JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

DECEMBER AMERICAN WATER RESOURCES ASSOCIATION 2005

ESTIMATION OF STREAMFLOW, BASE FLOW, AND NITRATE-NITROGEN

LOADS IN IOWA USING MULTIPLE LINEAR REGRESSION MODELS1

Keith E. Schilling and Calvin F. Wolter2

ABSTRACT: Nineteen variables, including precipitation, soils and Smakhtin, 2001). Although regional regression mod-
geology, land use, and basin morphologic characteristics, were eval- els are considered the most widely used method for
uated to develop Iowa regression models to predict total streamflow
(Q), base flow (Qb), storm flow (Qs) and base flow percentage (%Qb)
estimating low flow characteristics at ungauged sites
in gauged and ungauged watersheds in the state. Discharge records (Smakhtin, 2001), they are necessarily developed for
from a set of 33 watersheds across the state for the 1980 to 2000 specific physiographic conditions in the region of
period were separated into Qb and Qs. Multiple linear regression interest. It follows that watersheds in regions with
found that 75.5 percent of long term average Q was explained by different climate, soils, geology, and vegetation will
rainfall, sand content, and row crop percentage variables, whereas
88.5 percent of Qb was explained by these three variables plus per-
have different streamflow or base flow conditions.
meability and floodplain area variables. Qs was explained by aver- Hence, a review of variables included in various
age rainfall and %Qb was a function of row crop percentage, regression models would be rather exhaustive,
permeability, and basin slope variables. Regional regression models although Smakhtin (2001) does identify some basin
developed for long term average Q and Qb were adapted to annual and climate characteristics common to many models,
rainfall and showed good correlation between measured and pre-
dicted values. Combining the regression model for Q with an esti-
including basic watershed morphologic descriptors
mate of mean annual nitrate concentration, a map of potential (basin area, slope, perimeter), climate (mean annual
nitrate loads in the state was produced. Results from this study precipitation), vegetation patterns (forested areas),
have important implications for understanding geomorphic and and various soil and geologic indices.
land use controls on streamflow and base flow in Iowa watersheds Lacey and Grayson (1998) identified common topo-
and similar agriculture dominated watersheds in the glaciated Mid-
west.
graphic variables used by other investigators to
(KEY TERMS: streamflow; base flow; nitrate; Iowa; loads; agricul- describe base flow including catchment area, total
tural hydrology.) length of stream network, catchment relief, elevation,
and main stem stream length. Others have empha-
Schilling, Keith E. and Calvin F. Wolter, 2005. Estimation of Streamflow, Base sized stream density as an important variable in pre-
Flow, and Nitrate-Nitrogen Loads in Iowa Using Multiple Linear Regression
Models. Journal of the American Water Resources Association (JAWRA) 41(6):
dicting low flow conditions (Gregory and Walling,
1333-1346. 1968; Zecharias and Brutsaert, 1988) as well as basin
relief and average basin slope (Vogel and Kroll, 1992).
Lacey and Grayson (1998) developed a model for 114
catchments in Victoria, Australia, based on, among
INTRODUCTION other variables, the concept of “geology vegetation”
groups to link vegetation communities to underlying
rock types. They further utilized dimensionless topo-
Prediction of streamflow and base flow at
graphic parameters to model a dimensionless Base
ungauged sites has been recognized by hydrologists
Flow Index (BFI). A soil properties database, with
worldwide as an important issue, particularly with
properties such as drainage class, permeability, and
respect to low flow hydrology (see review by

1Paper No. 04165 of the Journal of the American Water Resources Association (JAWRA) (Copyright © 2005). Discussions are open until
June 1, 2006.
2 Respectively, Research Geologist and GIS Coordinator, Iowa Geological Survey, 109 Trowbridge Hall, Iowa City, Iowa 52242
(E-Mail/Schilling: kschilling@[Link]).

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1333 JAWRA

 17521688, 2005, 6, Downloaded from [Link] by Cochrane Thailand, Wiley Online Library on [01/08/2024]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SCHILLING AND WOLTER

major soil groups, has been used in the United King- pollutants like sediment, pesticides, or bacteria are
dom and Europe to study low flow behavior in delivered to streams primarily via overland runoff, so
streams (Gustard et al., 1992; Gustard and Irving, regression models are needed to predict regional Q
1994). and Qs for determination of these NPS loads.
Kroll et al. (2004) utilized a digital elevation model, The objectives of this paper are to: (1) evaluate cli-
the U.S. Department of Agriculture State Soil Geo- mate, soils and geology, land use, and basin morpholo-
graphic (STATSGO) grids, and climate data from the gy variables using multiple linear regression analysis
Spatial Climate Analysis Service (PRISM) grids to to develop regression models for Q, Qb, Qs, and %Qb
develop low flow regression models for regions across for Iowa watersheds; (2) determine the efficacy of
the United States. For Iowa, they found that drainage using the regression models to predict annual Q and
area (DA), precipitation intensity, and three monthly Qb in watersheds based on variable climate inputs;
precipitation variables predicted seven-day, 10-year and (3) couple the statewide Q estimates with nitrate
low flow with an adjusted R2 of 90.9 and a standard concentration estimates to provide a map of potential
error of 68.5 percent. A second regression equation nitrate pollution loads in the state.
with added BFI and base flow recession variables
improved the adjusted R2 and standard error (93.1
and 34.7 percent, respectively). Lara (1986) consid-
ered Iowa’s unique landform regions (Prior, 1991) METHODS
when he developed regional regression equations for
flood frequency prediction based on drainage area. A set of 33 Iowa watersheds (Figure 1) was identi-
In this paper, 19 variables were evaluated, includ- fied for analysis based on two criteria: their bound-
ing precipitation, soils and geology, land use, and aries were entirely within the state; and the
basin morphologic characteristics, to develop Iowa watershed had a stream gauge near their mouth, with
regression models to predict total streamflow (Q), a minimum of 20 years of daily streamflow record.
base flow (Qb), storm flow (Qs), and base flow per- The selected watersheds are located throughout the
centage (%Qb) in gauged and ungauged watersheds in state and are representative of the major landform
the state. The purpose for developing the models was regions. Daily streamflow records from the watershed
directed at providing understanding of the underlying gauging stations (USGS, 2004) were separated into
geomorphic and land use controls on streamflow and base flow and storm flow components using an auto-
base flow in Iowa. While studies of low flow are clear- mated hydrograph separation program (Sloto and
ly needed for water resources planning and wastewa- Crouse, 1996). Output from the program was
ter permitting, broader studies related to long term tabulated as annual means and an average value was
streamflow and base flow runoff from watersheds are computed for the 1980 to 2000 period. A 1980 to 2000
also of emerging importance to federal nonpoint period of streamflow record was analyzed to provide a
source (NPS) programs. One challenge posed under long term average streamflow condition that account-
the Total Maximum Daily Load (TMDL) program ed for variable climatic inputs but was not long
established by Section 303(d) of the Clean Water Act enough to capture the period of increasing streamflow
is to determine pollutant loads from nonpoint sources. and base flow in Iowa (Schilling and Libra, 2003). Qs
While waste loads from point sources (i.e., pipe efflu- was the difference between Q and Qb calculated by
ent) are relatively easy to assess, load allocations the program. %Qb was calculated by dividing Q by
from nonpoint sources are difficult to determine, Qb. All streamflow components (Q, Qb, Qs) are evalu-
requiring an understanding of the concentration and ated as unit discharges (Table 1).
rate of discharge of pollutants over large geographical Drainage basin characteristics were computed for
areas (Schilling and Wolter, 2001). A method to assess the 33 watersheds using a geographic information
discharge from large watershed areas is needed to system (GIS), a 30-m Digital Elevation Model (DEM)
couple with concentration data for development of tar- (USGS, 1999), and the computer program Basinsoft
geting strategies to reduce nonpoint source loads. (Eash, 1994). Although 27 characteristics are quanti-
Furthermore, the nature of NPS pollutant dis- fied by the Basinsoft program, only those properties
charges in Iowa demands that regression models con- that were either dimensionless or had unit values
sider relations to hydrograph components such as Q, were retained for analysis so comparisons could be
Qb, Qs, and %Qb. For example, nitrate-nitrogen is pri- made across watersheds of different sizes. Total
marily delivered to Iowa steams through ground drainage area was included in the analysis because of
water discharge and tile drainage (Hallberg, 1987; its significance in other regression model studies but
Schilling and Zhang, 2004), so prediction of regional was evaluated as a logarithm because of its log-
Qb or %Qb would aid development of NPS load alloca- normal distribution (skewness >1). All other basin
tions for nitrate. In contrast, other agricultural characteristics were normally distributed. Ten basin

