Environ. Toxicol. Chem. 21, 2002
W. Chen et al.
Table 1. Soil and field hydrological parameters of a Loring silt loam
soil, Yazoo (MS, USA), for the standard runoff scenario to define the
pesticide surface water mobility index (SWMI)
Soil erodability coefficient
0.49
K
ϭ
Bulk density ϭ 1.6 kg/L
Length–slope factor LS
Management practice factor
0.75
ϭ
0.4
p
Organic carbon
Incorporation depth ϭ
ϭ
1.16%
3 cm
ϭ
Field area Aha
Peak storm duration TR
ϭ
10 ha
Field capacity (
5.8 h Wilting point (␪ )
␪
)
ϭ
ϭ
0.294
0.094
f
ϭ
p
Standard application rate
kg/ha
ϭ
1.12 Initial soil water content (␪ ) ϭ
in
␪ ϭ 0.294
f
Curve number (CN) ϭ
97a
a Based on the method of antecedent moisture condition II (Haith et
al. [20]).
nerable to surface runoff as represented by its high CN value
(Table 1).
Further parameterization includes the runoff extraction co-
efficient ⑀, which has been reported as Koc dependent and with-
in the range from 0.1 to 0.5 [18]. To retain simplicity, a medium
value ⑀ ϭ 0.3 was used. Additionally, a 5.08-cm precipitation
event was assumed to occur 5 d after application. It should be
noted that the hypothetical scenario characterized by somewhat
arbitrary selection of environmental parameters may not re-
alistically exist. However, the objective here was to benchmark
pesticide runoff potential (mobility) solely by the compound
environmental fate properties.
Fig. 1. Contour plot of the pesticide surface water mobility index as
a function of Koc and Th. EPTC
ϭ s-ethyl dipropylthiocarbamate.
erally within the SWMI range of 0.1 to 0.7, with the contour
line at about 0.3 as a divider between insecticides and her-
bicides.
Substituting all the specified parameters into Equations 5
to 8 and redefining the mass loss as the fraction of total pes-
Our further postulation is that SWMI, as defined in Equa-
tion 10, is correlated to pesticide concentrations observed in
surface water resulting from agricultural field water runoff and
soil erosion. With the pesticide concentrations being directly
proportional to its use intensity in drainage basins as reported
in the literature [10–12], we assume a power function as a
plausible relationship to describe the correlation between
SWMI and pesticide concentrations:
ticide used (
Fm
h and
Fm) yields
Ϫ
3.466/Th
0.3
0.00348Koc
Koc that can only be positively valued, Equation
e
ϭ
(0.9
ϩ
0.000234Koc
)
(9)
(1
ϩ
)
Given
T
9 is mathematically bounded within 0 and 0.27. That is, Fm
asymptotically approaches 0 as Th approaches 0 and Koc ap-
proaches infinity, whereas Fm approaches 0.27 as Th approaches
infinity and Koc approaches 0. For convenience, Equation 9 is
rescaled to a mathematical bound between 0 and 1, and we
define
Muse
␤
CP(x)
ϭ
␣
·
SWMI or
(11)
Ab
ϭ ␤·Log(SWMI
where CP(x) is concentration at percentile
Log(CP
(x)/Rb
)
) ϩ Log(␣)
(12)
x
in a surface water
Ϫ3.466/Th
e
monitoring database ( g/L), Muse is total annual use of a pes-
␮
SWMI
ϭ
(1
ϩ
0.00026Koc
)
(10)
(1
ϩ 0.00348Koc)
ticide in a basin or watershed, Ab is total area of a drainage
basin or watershed (ha), Rb is basin application rate (kg/ha),
By definition, therefore, SWMI determines monotonically the
relative pesticide mobility scale, with value of 0 being the
asymptotically least mobile and 1 the asymptotically most mo-
bile to be transported by the processes of surface water runoff
and soil erosion.
A contour line plot of SWMI as a function of Koc and Th
is illustrated in Figure 1, where the contour lines represent
equal values of SWMI. Three regions may be differentiated
in the figure. For compounds with a relatively short half-life
and
␣ and ␤ are regression constants. The power function
model would predict a 0 concentration as SWMI approaches
0 (the most immobile compound).
The basin application rate
Rb in a given drainage watershed
characterized by the total area, Ab, may be obtained from the
total area treated (Atreated
Rlabel kg/ha):
,
ha) and the label application rate
(
,
Rlabel Atreated
Rb
ϭ
(13)
(
Ͻ
10 d), the mobility index is sensitive predominantly to half-
Ab
life variations rather than to changes in oc. Because of their
K
The ratio Atreated /Ab in Equation 13 is the fraction of treated
short-lived nature, it is often the temporal and spatial coin-
cidence of pesticide application and the occurrence of the run-
off event that governs the mobility for this type of compounds.
For compounds with a half-life between 10 and 40 d, the
mobility index is sensitive to both parameters (region II in
Fig. 1). For more persistent compounds, the index is deter-
cropland in the overall basin area. Denoting the ratio as
substituting Equation 13 back to Equation 11 gives
f
c and
(14)
␤
CP(x)
ϭ Rlabel fc ␣·SWMI
Equation 14 assumes that the treated area received the label
recommended use rate. The equation can be used when no
basin use rate is available for direct use of Equation 11. Hence,
mined primarily by
Koc (region III in Fig. 1). The 13 pesticides
monitored in the Lake Erie Basin as described later fall gen-

A pesticide surface water mobility index
Environ. Toxicol. Chem. 21, 2002
301
Table 2. Pesticide properties, values of the surface water mobility
index (SWMI), and total annual use in Ohio Agricultural Statistics
Service (OASS), USA, crop reporting Districts 1 and 2 in the Lake
Erie Basin, USA, 1986 [22]
Total use
(kg)
Pesticide
Type
Koc (L/kg)a Th (d)b SWMI
Alachlor
Metolachlor
Atrazine
Cyanazine
Metribuzin
Linuron
Terbufos
Butylate
Chlorpyrifos
EPTCe
Herbicide
Herbicide
Herbicide
Herbicide
Herbicide
Herbicide
Insecticide
Herbicide
Insecticide
Herbicide
Insecticide
Insecticide
Herbicide
170
200
100
15 0.5207 1,058,896
90 0.5969
60 0.7184
14 0.4932
40c 0.5453
60 0.4523
7d 0.1388
13 0.3536
30 0.1038
665,417
581,750
206,463
203,405
110,931
59,265
50,558
24,565
30,757
25,358
19,432
14,132
190
224c
370
1,504d
400
6,070
200
6
0.3482
Phorate
Fonofos
Simazine
540d
870
3d 0.1248
40 0.2792
60 0.6718
130
a Koc
b Th
ϭ
ϭ
organic carbon–normalized soil/water sorption coefficient.
degradation half-life in soil. Data from Richards and Baker
Fig. 2. Lake Erie tributary drainage basins and locations of pesticide
sample stations in the Ohio, USA, agricultural runoff monitoring pro-
gram. Listed along with the sampling stations are the U.S. Geological
Survey (USGS) stream gauging station numbers (courtesy of Peter
Richards).
[10] or as otherwise noted.
c [28]. Average Koc of 81, 65, and 526 calculated from multiple Kd
values: 1.32 (2.8% organic matter), 1.90 (5.0% organic matter), 1.53
(0.5% organic matter).
d Data from American Cyanamid (Wayne, NJ, USA).
e EPTC
ϭ s-ethyldipropylthiocarbamate.
either Equation 11, 12, or 14 is referred to as the SWMI model
in this paper.
scription of the monitoring program can be found in Richards
and Baker [10].
MONITORING DATA SETS AND ANALYSIS
Four input parameters are required to calibrate the SWMI
model. One parameter is the pesticide annual use data. Re-
gional annual use data for the 13 pesticides monitored by
Richards and Baker for 1986 in northwestern Ohio, USA, were
obtained from Waldron (Table 2) [22]. The other three SMWI
parameter inputs were the size of each individual watershed,
Lake Erie Basin data
The primary data set selected to establish the relationship
of SWMI with measured pesticide concentrations in surface
water was collected by Richards and Baker for six tributaries
of the Lake Erie Basin (Maumee River, Sandusky River, Honey
Creek, Rock Creek, River Raisin, and Lost Creek, USA) [10].
The monitoring locations and each of the tributary drainage
areas are illustrated in Figure 2. This monitoring program pro-
vided a complete database of pesticide riverine concentrations
monitored for nine consecutive years (April 1983–December
1991). More than 4,000 samples were analyzed for 13 different
pesticides (Table 2). The sampling frequency during the high-
runoff period (April 15–August 15) was usually three samples
per day and less frequently the rest of the calendar year (two
samples/month). Agriculture was the predominant land use
(67–83% row crop) in the six watersheds.
The Lake Erie Basin pesticide data set was selected to
calibrate the SWMI model (Eqn. 12) for three reasons. First,
the annual sample frequency was intentionally designed to
collect more samples during storm-related river flow than dur-
ing dry weather base flow. This time-weighted design opti-
mized the likelihood of capturing transient pesticide individual
peak concentrations subsequent to the springtime pesticide ap-
plication period. Second, the number of pesticides monitored
provided a wide range of soil degradation half-lives, sorption
coefficients, and agricultural crop use patterns (Table 2). Third,
the study duration was nine years (1983–1991). This extended
time period of continuous monitoring helps provide a range
of annual weather patterns that capture normal as well as
above- and below-normal annual and seasonal precipitation.
Thus, the annual and multiyear mean concentrations for each
pesticide should be representative of long-term exposure,
while seasonal short-term maximum exposure would be as-
sociated with the spring-period runoff events. A detailed de-
the individual pesticide’s soil degradation half-life (
organic carbon–normalized soil/water sorption coefficient
oc). These properties were obtained either from Richards and
Th), and
(
K
Baker [10] or from other sources as noted in Table 2.
Because crop-specific pesticide use information was not
available for each of the six agricultural watersheds and the
regional use data were available only for 1986, a uniform use
rate for each pesticide was applied over each of the six ag-
ricultural watersheds for the nine-year period. For each pes-
ticide, the basin use rate Rb was calculated on the basis of the
1986 annual use surveyed in the Ohio Agricultural Statistics
Service crop reporting districts 1 and 2, which encompass five
of the six Lake Erie tributaries in Ohio (except River Raisin
in Michigan, USA) (Fig. 2) [22]. The total county land area
is 2,341,468 ha in the two districts. The calculated average
basin use rate was applied to River Raisin.