JAWRA 1334 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

 17521688, 2005, 6, Downloaded from [Link] by Cochrane Thailand, Wiley Online Library on [01/08/2024]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ESTIMATION OF STREAMFLOW, BASE FLOW, AND NITRATE-NITROGEN LOADS IN IOWA USING MULTIPLE LINEAR REGRESSION MODELS

Figure 1. Location of 33 Watersheds Evaluated in Regional Regression Models. Watershed

numbers identified in Table 1. Precipitation trends from 1960 to 1990.

characteristics were eventually retained for regres- land area (Schilling and Libra, 2000). Because of this
sion modeling (Table 2). dominant land use type in Iowa, the percentage of
Digital soil information was obtained from the Iowa row crop land in each watershed (%RC) was deter-
Soil Properties and Interpretations Database mined by summing 30 m pixels in a 2002 statewide
(ISPAID) (ISU, 1996), a GIS grid coverage of state- digital land cover map (IGS, 2004). A 2002 GIS cover-
wide soil series available at 30 m resolution. For each age was used because of its increased resolution of
watershed, the percentage of sand, clay, and organic vegetation compared to earlier maps, recognizing that
matter in the soil horizon, the percentage of the land use patterns have not changed appreciably since
watershed underlain by hydric soils, the percentage of an earlier 1992 land use map of Iowa was generated
rock as parent material in the watershed, and the (Giglierano, 1999).
average permeability of soils in the watershed were Precipitation data for the 1960 to 1990 period sum-
computed. The permeability class abbreviations con- marized by the Natural Resources Conservation Ser-
tained in the database were modified to reflect actual vice (NRCS, 2000) were obtained as a GIS grid
infiltration values in inches per hour. The percentage coverage. Although the period of record for the rain-
of alluvium (%ALLUV) contained in each watershed fall average was not the same as the hydrograph peri-
was calculated by summing the areas of soil series od, the 30-year record of rainfall was sufficiently long
units that typified floodplain alluvium as parent to provide a value for long term average rainfall in
material. This property was analogous to the percent- Iowa. As shown in Figure 1, average rainfall increases
age of the watershed in a floodplain landscape posi- from northwest to southeast, from less than 28 inches
tion. (711 mm) to more than 36 inches (914 mm) annually.
Land use in Iowa is characterized by agriculture, Average annual precipitation was determined for the
consisting primarily of annual row crops of corn and 33 watersheds by intersecting the watershed bound-
soybeans. Between the years of 1980 and 2000, Iowa aries with the precipitation coverage and calculating
averaged approximately 12 to 14 million acres (48,600 a weighted average value.
to 55,700 km 2 ) of corn and 8 to 11 million acres Multiple linear regression (MLR) was performed to
(32,400 to 44,550 km2) of soybeans planted each year identify the important morphologic, soil and geologic,
(ISU, 2004), averaging about 60 percent of the total land use, and climate variables that would best

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1335 JAWRA

 17521688, 2005, 6, Downloaded from [Link] by Cochrane Thailand, Wiley Online Library on [01/08/2024]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SCHILLING AND WOLTER

TABLE 1. Summary of Watersheds Evaluated in Multiple Linear Regression Model.

(Numbers correspond to watershed locations shown on Figure 1.)