NAWQA Program
Monitoring data of 12 pesticides (Table 3) were obtained
from the U.S. Geological Survey NAWQA Program for the
sampling site on White River at Hazelton. These data were
used to evaluate the SWMI model, which had been calibrated
on the basis of the monitoring data from the six river systems
in the Lake Erie Basin. The data set was downloadable from
the U.S. Geological Survey Internet site (http://www.dinind.
er.usgs.gov/nawqa/). The Hazelton location is approximately
19 miles upstream of the White River’s confluence with the
Ohio River. This location on the White River receives flow
from essentially the entire watershed drainage area (2,938,382

302
Environ. Toxicol. Chem. 21, 2002
W. Chen et al.
Table 3. Pesticide annual use and concentration data from U.S. Geological Survey, National Water Quality Assessment Program White River
Study Unit, at Hazelton (IN, USA) (May 1, 1991, to September 23, 1996)a
Basin annual
use (kg)
Basin annual use
rate (kg/ha/year)
TWMC
(ppb)b
95th
Peak
Pesticide
(ppb)c
(ppb)d
Alachlor
Metolachlor
Atrazine
Cyanazine
Metribuzin
Linuron
Terbufos
Butylate
Chlorpyrifos
EPTCf
562,500
931,500
999,000
355,950
33,300
35,550
38,250
399,150
69,300
37,350
73,350
15,750
0.1930
0.3195
0.3427
0.1221
0.0114
0.0122
0.0131
0.1369
0.0238
0.0128
0.0252
0.0054
0.1052
0.5164
1.3476
0.3056
0.0219
0.5000
2.2800
6.3800
1.3800
0.0990
3.2000
5.3000
13.4000
5.1000
0.3800
0.1230
0.0500
0.0830
0.1300
0.0270
0.3200
1.2000
Ͻ
Ͻ
0.05e
Ͻ
Ͻ
0.05e
0.05e
0.05e
0.0055
0.056e
0.008e
0.005e
0.1333
0.0240
0.056e
0.008e
0.005e
0.4700
Ͻ
Ͻ
Ͻ
Ͻ
Ͻ
Ͻ
Fonofos
Simazine
a For some compounds in the table, half the limit of detection (LOD) is applied when the measured values
specifically.
Ͻ
LOD; otherwise, it is noted
b TWMC
ϭ annual time-weighted mean concentration.
c 95th
d Peak
ϭ
ϭ
concentration at which 95% of time the value is below or equal to.
maximum concentration value.
e Less than LOD.
f EPTC
ϭ s-ethyldipropylthiocarbamate.
ha) [13]. The monitoring period was about six years (May
1991–September 1996). The annual sample frequencies were
twice weekly to twice monthly from May to August and ap-
proximately monthly during other times of the year [13]. Pes-
ticide analysis was by gas chromatography/mass spectrometry.
More detailed descriptions of the monitoring program are in
Crawford [13] and Shelton [23]. Pesticide use data (annual
average 1992–1994) for the White River Basin were obtained
from Anderson and Gianessi [24].
the limit of detection (LOD) 0.1 ppb were reported as 0.05
ppb (half the LOD) in the calculations of annual means.
Data processing
The three monitoring programs were designed to sample
rivers and reservoirs more frequently in high-runoff periods
subsequent to the field application of the pesticides (usually
late spring to early summer), with less frequent samples col-
lected the rest of the year. The unequal sampling interval was
selected to more accurately monitor the seasonal variation in
pesticide concentrations prior to, during, and subsequent to
the pesticide application period in the Great Lakes, upper Mis-
sissippi, and Ohio River systems. The uneven sample fre-
quency within a calendar year requires the measured pesticide
concentrations to be organized in a time-weighted fashion.
Assuming that each sampling point represents half the time to
the preceding sample and half to the following sample as de-
scribed in Richards and Baker [10], a time weight for each
sample was calculated as
Higginsville Community Water Systems monitoring data
Concentrations of the herbicide atrazine at the Higginsville
City Lake were monitored through the Syngenta (previously
Novartis) Voluntary Monitoring Program with Community
Water Systems (CWS) during 1995–1997 [14]. The drainage
area of the city lake is about 1,386 ha and had annual atrazine
use in the range 107 to 116 kg during the three monitoring
years [25] (Table 4). Water samples were taken both at the
point of distribution (finished drinking water) and at the intake
point (raw water). Two sampling frequencies were established
for a 12-month period (January–December) to cover potential
seasonal variations in concentrations, with weekly samples
from May to July and two samples per month the other nine
months. Sample analysis included both gas chromatography
n
wi
ϭ
[(tiϩ
Ϫ
tiϪ1)/2]
t
iϭ1
(15)
͸
1
i
΋
where
t
i is the sampling time point of sample
i
and
n
is number
and immunoassay with limit of detection at 0.1 ppb (␮g/L).
of samples during the total sampling period. The weights were
then organized by their associated corresponding concentra-
tions in an ascending order. The cumulative time-weighted
The parent atrazine in the immunoassay analysis was adjusted
from a relationship between the gas chromatography–detected
atrazine and the immunoassay results. Concentrations below
Table 4. Atrazine annual use and concentrations monitored at the Higginsville City Lake (MO, USA), 1995–1997
Total annual use
Measured water (ppb), finished
Measured water (ppb), raw
in drainage basin Basin application
Year
(kg)
rate (kg a.i./ha)
TWMCa
95thb
Peakc
TWMC
95th
Peak
1995
1996
1997
113.5
106.8
116
0.0819
0.0771
0.0837
0.72
0.83
1.72
1.38
1.34
3.14
1.64
1.46
3.93
2.79
1.88
2.95
4.27
2.99
4.72
4.71
2.99
5.50
a TWMC
ϭ annual time-weighted mean concentration.
concentration at which 95% of time the value is below or equal to.
maximum concentration value.
b 95th
c Peak
ϭ
ϭ

A pesticide surface water mobility index
Environ. Toxicol. Chem. 21, 2002
303
Table 5. The 95th-percentile concentrations (
␮
g/L) of 13 pesticides monitored in the six Lake Erie Basin tributaries, USA, 1986
Pesticides
Maumee River
Sandusky River
Honey Creek
Rock Creek
River Raisin
Lost Creek
Alachlor
Metolachlor
Atrazine
Cyanazine
Metribuzin
Linuron
4.0890
4.3858
7.9799
2.7404
2.3862
0.2369
0.0073
0
8.9863
10.9704
15.0301
4.5416
2.4816
1.1338
0.0180
0.1503
NA
0.0508
0.0100
0.0397
0.5532
75
10.8784
11.8958
15.0658
2.1908
2.1109
2.0831
0.0113
0.0503
6.1818
10.5550
12.2378
1.8529
2.5859
1.7921
0.0070
0.0300
3.5783
6.1903
15.2498
1.5562
1.2928
0.4499
0.0060
0
3.9996
1.6824
6.0666
1.1928
1.7664
0.4816
0.0288
0.1808
Terbufos
Butylate
Chlorpyrifos
NAa
NA
NA
NA
NA
EPTCb
0
0.0412
0.0030
0.0303
0.4926
125
Jan. 13–Sept. 29
0.0000
0.0040
0.0279
0.2299
124
Jan. 13–Sept. 29
0.0000
0.0048
0.0093
0.1802
17
0.0998
0.0024
0.0174
0.9204
135
Jan. 13–Sept. 22
Phorate
Fonofos
Simazine
No. of samples
Sampling period
0.0080
0.0335
0.7068
75
Jan. 13–Sept. 15
Jan. 13–Sept. 8
Apr. 6–Sept. 27
a NA
b EPTC
ϭ
no data available.
-ethyl dipropylthiocarbamate.
ϭ
s
percentile associated with each concentration thus was deter-
mined as
other compounds with smaller uses were not reported in the
publication of Richards and Baker [10] because the small
TWMC values calculated using uncensored analytical results
were far below their corresponding LODs.
n
Pi
ϭ
100
w
͸
iϭ1
(16)
i
Regression results of the proposed SWMI relationship
(Eqn. 12) to the concentrations normalized for each pesticide
basin scale use rate, including the 95th percentile, peak, and
TWMC data, are shown in Figure 3A to C and Table 6 (Yr
83–91 models). Data for all the six tributaries were plotted
and used in the regression. Thus, the spread of data points at
a given SWMI value represents the variations among the six
rivers. A few zero values in the original uncensored data were
excluded because of uncertainty in the true value. As shown
in Figure 3A and Table 6, a significant relationship exists
between the proposed SWMI and the 95th-percentile concen-
trations obtained from the six agricultural drainage basins
where Pi corresponds to the time-weighted percentile at con-
centration Ci of sample i. An exact percentile, for example,
the 95th, was estimated by linear interpolation between two
adjacent samples. A time-weighted percentile, therefore, rep-
resents the percentage of the total time during the observation
period over which a particular concentration is not exceeded.
For example, a 95th percentile of 1 ppb indicates that a con-
centration
period.
Ͻ1 ppb occurred during 95% of the observation
Time-weighted mean concentration (TWMC) was calcu-
lated as
throughout the entire monitoring period (1983–1991) (r2
ϭ
n
0.71). In comparison, the TWMC and the peak data are rather
scattered with much variability exhibited among different trib-
utary drainage basins (more discussions follow).
TWMC
ϭ
w C
(17)
͸
i
i
i
ϭ
1
where Ci is the concentration of sample i.
The regressions of multiyear monitoring data (1983–1991)
were based on the single-year survey of pesticide use infor-
mation conducted in 1986 by Waldron [22]. To further explore
the SWMI relationship when more accurate pesticide use in-
formation is available, the original database of the Lake Erie
Basin was reanalyzed specifically for 1986 (personal com-
munication, P. Richards, Heidelberg College, OH, USA). The
sampling period in 1986 covered generally from January to
September across the six Lake Erie tributaries (Table 5). Data
of insecticide chlorpyrifos data were not available during that
monitoring year.
The 95th-percentile concentrations for the 13 pesticides
over the nine-year period (1983–1991) in the six flowing water
bodies in the Lake Erie Basin can be found in Richards and
Baker [10], while the 95th percentile for the same rivers in
1986 are shown in Table 5. The atrazine time-weighted means
and the 95th-percentile and peak individual concentrations for
the CWS at the Higginsville City Lake during 1995–1997 are
illustrated in Table 4. The monitoring data of time-weighted
means and the 95th-percentile and peak individual concentra-
tions for 12 pesticides from the U.S. Geological Survey–
NAWQA site at Hazelton are shown in Table 3.
Regressions of the 1986 single-year monitoring and product
use data were conducted for each of the six tributaries indi-
vidually as well as all rivers combined (Table 6). Results of
the combined regression are illustrated in Figure 4A to C (Yr
86 models in Table 6). An example for one of the six individual
river regressions (Honey Creek) is shown in Figure 5A to C.
Compared to the results based on the multiyear data, all the
new regressions were improved significantly (Table 6). With
the data for all rivers combined, the r2 value of the 95th-
RESULTS AND DISCUSSION
Regression
The first regression analysis was conducted on the entire
database obtained over the period of 1983 to 1991 from the
six Lake Erie Basin tributaries. The time-weighted 95th-per-
centile concentrations of 13 pesticides were obtained directly
from Richards and Baker [10]. The time-weighted mean and
peak concentrations were also available from the same source.