Drainage Total Base Average Row

Map Area Streamflow Flow Precipitation Crop Basin
Location STORET_ID Name (m2) (in) (in) (in) (percent) Slope

1 10070001 Beaver Creek Near Cedar Falls 767.4 11.1 6.5 33.0 81.9 2.48
2 10070003 West Fork Cedar River at Finchford 850.2 11.3 7.3 33.3 79.2 1.97
3 10220001 Turkey River Near Garber 1552.8 10.9 6.9 33.3 57.4 6.34
4 10250001 South Racoon River Near Redfield 979.5 7.9 4.6 33.0 63.2 5.35
5 10270001 Thompson Fork of Grand River 695.1 8.3 2.8 35.1 30.3 7.53
6 10360001 East Nishnabotna River NearShenandoah 1021.1 9.1 5.6 33.1 65.7 6.2
7 10370001 North Racoon River Near Jefferson 1586.9 8.7 5.0 31.1 84.1 1.61
8 10400001 Boone River Near Stratford 887.4 10.3 5.2 31.1 84.9 1.34
9 10430002 Soldier River Near Pisgah 408.3 6.9 4.3 29.7 67.8 8.57
10 10440001 Cedar Creek Near Oakland Mills 532.9 10.5 2.8 35.7 58.2 3.46
11 10490001 North Fork Maquoketa River Near Maquok 589.4 9.1 6.1 34.8 49.8 7.33
12 10490002 Maquoketa River Near Maquoketa 956.6 10.4 7.0 34.9 61.1 5.25
13 10500001 Indian Creek Near Colfax 956.6 11.6 6.2 33.4 73.0 2.95
14 10520001 Old Mans Creek Near Iowa City 589.4 10.3 5.0 35.6 57.9 6
15 10540001 North Skunk River Near Sigourney 636.7 10.3 5.3 34.9 60.2 5.52
16 10620001 South Skunk River Near Oskaloosa 1639.3 10.4 6.0 33.4 70.7 3.25
17 10630001 Whitebreast Creek Near Knoxville 1639.3 10.0 2.3 35.3 23.6 7.65
18 10630002 Cedar Creek Near Bussey 369.4 9.7 2.5 36.2 23.4 8.76
19 10640003 Iowa River Upstream Marshalltown 1467.4 10.5 6.5 32.6 80.0 2.24
20 10650001 West Nishnabotna River Near Malvern 968.2 8.6 5.6 32.6 73.9 6.23
21 10670002 Maple River Near Mapleton 643.8 8.0 5.4 30.3 79.6 4.68
22 10730001 West Nodaway River Near Shambaugh 789.9 8.6 4.1 34.6 50.8 6.18
23 10750001 Floyd River Near Sioux City 885.9 6.0 3.9 27.0 85.4 3.07
24 10770001 Beaver Creek Near Grimes 885.9 9.1 4.8 32.9 79.3 1.84
25 10810001 North Racoon River Near Sac City 708.6 9.7 5.5 30.7 85.4 1.34
26 10850002 South Skunk River Near Cambridge 583.8 10.5 6.7 33.0 76.9 1.95
27 10850003 Skunk River Upstream Ames 2333.3 11.7 6.1 33.0 80.2 1.79
28 10910001 Middle River Near Indianola 489.2 8.4 3.5 34.1 38.6 6.95
29 10910002 North River Near Norwalk 489.2 8.9 3.8 33.1 52.0 5.15
30 10910003 South River Near Ackworth 474.3 8.3 2.3 35.0 23.5 7.23
31 10920001 English River at Riverside (Kalona) 626.7 9.7 4.0 35.0 57.4 5.15
32 10970002 West Fork Ditch at Hornik 402.4 5.7 3.7 27.5 80.8 4.4
33 10990001 Iowa River Near Rowan 426.9 10.1 6.0 31.4 82.8 1.62

explain average Q, Qb, Qs, and %Qb in the water- explanatory variable and ∈ is the remaining unex-
sheds. The regression model has the form (Helsel and plained noise in the data (the error). A combination of
Hirsch, 2002) stepwise, forward, and backward regression proce-
dures was performed using MINITAB Release 13 sta-
y = β0 + β1x1 + β2x2 + … + βkxk + ∈ (1) tistical software package (Minitab, 2000) to identify
potential explanatory variables for the regional
where y is the response variable, β0 is the intercept, regression models. Stepwise procedures generally
β1 is the slope coefficient for the first explanatory involve adding or removing variables in the model
variable, β k is the slope coefficient for the kth and checking the significance (α = 0.15 to enter or

JAWRA 1336 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

 17521688, 2005, 6, Downloaded from [Link] by Cochrane Thailand, Wiley Online Library on [01/08/2024]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ESTIMATION OF STREAMFLOW, BASE FLOW, AND NITRATE-NITROGEN LOADS IN IOWA USING MULTIPLE LINEAR REGRESSION MODELS

TABLE 2. Description of Explanatory Variables.

Class of
Explanatory
Variable Symbol Variable Description Source

Climate RAIN 30-year average rainfall 1960 to 1990 NRCS (2000)

Land Use RC Percentage of watershed area in annual row crop production IGS (2004)

Soils/Geology %HYDRIC Percentage of watershed area with hydric soil classification IGS (2003)
%SAND Average sand content of soil mapping units in watershed IGS (2003)
%CLAY Average clay content of soil mapping units in watershed IGS (2003)
%ORG Average organic matter content of soil mapping units in watershed IGS (2003)
%ROCK Percentage of watershed area with rock as soil parent material IGS (2003)
PERM Average soil infiltration rate (in/hr) x 100 IGS (2003)
%ALLUV Percentage of watershed area with alluvium or sediment as soil parent material IGS (2003)

Basin Characteristics LOGTDA Logarithm of total drainage area (square mile) of watershed USGS (1999)
BS Average basin slope determined by average slope of 30 m cells in basin USGS (1999)
BW Effective basin width (mile), BW = CDA/basin length USGS (1999)
RR Relative relief, (foot/mile), RR = basin relief/basin perimeter USGS (1999)
CR Compactness ratio (dim), ratio of basin perimeter to the circumference of a circle USGS (1999)
of equal area USGS (1999)
SF Ratio of basin length to effective basin width USGS (1999)
MCS Main channel slope, (foot/mile), river coverage IGS (2000)
MCSR Main channel sinuosity ratio, (dim), MCSR = main channel length/basin length IGS (2000)
SD Stream density (mile/mile2), SD = total stream length/CDA IGS (2000)
LOGSR Slope ratio of main-channel slope to basin slope (dim) IGS (2000)

Notes: NRCS = Natural Resources Conservation Service, IDNR = Iowa Department of Natural Resources, DEM = digital elevation model,
ISPAID = Iowa Soil Properties and Interpretations Database.