The reported TWMC values, however, were limited only to
the six herbicides of highest usage in the region (Table 2). The
percentile regression increased from 0.71 (
to 0.87 ( 67, Fig. 4A), while the TWMC regression in-
creased from 0.33 ( 36, Fig. 3B) to 0.67 ( 72, Fig.
n ϭ 64) (Fig. 3A)
n
ϭ
n
ϭ
n ϭ

304
Environ. Toxicol. Chem. 21, 2002
W. Chen et al.
best with the 95th-percentile concentrations, particularly when
data for all rivers were combined (Table 6). Percentiles either
above or below the 95th, including the TWMC and the peak
data, appear more scattered, with a few cases of weak corre-
lation, particularly to the peak concentrations (Table 6). The
variable ability of SWMI to explain different percentile data
can be attributed to a number of reasons. First, the overall
wide range of peak concentrations among different pesticides
and watersheds (0.09–97 ppb in the multiyear database and
0.004–96 ppb in the single year, 1986), on one hand, appears
to provide a sensitive basis responding to the variations of the
compound properties (SWMI). The occurrence of individual
peak concentrations, on the other hand, is strongly dependent
on the nature of the rainfall-generated runoff events (the mag-
nitude as well as the temporal and spatial concurrence with
pesticide applications). It can be seen from Figures 3B and 4B
that the peak variations among different compounds are about
the same in significance as the relative variations of a single
compound within the six different watersheds. This would
weaken the response of the compound properties to the overall
variation of the peak data across different pesticides. The typ-
ically heavy soil with high clay content and low infiltration in
the Lake Erie drainage basin is of particular relevance to the
production of rapid and strong overland flow that directly
washes into the tributaries [10]. Peak concentrations usually
occur at the time of a pesticide’s maximum availability to
runoff, often shortly after application. This means that the
dependency of peak concentrations on individual chemical
properties is reduced regardless of a soil short or long half-
life compound even though the trend of correlation (increases
as SWMI increases) still remains valid as demonstrated by the
overall improved agreement with the 1986 as well as the in-
dividual river data.
Second, similar to the peak concentrations, SWMI ex-
plained the variation of TMWC data relatively well except for
the 1983–1991 data of the six herbicides (Table 6). Compared
to the 95th-percentile data, the relatively low r2 values (even
though not statistically different) were probably partly due to
the fact that the overall variation in the observed TWMC val-
ues were low across the pesticides monitored (range 0.05–2.33
ppb during 1981–1991 and 0.0004–2.99 ppb in 1986). It is
expected that the precision of the lower TWMC values could
also be adversely affected by the lower precision of the un-
censored analytical results near the method LODs. This effect
can propagate to the TWMC calculations since the low and
relatively less accurate concentrations often have higher time
weights representing a wider time period because of the sparser
off-season sampling events.
Fig. 3. Regressions between the proposed pesticide surface water
mobility index (SWMI) and the basin use–normalized concentrations.
Data are derived from the multiyear database of the six agricultural
tributaries in the Lake Erie Basin, USA, sampled from April 1983 to
December 1991. TWMC
ϭ time-weighted mean concentration.
Comparison of the two regression constants
and in Table 6 reveals that Log( ) increases as the concen-
tration data used for regression changed from TWMC to peak,
whereas the general trend for constant is decreasing. Con-
stant in Equation 11, or Log( ) in Equation 12, is a scaling
parameter that determines the elevation of the model predic-
tions. That is, it adds a linear multiplicity to the overall
regression. Constant is the exponent of SWMI that defines
the shape of the regression model. Generally, as decreases,
the regression model in the form of a power function (Eqn.
11) becomes less concave and thus less sensitive to SWMI.
To illustrate variations among different-size watersheds, the
six individual tributary models for the peak and TWMC were
plotted in Figure 6A and B. As compared in Figure 6A, the
peak models of the three relatively smaller tributaries (Lost,
␣ or Log(␣)
␤
␣
4B), and the regression of peak data increased from 0.24 (
78, Fig. 3C) to 0.53 ( 72, Fig. 4C). All results were
significant at the 95% level of statistics [26]. Individual river
n
ϭ
n ϭ
␤
Z
␣
␣
regressions were generally better than all rivers combined be-
cause of reduced variability among different tributary water-
sheds. The overall improvement of the SWMI relationship as
more relevant pesticide use information was incorporated thus
supports the concept of SWMI as a generic chemical mobility
index. It also underlines the importance of obtaining accurate
annual pesticide use data in order to allow quantitative analysis
of regional scale monitoring programs.
␣
␤
␤
Although a general improvement in the SWMI regression
on the monitoring data was achieved with more accurate annual
pesticide use information, it remains true that SWMI correlates

A pesticide surface water mobility index
Environ. Toxicol. Chem. 21, 2002
305
Table 6. Regression results of the multiyear (1981–1991) and single-year (1986) data for the six Lake Erie, USA, agricultural tributary drainage
basins
TWMC (
␮
g/L)a
95th percentile (
␮
g/L)b
Peak (
␮
g/L)c
River name
(regression group)
Drainage
area (ha) Log(
␣
)
␤
r2
nd
Log(
␣
)
␤
r2
nd
Log(
␣
)
␤
r2
nd
All river (1983–1991 models)
All rivers (1986 models)
Lost Creek
Rock Creek
Honey Creek
River Raisin
Sandusky River
Maumee River
2,233,400 1.128 2.720 0.33
2,233,400 1.253 2.854 0.67
1,130 1.057 3.049 0.72
8,800 1.411 3.359 0.87
38,600 1.577 3.478 0.86
269,900 1.063 2.490 0.76
324,000 1.224 1.712 0.73
1,639,500 1.186 3.037 0.58
36e
72
12
12
12
12
12
12
1.725 2.408 0.71
2.093 2.825 0.87
1.963 2.865 0.94
2.165 3.049 0.90
2.301 3.207 0.92
1.896 2.507 0.79
2.205 2.683 0.93
2.012 2.661 0.86
64
67
10
11
12
12
12
10
2.522 1.157 0.24
2.492 2.267 0.53
2.680 1.960 0.76
2.782 2.950 0.85
2.952 2.983 0.86
2.097 2.486 0.83
2.346 0.799 0.21
2.095 2.425 0.74
78
72
12
12
12
12
12
12
a TWMC
b 95th percentile
c Peak
maximum concentration value.
ϭ
annual time-weighted mean concentration.
concentration at which 95% of time the value is below or equal to.
ϭ
ϭ
d Number of data points. The different number of data points is due to no samples or values excluded when below limit of detection.
e Only six herbicides were included for the TWMC 1983 to 1991 regression.
Rock, and Honey) deviate clearly from the larger rivers, par-
ticularly River Raisin and Maumee River, projecting much
higher peak concentrations. The noticeably flatter predictions
by the Sandusky peak model were probably due to its low r2
The TWMC and 95th-percentile and peak concentrations
of the White River Basin monitoring data obtained from Ha-
zelton, Indiana, are compared to the SWMI model predictions
in Figure 7A to C. Since no calculable TWMC and 95th-
(
ϭ
0.21, Table 6). Compared to the peak models, the TWMC
percentile data were available for chlorpyrifos, s-ethyl dipro-
models did not exhibit a similar trend of consistent high pre-
dictions for the small rivers than for the large ones (Fig. 6B).
Predictions of TWMCs were relatively close among different-
size rivers. For example, Lost Creek had TWMC predictions
only slightly lower than Maumee River even though it is a
much smaller tributary of the latter.
Larger watersheds tend to have lower observed peak con-
centrations while longer duration of moderate concentrations
because of mixing of storm runoff at different stages from
their various tributaries [10]. As discussed previously, higher-
pylthiocarbamate (EPTC), fonofos, terbufos, and linuron, only
peak concentrations were presented in Figure 7C for these
compounds. Three groups of the SWMI regressions in Table
6 were used: the multiyear databased Yr 1983–91 model, the
1986 databased Yr 86 model, and the 1986 Maumee River
model. The Maumee model was selected primarily because it
was the largest Lake Erie tributary with drainage area about
half of the White River watershed. As shown in Figure 7A to
C, all three models, in the majority of the cases, had ratios of
predicted percentiles of monitored concentrations in the range
of 0.5 to 5 even though the Yr 86 and Maumee models were
noticeably better than Yr 83–91 for the peak predictions. The
Yr 83–91 model predicted considerably higher peaks than the
other two models (1–13 times for most of the compounds, 46
and 165 times for EPTC and butylate, respectively). As dis-
cussed previously, the data on which the Yr 83–91 regression
is based were obtained over the nine-year period (1983–1991)
from the Lake Erie tributaries. The regression, and conse-
quently its predictions, are likely to be biased toward the ex-
treme runoff event values.
Although overpredictions of the peak concentrations by the
Yr 86 and Maumee models were not as severe as by the Yr
83–91 model, the bias to higher predicted concentrations was
quite consistent for most compounds (Fig. 7C). Specific rea-
sons for consistent over- or underprediction are not readily
apparent even though potential influence by factors such as
watershed characteristics and agricultural practices between
the two basins may be expected. The entire White River Basin
has a much larger drainage area than any of the SWMI model
calibration tributaries and covers a mix of diverse hydrogeo-
morphic regions from the relatively impervious northern till
plain to the better-drained southern glacial lowland [27]. Tile
drain contribution to pesticide concentrations in surface water
is also common in the region. As discussed previously, larger
drainage basins with diverse hydrogeological properties tend
to have lower peak concentrations. Additionally, some highly
nonuniform pesticide use patterns in the watershed may have
made it difficult to compare monitoring data from a given site
to the SMWI models that are based on homogeneous use rates
percentile concentrations (
storm driven and relatively less dependent on specific pesti-
cides properties. Similarly, low concentrations ( 95th per-
Ͼ95th percentile) are primarily
Ͻ
centile) and TWMC may be less associated with basin-scale
hydrology in the sense that low concentrations often occur in
the postapplication season (winter) and that the pesticide trans-
port, if any, is controlled more by the aged residues in soil
and thereby by the pesticide properties. For a given use, the
relative importance of the basin-scale characteristics, such as
watershed sizes, in determining pesticide occurrence varies
with time, equivalently in determining different percentile con-
centrations. This observation agrees with the results of Rich-
ards and Baker [10], who compared concentration exceedency
curves of different-size tributaries and found that smaller trib-
utaries deviate from larger ones significantly only at concen-
trations exceeding the 99th percentile.
Preliminary model evaluation
The transportability of the SWMI regressions to watersheds
other than the Lake Erie tributaries was evaluated by com-
paring the SWMI model predictions to monitored pesticide
concentrations from two other monitoring data sets in mid-
western corn states: one from a river system (White River at
Hazelton, IN, USA) and the other from a reservoir system
(Higginsville City Lake, MO, USA). The evaluation was to
test the robustness/sensitivity of the regression models to var-
iations in geohydrological/meteorological conditions, water-
shed size, and agricultural practices that are potentially dif-
ferent from the reference tributaries of the Lake Erie Basin.

306
Environ. Toxicol. Chem. 21, 2002
W. Chen et al.
Fig. 4. Regressions between the proposed pesticide surface water
mobility index (SWMI) and the basin use–normalized concentrations.
Data are derived from the single-year (1986) database of the six ag-
ricultural tributaries in the Lake Erie Basin, sampled from January to
Fig. 5. Regressions between the proposed pesticide surface water
mobility index (SWMI) and the basin use–normalized concentrations.
Data are derived from the single-year (1986) database of Honey Creek,
USA, sampled from January to September 1986. TWMC
weighted mean concentration.