remove from the model). To address a potential prob- RESULTS AND DISCUSSION
lem with multicolinearity among the explanatory
variables, a correlation matrix among all variables Explanatory Variable Classes
was used to judge the significance of the relation
between two variables. Any two variables with a sig-
nificant correlation (p < 0.1) were not included togeth- The 19 explanatory variables were divided into
er in the regression model. For example, models were four general classes of variables to evaluate how well
not allowed to contain both %ORG and %HYDRIC, or each class of variable explains Q, Qb, Qs, and %Qb
%SAND and %CLAY variables since they were each (Table 3). Linear regressions were conducted for pre-
highly correlated and measure similar attributes. cipitation and %RC variables and the four streamflow
Furthermore, professional judgment was used variables (i.e., linear regression between Q and pre-
throughout the process to identify potential linkages cipitation, Qb and precipitation, etc.), whereas for-
among variables. The goodness of fit of the regression ward, backward, and stepwise MLRs were conducted
models was judged by minimizing Mallow’s Cp statis- to identify important explanatory variables within
tic and maximizing the adjusted R2 value (Helsel and soil/geology and basin characteristics classes. Results
Hirsch, 2002). Mallow’s Cp is a statistical measure suggest that precipitation explained nearly 60 percent
recommended to assess the balance between explain- of average storm flow and 24 percent of Q, but had lit-
ing as much variability as possible with fewest num- tle relation to average base flow (Table 3). On the
bers of coefficients (Helsel and Hirsch, 1991). Overall, other hand, row crop land use explained approximate-
the advice of Helsel and Hirsch (2002) was followed to ly 41 percent of base flow and 55 percent of base flow
use the computer to perform large computations asso- percentage (Table 3) but had little relation to average
ciated with many regression models, but retained sci- Q and negative relation to average Qs. It would seem
entific judgment to select the regression model that counterintuitive that row crop land use would posi-
explained the most variability in the streamflow data tively affect base flow and negatively affect storm flow
with the fewest number of easily measured explanato- until one considers how row crop lands are positioned
ry variables. on the Iowa landscape. Row crops tend to be located

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1337 JAWRA

 17521688, 2005, 6, Downloaded from [Link] by Cochrane Thailand, Wiley Online Library on [01/08/2024]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SCHILLING AND WOLTER

TABLE 3. Results of Best Model Fit of Variables in Explanatory Variable Classes to

Explain Q, Qb, Qs, and %Qb. Linear regression used in climate and land use
evaluation and MLR used in soils/geology and basin characteristics evaluation.

Class of
Explanatory Streamflow
Variable Condition Variable in the Model Adjusted R2 p-Value

Climate Q RAIN 23.6 0.002

Qb RAIN 3.5 0.152
Qs RAIN 59.9 0.000
%Qb RAIN 44.7 0.000

Land Use Q %RC 0.0 0.696

Qb %RC 41.3 0.000
Qs -%RC 32.5 0.000
%Qb %RC 55.0 0.000

Soils/Geology Q %CLAY, %HYDRIC 49.8 0.000

Qb %ORG, PERM, %ALLUV 74.2 0.000
Qs -PERM, -%CLAY, -%ALLUV, %ROCK 59.1 0.000
%Qb PERM, %ALLUV, %CLAY, %ORG 68.4 0.000

Basin Characteristics Q SD, BS 10.3 0.068

Qb -BS, LOGTDA 31.7 0.002
Qs -LOGTDA, MCSR 17.1 0.023
%Qb LOGTDA, LOGSR, -MCSR 29.0 0.005

on flat landscapes (i.e., recently glaciated terrain, relations with PERM, %CLAY, and %ALLUV, and a
upland divide areas, floodplains) or on hillslopes with positive %ROCK coefficient (56.1 percent). The MLR
some degree of conservation (i.e., terraces, conserva- model results using soil and geology variables indi-
tion tillage contour cropping), where substantial infil- cate that these variables play an important role in
tration of rainfall would be expected to occur. Greater generating and partitioning streamflow in Iowa.
infiltration increases ground water levels and sus- As a class of variable, basin characteristics were
tains higher base flow in streams. Other studies have not highly related to Q, Qb, Qs, and %Qb (Table 3).
shown that increasing row crop land use in Iowa over The best MLR model for Q included stream density
the last 60 years is significantly related to increasing (SD) and basin slope (BS) but had an adjusted R2 of
base flow (Schilling and Libra, 2003; Schilling, 2005). only 10.7 percent of Q. For Qb, variables, LOGTDA
Several soil and geology variables explain as much and negative BS, had an adjusted R2 of 31.7 percent,
as 74.2 percent of the streamflow variables (Table 3). which suggests that larger, low relief watersheds
For Q, %SAND alone had an adjusted R2 of 35.6 per- would tend to contribute more base flow. Likewise,
cent, and the combination of negative %CLAY and basin characteristics (-) LOGTDA and MCSR suggest
%HYDRIC had an adjusted R2 of 49.8 percent. The that smaller, steeply sloping watersheds contribute
coefficient on %CLAY was negative whereas the coef- more storm flow runoff.
ficient on %SAND was positive, which indicates the
inverse relation between the two textural variables.
Three variables, %ORG, PERM, and %ALLUV Regional Regression Models (Objective 1)
explained 74.2 percent of %Qb, whereas four soil and
geology variables (PERM, %ALLUV, %CLAY, and Final regional regression models for Q, Qb, Qs, and
%ORG) explained 68.4 percent of %Qb. The three %Qb represented a balance between explaining as
variables included in the Qb equation suggest that much variability as possible versus including only
greater base flow is associated with watersheds with highly significant and easily measured variables. For
high infiltration potential in uplands, larger flood- Q, 19 possible explanatory variables were reduced to
plains, and greater water retention capacity in organ- three, so that long term average Q was explained by
ic rich soils (e.g., wetlands, lowland areas). The rainfall, sand content, and amount of row crop in a
highest adjusted R 2 for Qs contained negative watershed by the expression

JAWRA 1338 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

 17521688, 2005, 6, Downloaded from [Link] by Cochrane Thailand, Wiley Online Library on [01/08/2024]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ESTIMATION OF STREAMFLOW, BASE FLOW, AND NITRATE-NITROGEN LOADS IN IOWA USING MULTIPLE LINEAR REGRESSION MODELS

Q = 0.645(RAIN) + 0.0607(%SAND) (2) which has an adjusted R2 of 86.8 percent and Cp fac-
+ 0.0519(%RC) - 16.3 tor of 6.0. The prediction of long term Qb with the
equation is highly significant for the 33 watersheds
The adjusted R2 and Cp for the model was 75.5 per- (Figure 2). The added PERM and %ALLUV variables
cent and -1.5, respectively, and the relation between in the Qb equation, along with the RAIN, %SAND,
measured and predicted values was highly significant and %RC variables, clearly points to factors that
(Figure 2). The model variables appear to reflect dom- influence ground water recharge in Iowa, since base
inant hydrologic processes in Iowa. While rainfall pat- flow and ground water recharge are generally consid-
terns in Iowa clearly play a large role in predicting Q ered synonymous. The inclusion of the PERM variable
(Table 3), the inclusion of the %SAND variable sug- suggests that greater ground water recharge occurs in
gests that Q is at least partly dependent on the areas underlain by soils with higher infiltration
hydraulic conductivity of the soil surface layer to capacity, whereas the %ALLUV variable suggests that
transmit the rainfall runoff to streams. The %RC watersheds with more proportion of their land area in
variable further suggests that regions with greater floodplain contribute more base flow to streams.
streamflow in Iowa are under row crop cultivation. In Again, the amount of land use under row crop cultiva-
greater %RC regions, higher Q is largely a water bud- tion plays an important role in determining how
get issue. Because annual ET losses from annual row much base flow is delivered to Iowa streams. Identifi-
crops are limited to a shorter summer growing cycle, cation of the important variables controlling Qb in
rainfall falling on fallow fields throughout much of Iowa may have important implications for targeting
the year is free to migrate to streams without capture best management practices (BMPs) in areas where
and uptake by vegetation. The strong relation excessive ground water recharge may occur and thus
between the base flow portion of total streamflow and where greater nitrate leaching is possible.
row crop land use is captured by inclusion of the %RC The MLR model for Qs is, for the most part, an
variable in total Q equation since the relation of Q inverse of Qb, given by the expression:
and %RC was not significant (Table 3).
Qs = 0.529(RAIN) - 0.0121(PERM)
- 0.045(%ALLUV) - 10.99 (4)

which has an adjusted R2 of 79.9 percent and Cp fac-

tor of 6.0 (Figure 2). The Qs relation suggests that
watersheds with low permeability and low percentage
of land area in floodplain contribute more storm flow
to streams. RAIN was clearly the dominant variable
in the Qs equation, indicated by the amount of vari-
ance explained by rainfall alone (Table 3).
The RAIN variable was not included in the %Qb
MLR model, which was reduced to three variables in
the expression

%Qb = 0.838(%RC) + 0.113(PERM)

+ 3.87(BS) - 35.7 (5)

where the adjusted R2 was 88.5 percent with a Cp of

11.1. While the fit between measured and predicted
values was very good (Figure 2), the adjusted R2 for
Figure 2. Relationship Between Measured Annual Q, Qb, the equation could have been improved to more than
Qs, and %Qb to Values Predicted Using Multiple 92 percent with addition of more explanatory vari-
Linear Regression Models in 33 Watersheds. ables. However, this is an example of how scientific
judgment was used to winnow the variables to only
The MLR model for prediction of long term base those deemed most important and relevant. The MLR
flow in Iowa includes the three Q variables and adds model with the highest adjusted R 2 included
PERM and %ALLUV in the expression (-)%HYDRIC, %SAND, %CLAY, and RR variables,
but these variables contributed relatively little
Qb = 0.247(RAIN) + 0.047(%SAND) + 0.063(%RC) to improvement in the overall R 2 . Furthermore,
+ 0.0142(PERM) + 0.0528(%ALLUV) – 11.2 the inclusion of two particle size classes in the same
(3)

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1339 JAWRA

 17521688, 2005, 6, Downloaded from [Link] by Cochrane Thailand, Wiley Online Library on [01/08/2024]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SCHILLING AND WOLTER

equation violated the a priori guidelines against using Western Corn Belt Plains have been delineated, and
collinear variables in this MLR study. portions of four other Level IV ecoregions are included
The %RC variable largely governs the partitioning in Iowa (Chapman et al., 2002). Eight of the ten Iowa
of total streamflow into base flow or storm flow com- regions contain watershed areas that can be summa-
ponents (i.e., %Qb) in the MLR model evidenced by rized to assess average Q and Qb in each ecoregion
the amount of variance explained by the %RC vari- (Table 4). Highest average Q and Qb (11.8 and 7.7 in
able alone (Table 3). The addition of a basin charac- or 300 m and 196 mm, respectively) was found in the
teristics variable (basin slope) in the model improved Iowan Surface ecoregion, a region characterized by
the adjusted R2 value and suggests that greater relief thin glacial drift overlying permeable Devonian and
in watersheds contributes to a higher percentage of Silurian carbonate bedrock. The Des Moines Lobe
base flow in streams. This factor differs from basin (DML) ecoregion contained the second highest region
relief (BR) or relative relief (RR) in that it was deter- of Q and Qb (11.0 and 6.1 in or 279 mm and 155 mm).
mined as the average slope of all the 30-m cells in a This region is characterized by a low relief, recently
watershed, and not by large scale basin measure- glaciated landscape with poor natural surface
ments. As such, the BS factor points to more local drainage. However, it is noteworthy that the hydrolo-
topographic control on %Qb rather than the water- gy of the DML has been so extensively altered by arti-
shed scale BR and RR factors, which are often biased ficial drainage (tiles and ditches) for extensive row
by the presence of a topographic high in the basin crop production that it now produces more Q and Qb
that does not reflect the overall landscape relief. A per unit area than nearly any region in Iowa.
prime example in Iowa is a watershed located on the Lowest Q was noted in the Northwest Iowa Loess
recently glaciated Des Moines Lobe landform region Prairies and Loess Hills where average annual rain-
that contains a portion of a moraine. While the basin fall is least in Iowa. Lowest Qb in Iowa is sharply
relief would indicate substantial relief in the basin, defined in southern Iowa in the Central Irregular
the average basin slope would be low since the Plains where average Qb is 2.0 in (50.8 mm). This
drainage area is predominantly till plain. region is characterized by rolling hills of glacial till
and loess that contains significantly more grassland,
forest, and pasture than other Iowa regions. Although
Distribution of Iowa Q and Qb and Relation to Iowa precipitation is relatively high in this region, most is
Ecoregions directed as Qs, as the average base flow percentage is
approximately 20.4 percent based on average Q (9.8
The regional regression models for Q and Qb devel- in or 149 mm) and Qb (2.0 in or 50.8 mm).
oped with data from the initial 33 watersheds (Figure The average Qb estimated for the various ecoregion
1) were extended to other Iowa watersheds of varying areas may also be considered average ground water
size and statewide maps of Q and Qb were produced recharge values in these areas. Information related to
(Figure 3). For purposes of producing Figure 3, it was average ground water recharge may be useful for
assumed that the regression models developed from water supply operators, ground water modelers,
the 33 watersheds would also describe Q and Qb in watershed planners, and others interested in estimat-
other watersheds. Watershed sizes ranged from the ing how much ground water recharge occurs each
U.S. Geological Survey’s 8-digit hydrologic unit codes year.
(HUCs) (390 to 1,953 mi2 or 1,010 to 5,058 km2) to 12-
digit HUCs (15.6 to 62.5 mi2 or 40.4 to 161.9 km2) and
Prediction of Annual Q and Qb with MLR Models
all were combined in Figure 3 to provide a compre-
(Objective 2)
hensive statewide coverage of average streamflow and
base flow. Although a lack of stream gauges in smaller
watersheds prevented calibration of the regression The MLR models presented above combined a long
models across different watershed sizes, the data in term average rainfall variable with static basin prop-
Figure 3 are nonetheless informative and show sys- erties to predict long term Q and Qb. The process
tematic variations in Q and Qb across the state. essentially identified the constant with which to mod-
Patterns of average Q and Qb show remarkable ify long term precipitation data to obtain Q and Qb.
alignment with Iowa ecoregion boundaries (Figure 3). For example, based on the %SAND and %RC in the
Ecological regions have been defined as “regions of Boone River in north-central Iowa (84.9 percent and
relative homogeneity in ecological systems and rela- 24.5 percent, respectively), Equation (2) can be
tionships between organisms and their environments” reduced to:
(Griffiths et al., 1994), and in Iowa, ten regions have
been identified. Seven Level III subregions of the Q (in) in Boone River = 0.645(RAIN) - 10.17 (6)