ϭ time-
September 1986. TWMC
ϭ time-weighted mean concentration.
throughout a drainage basin. For example, the percentage of
acres treated with the corn herbicide butylate in 1993 was eight
times (four times in total amount) greater in southern Indiana
than in the central, resulting in observed peak concentrations
27- to 59-fold different among different subdrainage basins in
the White River watershed [27]. Other potential factors may
also include the nonagricultural uses of several pesticides with-
in the watershed as well as the sampling regimen for the White
River that might not capture concentration peaks as in the Lake
Erie Basin monitoring program.
shown in Figure 8. The Yr 86 model predicted the 95th-per-
centile and peak concentrations very well, particularly for the
raw water, with the prediction-to-measurement ratio ranging
0.9 to 1.2 for the 95th-percentile data and less than fourfold
for the peak concentrations. The overall performance of the
Yr 86 model was noticeably better than the Yr 83–91 model
even though both underestimated the annual TWMCs by a
factor of 2 to 5. The raw water annual TWMC was below 3
␮
g/L and the finished water below 2.0
␮g/L during the three-
year monitoring period.
The atrazine monitoring data from a much smaller water-
shed, the Higginsville City Lake, are presented in Figure 8
with annual TWMC and 95th-percentile and peak concentra-
tions in both raw and finished water systems during 1995–
1997. The corresponding SWMI model predictions are also
The relatively high runoff potential in the Higginsville City
Lake drainage basin is generally attributed to the dominant
soil types that had clay content in the range of 21 to 23% and
the common presence of silty loess in the soils [25]. These
relatively heavy soil types and gently to strongly sloping geo-

A pesticide surface water mobility index
Environ. Toxicol. Chem. 21, 2002
307
Fig. 7. Comparison of model-predicted concentrations to the observed
data of 12 pesticides in White River at Hazelton (IN, USA) 1991–
Fig. 6. The effect of river size on the pesticide surface water mobility
index (SWMI) model predictions for the peak and annual time-weight-
ed mean concentrations (TWMC). Refer to Figure 2 for creek and
river locations.
1996. TWMC
ϭ time-weighted mean concentration; EPTC ϭ s-ethyl
dipropylthiocarbamate.
peared to play an important role in determining pesticide oc-
currence in surface water, especially for the pesticide peak and
morphology appear to resemble the general characteristics of
the Lake Erie tributary drainage basins, which may have led
to the closer model agreement with the 95th-percentile and
peak concentrations. Compared to the river systems, however,
water in the Higginsville City Lake is expected to be relatively
static, resulting in longer residence time for pesticide residues
in the lake than in a river system. Since TWMC is a long-
term parameter while peak concentrations are temporary, un-
derestimation of TWMC in reservoirs/lakes by the flow
stream–based SWMI models may reflect the effect of water
flow on the annual averages of potential pesticide residues
(dilution factor).
higher-percentile (e.g.,
Ͼ95th percentile) concentrations.
However, specific pesticide properties become more important
at low concentration ranges. A preliminary model evaluation
was conducted with two independent pesticide surface water
CONCLUSIONS
This paper describes a pesticide surface water mobility in-
dex to quantify the movement of pesticides to surface water
via overland runoff and erosion on various watershed scales.
The proposed SWMI was found to statistically correlate with
the various percentiles of pesticide concentrations in rivers
obtained from a nine-year monitoring database for the Lake
Erie Basin tributaries. The regression improved significantly
with more accurate annual pesticide use information. This ob-
servation underlines the importance of collecting relevant pes-
ticide use data within the watersheds in order to obtain a more
realistic quantitative analysis of regional scale pesticide sur-
face water monitoring programs. Watershed hydrology ap-
Fig. 8. Comparison of model-predicted concentrations to the observed
data of atrazine in raw and finish waters at the Higginsville City Lake
(MO, USA) 1995–1997. TWMC
tion.
ϭ time-weighted mean concentra-

308
Environ. Toxicol. Chem. 21, 2002
W. Chen et al.
1995. Relations between pesticide use and riverine flux in the
Mississippi River Basin. Chemosphere 31:3305–3321.
12. Battaglin WA, Goolsby DA, 1997. Statistical modeling of agri-
cultural chemical occurrence in midwestern rivers. J Hydrol 196:
1–25.
13. Crawford CG. 1995. Occurrence of pesticides in the White River,
Indiana, 1991–95. U.S. Geological Survey Fact Sheet 233-95.
White River Basin Study, U.S. Geological Survey, Indianapolis,
IN.
14. Syngenta Crop Protection. 1997. Voluntary atrazine monitoring
program at selected community water systems: Missouri 1996.
Technical Report 6-97. Syngenta Crop Protection, Greensboro,
NC, USA.
15. Jury WA, Focht DD, Farmer WJ. 1987. Evaluation of pesticide
groundwater pollution potential from standard indices of soil-
chemical adsorption and biodegradation. J Environ Qual 16:422–
428.
monitoring data sets, one obtained from a much larger scale
river system (drainage area 2,938,382 ha, White River at Ha-
zelton) and the other from a small reservoir (1,386 ha, Hig-
ginsville City Lake). The SWMI models, based on the 1986
monitoring data and the 95th-percentile model based on mul-
tiyear data (1983–1991) from the six Lake Erie tributaries,
demonstrated predictions of TWMC and peak and 95th-per-
centile concentrations mostly within fivefold of the observed
data. It is suggested that these regression models may be used
as simple screening tools to provide quick and reasonably
conservative estimates of pesticide concentrations in surface
water. For example, when monitoring data are not available,
the TWMC and the 95th-percentile estimates by the SWMI
models for a surface water system can be used for chronic and
acute exposure assessments for ecological or human health
purposes if the water system is used as a public drinking water
supply.