JAWRA 1340 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

 17521688, 2005, 6, Downloaded from [Link] by Cochrane Thailand, Wiley Online Library on [01/08/2024]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ESTIMATION OF STREAMFLOW, BASE FLOW, AND NITRATE-NITROGEN LOADS IN IOWA USING MULTIPLE LINEAR REGRESSION MODELS

Figure 3. Spatial Distribution of Estimated Q and Qb in Iowa Watersheds. Ecoregion numbers correspond to:
40 – Central Irregular Plains; 47a – Northwest Iowa Loess Prairies; 47b – Des Moines Lobe; 47c – Iowan Surface;
47e – Loess Hills and Rolling Prairies; 47f – Southern Iowa Rolling Loess Prairies; and 52 – Paleozoic Plateau.

where the -10.17 term represents the basin specific rainfall totals, instead of 20-year rainfall averages,
number needed to modify the long term annual pre- combined with the static basin properties would pro-
cipitation muliplier in the Boone River watershed to vide a reasonable estimate of annual Q and Qb in
obtain long term Q. The purpose under Objective 2 watersheds.
was to determine whether substitution of annual

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1341 JAWRA

 17521688, 2005, 6, Downloaded from [Link] by Cochrane Thailand, Wiley Online Library on [01/08/2024]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SCHILLING AND WOLTER

TABLE 4. Summary of Average Streamflow (Q), Base Flow (Qb), and Nitrate Loss From Ecoregion Areas.

Mean
Mean Mean NO3-N
Number of Q Qb Load
Ecoregion General Description Watersheds (in) (in) (b/ac)

Central Irregular Irregular plains, open low hills, moderate loess over 82 9.8 2 3.6
Plains (40) loamy till

Northwest Iowa Irregular plains, moderate to thick loess over clay 58 7.8 4.4 14.8
Loess Prairie (47a) loam till

Des Moines Lobe (47b) Smooth to irregular plains, loamy till with no loess 179 11 6.1 20.7

Iowan Surface (47c) Irregular to smooth plains, thin loess over loamy till 114 11.8 7.7 20.0

Loess Hills and Open low hills, thick loess 54 9.4 5.3 14.7
Rolling Prairies (47e)

Southern Iowa Rolling Irregular plains to open low hills, moderate to thick 190 10.4 4.5 12.5
Loess Prairies (47f) moderate to thick loess over clay till

Loess Hills (47m) Steep hills, thick loess 3 7.0 3.8 6.7

Paleozoic Plateau (52) Open hill, irregular plains, thin loess and patches of 30 8.7 4.1 6.7
drift over bedrock

Results of Q and Qb for four Iowa watersheds, discharge and concentration. To estimate nitrate con-
Floyd River, Boone River, Turkey River, and Cedar centrations, the linear relation established between
Creek, representing four different landform regions of mean annual nitrate concentrations in Iowa streams
the state are presented in Figure 4. Over a 20-year and row crop land use was used (Schilling and Libra,
period from 1980 to 2000, the results indicate correla- 2000), where
tion between actual annual Q and Qb with predicted
annual Q and Qb. Correlation coefficients for Q Mean Annual NO3-N Conc. (mg/l) = 0.1077(%RC)
ranged from 0.79 to 0.89, whereas the correlation - 0.812 (7)
coefficients for Qb were slightly less (0.72 to 0.81).
Although only four examples are presented herein, The %RC variable was used in both Q and concentra-
the ranges of these correlation values are similar to tion predictions, but this is valid for several reasons.
those determined for other Iowa watersheds. In gen- Row crop land use affects both the hydrology of a
eral for both Q and Qb, the predicted results tended to region, as shown in the present study, as well as the
mimic the overall pattern of annual fluctuations but source of nitrate leaching to streams. The major non-
often to a dampened level, not quite reaching the point source of nitrate is agriculture, primarily the
extremes of wet or dry years. It should be noted that widespread use of nitrogen fertilizers, application of
the 20-year averages of measured and predicted livestock manure, legume fixation, and mineralization
annual values were close, ranging from 92 to 113 per- of soil nitrogen (Burkart and James, 1999). All of
cent for Q and 80 to 101 percent for Qb. Correlation these major nitrate source terms are concentrated on
coefficients were uniformly better for Q than Qb, row crop lands. Hence, using the percentage of row
probably due to several factors, including the effects crop to estimate mean annual concentrations in a
of antecendent moisture conditions from year to year. watershed apportions nitrate concentration to areas
where nitrate sources are greatest. The estimated
concentration produced from Equation (7) is probably
Prediction of Iowa N-Loads (Objective 3) lower than a true average value since it does not con-
sider how nitrate concentrations vary according to
The final objective in this study was to use the flow. For example, nitrate concentrations in the
regional regression model for Q to estimate potential Raccoon River in central Iowa averaged 10 mg/l
nitrate pollution loads in the state. While Q has when streamflow was in the upper 25 percent of flow
been estimated, pollution loads are the product of but 2.4 mg/l when streamflow was in the lowest

JAWRA 1342 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

 17521688, 2005, 6, Downloaded from [Link] by Cochrane Thailand, Wiley Online Library on [01/08/2024]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ESTIMATION OF STREAMFLOW, BASE FLOW, AND NITRATE-NITROGEN LOADS IN IOWA USING MULTIPLE LINEAR REGRESSION MODELS

Figure 4. Prediction of Annual Q and Qb Using the Regression Models.

Filled circle = measured value; open circle = estimated value.

25 percent (Schilling and Lutz, 2004).A truer repre- 20.7 and 20.0 lbs/ac (23.2 and 22.4 kg/ha) (Table 4).
sentation of mean annual nitrate concentrations Nitrate export averaged between 12.5 and 14.8 lbs/ac
would consider flow weighted values as opposed to (14.0 and 16.6 kg/ha) in three other ecoregions and
periodic grab samples used in development of Equa- between 6.7 and 7.0 lbs/ac (7.5 and 7.8 kg/ha) in the
tion (7). Thus, the estimated loads based on the esti- loess hills and Paleozoic Plateau regions. Unfortu-
mated nitrate concentration are likely lower than nately, few watersheds were located within the nar-
measured nitrate loss rates. row geographic range of the Loess Hills ecoregions to
Estimated nitrate export (in lbs/acre) ranged from better characterize average nitrate export. Lowest
essentially zero in several small southern Iowa water- average nitrate export was observed in southern Iowa
sheds, where row crop percentages are less than in the Central Irregular Plains region where nitrate
8 percent, to a high of 26.4 lbs/ac (29.6 kg/ha) in a export from 82 watersheds areas averaged 3.6 lb/ac
small southeast Iowa watershed (Figure 5). Nitrate (4.0 kg/ha) (Table 4).
loads were highest in the Des Moines Lobe and Iowan The estimated loads presented in this study appear
Surface ecoregions where average annual export was to be reasonable based on available long term

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1343 JAWRA

 17521688, 2005, 6, Downloaded from [Link] by Cochrane Thailand, Wiley Online Library on [01/08/2024]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SCHILLING AND WOLTER

monitoring data in the state. In the Raccoon River in that the regression model is based on long term pre-
west-central Iowa, draining a watershed partly in the cipitation averages, short term studies with above
Des Moines Lobe and Southern Iowa Rolling Loess average rainfall will be biased toward higher than
Prairie regions, the long term average nitrate loss average loads. However, it must be emphasized that
over a 28-year period was found to be 23.2 lbs/ac (26.0 the purpose of this study was not to necessarily pre-
kg/ha) (Schilling and Zhang, 2004). For the same dict annual nitrate losses in a given watershed
watershed area, the estimated nitrate loss averaged because measured losses vary considerably, but rather
17.2 lbs/ac (19.3 kg/ha), based on a weighted averaged to examine variations in average nitrate losses across
of watersheds located in the Des Moines Lobe region a spatial scale. Figure 5 illustrates that while nitrate
(average 19.2 lbs/ac or 21.5 kg/ha) and Southern Iowa losses are highly variable across the state of Iowa,
Rolling Prairie (average 13.6 lbs/ac or 15.2 kg/ha). regional patterns are evident, and some watersheds
Long term monitoring projects in two HUC 12 stand out as potentially contributing more nitrate
watersheds, the Walnut Creek watershed in Story loads to streams. These watersheds can then be sin-
County (Des Moines Lobe ecoregion) and Squaw gled out for more detailed monitoring and analysis in
Creek watershed in Jasper County (Southern Iowa any future TMDL determinations.
Rolling Loess Prairie) indicate reasonable agreement
in average nitrate loads. In Walnut Creek, a nine-year
average nitrate flux (16.7 lbs/ac or 18.7 kg/ha) (Tomer
et al., 2004) compared favorably to our regression CONCLUSIONS
model estimate (19.0 lbs/ac or 21.3 kg/ha). The esti-
mated load determined in this study (17.9 lbs/ac or Multiple linear regression was conducted to identi-
20.0 kg/ha) was lower than a five-year average flux in fy which of the 19 rainfall, land use, soil and geology,
Squaw Creek (25.4 lbs/ac or 28.4 kg/ha) (Schilling, and basin characteristics best explained 20-year aver-
2002) but within the range of measured annual export age Q, Qb, Qs, and %Qb in Iowa watersheds. The
(11.6 to 50.3 lbs/ac or 13.0 to 56.3 kg/ha). Considering results indicated that rainfall, row crop percentage,

Figure 5. Spatial Distribution of Estimated Nitrate Losses in Iowa Watersheds.

Ecoregion boundaries and nomenclature same as Figure 3.

JAWRA 1344 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

 17521688, 2005, 6, Downloaded from [Link] by Cochrane Thailand, Wiley Online Library on [01/08/2024]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ESTIMATION OF STREAMFLOW, BASE FLOW, AND NITRATE-NITROGEN LOADS IN IOWA USING MULTIPLE LINEAR REGRESSION MODELS