16. Gustafson DI. 1989. Ground water ubiquity score: A simple meth-
od for assessing pesticide leachability. Environ Toxicol Chem 8:
339–357.
17. Kleveno JJ, Loague K, Green RE. 1992. Evaluation of a pesticide
mobility index: Impact of recharge variation and soil profile het-
erogeneity. J Contam Hydrol 11:83–99.
18. Leonard RA, Knisel WG, Still DA. 1987. GLEAMS: Ground-
water loading effects of agricultural management systems. Trans
ASAE (Am Soc Agric Eng) 30:1403–1418.
19. U.S. Department of Agriculture Soil Conservation Service. 1985.
National engineering handbook, section 4—Hydrology (Part 1 of
2). PB86180494. Technical Report. U.S. National Technical In-
formation Service, Springfield, VA.
20. Haith DA, Mandel R, Wu RS. 1996. Generalized Watershed
Loading Functions, Ver 2.0. User’s Manual. Department of Ag-
ricultural and Biological Engineering, Cornell University, Ithaca,
NY, USA.
21. Burns LA, Allen BW, Barber MC, Bird SL, Cheplick JM, Fendley
MJ, Hartel DR, Kittner CA, Mayer FL, Suarez LA Jr, Wooten
SE. 1992. Pesticide and Industrial Chemical Risk Analysis and
Hazard Assessment (PIRANHA), Ver 3.0. User’s Manual. En-
vironmental Research Laboratory, Office of Research and De-
velopment, U.S. Environmental Protection Agency, Athens, GA.
22. Waldron AC. 1989. Survey application of potential agricultural
pollutants in the Lake Erie Basin of Ohio: Pesticide use on major
crops—1986. Special Circular 120. Cooperative Extension Ser-
vice Bulletin 787. Ohio State University, Columbus, OH, USA.
23. Shelton LR. 1994. Field guide for collecting and processing
stream-water samples for the National Water-Quality Assessment
Program. U.S. Geological Survey Open-File Report 94-455.
White River Basin Study, U.S. Geological Survey, Indianapolis,
IN.
24. Anderson JE, Gianessi LP. 1995. Pesticide use in the White River
Basin. Technical Report. National Center for Food and Agricul-
tural Policy, Washington, DC.
25. Burr SA. 1998. Modeling of atrazine in the Higginsville City
Lake Watershed. MS thesis. University of Missouri—Columbia,
Columbia, MO, USA.
Acknowledgement—We thank P. Richards and D. Baker for providing
the Lake Erie Basin monitoring data and insightful discussions during
preparation of this work. Comments from M. Barrett, R. Parker, P.
Coody, P. Hendley, K. Travis, N. Poletika, S. Jackson, R. Jones, D.
Gustafson, and other colleagues are appreciated.
REFERENCES
1. Mullins JA, Carsel RF, Scarbrough JE, Ivery AM. 1993. PRZM-
2, a model for predicting pesticide fate in the crop root and un-
saturated soil zones: Users manual for release 2.0. EPA/600/R-
93/046. Technical Report. Environmental Research Laboratory,
U.S. Environmental Protection Agency, Athens, GA.
2. Knisel WG. 1993. GLEAMS: Groundwater loading effects of
agricultural management systems, version 2.10. UGA-CPES-
BAED Publication 5. Technical Report. University of Georgia,
Coastal Plain Experimental Station, Tifton, GA, USA.
3. Luxmoore RJ, Sharma ML. 1980. Runoff responses to soil het-
erogeneity: Experimental and simulation comparisons for two
contracting watersheds. Water Resour Res 16:675–684.
4. Freez RA. 1980. A stochastic-conceptual analysis of rainfall-run-
off processes on a hill slope. Water Resour Res 16:319–408.
5. Seyfried MS, Wilcox BP. 1995. Scale and the nature of spatial
variability: Field examples having implications for hydrologic
modeling. Water Resour Res 31:173–184.
6. Merz B, Plate EJ. 1997. An analysis of the effects of spatial
variability of soil and soil moisture on runoff. Water Resour Res
33:2909–2922.
7. Dooge JCI, Bruen M. 1997. Scaling effects on moisture flues at
unvegetated land surfaces. Water Resour Res 33:2923–2927.
8. Solomon KR, Baker DB, Richards RP, Dixon KR, Klaine SJ, La
Point W, Kendall RJ, Weisskopf CP, Giddings JM, Giesy JP, Hall
LW, Williams WM. 1996. Ecological risk assessment of atrazine
in north American surface waters. Environ Toxicol Chem 15:31–
76.
26. Sachs L. 1984. Angewandte Statistik, Anwendung Statistischer
Methoden, 6th ed. Springer-Verlag, New York, NY, USA.
27. Crawford CG. 1996. Occurrence of pesticides in the White River,
Indiana, 1991–95. U.S. Geological Survey Fact Sheet 119-96.
Influence of natural and human factors on pesticide concentrations
in surface waters of the White River Basin, Indiana. U.S. Geo-
logical Survey, Indianapolis, IN.
28. U.S. Environmental Protection Agency. 2001. One Liner Data-
base, Ver 3.04. Environmental Fate and Ground Water Branch,
Office of Pesticide Programs, Washington, DC.
9. Wauchope RD. 1978. The pesticide content of surface water drain-
ing from agricultural fields—A review. J Environ Qual 7:459–
472.
10. Richards RP, Baker DB. 1993. Pesticide concentration patterns
in agricultural drainage networks in the Lake Erie Basin. Environ
Toxicol Chem 12:13–26.
11. Larson SJ, Capel PD, Goolsby DA, Zaugg SD, Sandstrom MW.