and several soil and geology variables could explain study have important implications for understanding
75.5 percent to 88.5 percent of the variability in the geomorphic and land use controls on streamflow and
four streamflow variables. Basin characteristics did base flow in Iowa watersheds and similar agriculture-
not improve the regression models significantly, dominated watersheds in the glaciated Midwest. An
except for the addition of basin slope to the %Qb understanding of these underlying controls will assist
regression equation. However, other investigators managers with identifying and targeting watershed
have observed basin characteristics, like total areas in need of NPS pollution reductions.
drainage area and stream length, to be important
variables in predicting volume discharge (cubic feet
per second) rather than unit discharge (in) considered LITERATURE CITED
in this study.
The linking of agricultural row crop land use with Burkart, M.R. and D.E. James, 1999. Agricultural-Nitrogen Contri-
soil texture (%SAND), permeability, and floodplain butions to Hypoxia in the Gulf of Mexico. Journal of Environ-
mental Quality 28:850-859.
variables to predict Q and Qb supports the general
Chapman, S.S., J.M Omernick, G.E. Griffith, W.A. Schroeder, T.A.
concept of “geology vegetation” groups proposed by Nigh, and T.F. Wilton, 2002. Ecoregions of Iowa and Missouri
Lacey and Grayson (1998) to explain hydrologic vari- (color poster with map, descriptive text, summary tables and
ability in Australian streams. In Iowa, streamflow photographs). U.S. Geological Survey, Reston, Virginia.
conditions are governed by key climate, geologic, and Eash, D.A., 1994. A Geographic Information System Procedure to
Quantify Drainage-Basin Characteristics. Water Resources Bul-
land use variables. In particular, base flow and base
letin 30(1):1-8.
flow percentage are largely governed by vegetation Giglierano, J.D., 1999. Iowa’s Statewide Land Cover Inventory.
and soil factors that contribute to greater ground Iowa Geology 1999, Iowa Geological Survey, Iowa City, Iowa.
water recharge in the watershed. In an agricultural Gregory, K.J. and D.E. Walling, 1968. The Variation of Drainage
state like Iowa, the dominant vegetation of row crop Density Within a Catchment. International Association of Sci-
ence Hydrology Bulletin 13:61-68.
clearly affects the basin scale hydrologic cycle by
Griffiths, G.E., [Link], T.F. Wilton and S.M. Pierson, 1994.
delivering more total runoff and base flow to streams Ecoregions and Subecoregions of Iowa: A Framework for Water
per unit area. Statewide patterns of average Q and Quality Assessment and Management. Journal of the Iowa
Qb show remarkable alignment with Iowa ecoregion Academy of Sciences 101:5-13.
boundaries and reinforce the close connection between Gustard, A., A. Bullock, and J.M. Dixon, 1992. Low Flow Estima-
tion in the United Kingdom. Institute of Hydrology, Report No.
hydrologic and ecologic systems. Two regions with
108, Oxfordshire, United Kingdom, 88 pp.
highest Q and Qb are both dominated by row crop but Gustard, A. and K.M. Irving, 1994. Classification of the Low Flow
contain entirely different hydrological plumbing sys- Response of European Soils. In: FRIEND: Flow Regimes From
tems, with one underlain by porous and fractured car- the International Experimental and Network Data, P. Seuna,
bonate bedrock (Iowan Surface), whereas the other A. Gustard, N.W. Arnell, and G.A. Cole (Editors). International
Association of Hydrological Sciences (IAHS) Publication No.
(Des Moines Lobe) is dominated by artificial drainage.
221, Oxfordshire, United Kingdom, pp. 113-117.
The region with the least amount of row crop (Central Hallberg, G.R., 1987. Nitrates in Ground Water in Iowa. In: Rural
Irregular Plain) was sharply defined by a very low Ground Water Contamination, F.M. D’Itri and L.G. Wolfson
annual base flow. (Editors). Lewis Publishers, Chelsea, Michigan, pp. 23-68.
Regional regression models developed for long term Helsel, D.R. and R.M. Hirsch, 2002. Statistical Methods in Water
Resources. Techniques of Water-Resources Investigations Book
average Q and Qb were adapted to annual rainfall
4, Chapter A3, U. S. Geological Survey. Available at [Link]
values and showed good correlation between mea- [Link]/pubs/twri/twri4a3/. Accessed in August 2004.
sured and predicted trends. Predicted Q and Qb tend- IGS (Iowa Geological Survey), 2000. Strahler Stream Order of
ed to mimic the overall pattern of annual fluctuations Streams in Iowa. Available at [Link]
but often to a dampened level. gis_library/IA_State/Hydrologic/Surface_Waters/stream_order.
zip. Accessed in July 2004
By combining the regression model for Q with an
IGS (Iowa Geological Survey), 2003. Soils of Iowa. Available at
estimate of mean annual nitrate concentration devel- [Link]
oped elsewhere, a map of potential nitrate loads for zip. Accessed in July 2004.
Iowa was produced. Nitrate losses estimated with the IGS (Iowa Geological Survey), 2004. Natural Resources Geographic
coupled flow and concentration models appear reason- Information Systems Library: Land Cover of the State of
able and well within normal ranges of nitrate losses Iowa in the Year 2002. Available at [Link]
nrgislibx/. Accessed in July 2004.
measured by others. While nitrate losses appeared ISU (Iowa State University), 2004. Iowa Soil and Land Use: Crop
highly variable across the state of Iowa, regional pat- and Land Use Tables and Charts. Iowa State University Agrono-
terns were evident, and again appear correlated to my Extension. Available at [Link]
ecoregion boundaries. Highest potential losses of soils/tables_charts.html. Accessed in August 2004.
nitrate were located in regions with highest Q and ISU (Iowa State University), 1996. Iowa Soil Properties and Inter-
pretations Database (ISPAID 6.0). Iowa State University Agri-
Qb, emphasizing the importance of “following the culture Extension Service, Ames, Iowa, 28 pp.
water” in any study involving loads. Results from this

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1345 JAWRA

 17521688, 2005, 6, Downloaded from [Link] by Cochrane Thailand, Wiley Online Library on [01/08/2024]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SCHILLING AND WOLTER

Kroll, C., J. Lutz, B. Allen, and R.M. Vogel, 2004. Developing a

Watershed Characteristics Database to Improve Low Stream-
flow Prediction. Journal of Hydrologic Engineering 9:116-125.
Lacey, G.C. and R.B. Grayson, 1998. Relating Baseflow to Catch-
ment Properties in South-Eastern Australia. Journal of Hydrol-
ogy 204:231-250.
Lara, O.G., 1986. Method for Estimating the Magnitude and Fre-
quency of Floods at Ungaged Sites on Unregulated Rural
Streams in Iowa. U.S. Geological Survey Water-Resources Inves-
tigation Report 87-4132, Iowa City, Iowa.
Minitab, 2000. Release 13. MINITAB Statistical Software, Minitab,
Inc. State College, Pennsylvania.
NRCS (Natural Resources Conservation Service), 2000. Iowa Pre-
cipitation Data/Maps. Available at [Link]
[Link]/products/datasets/climate/data/precipitation-state/
[Link]. Accessed in August 2004.
Prior, J.C., 1991. Landforms of Iowa. University of Iowa Press, Iowa
City, Iowa, 154 pp.
Schilling, K.E., 2002. Chemical Transport From Paired Agricultural
and Restored Prairie Watersheds. Journal of Environmental
Quality 31:1184-1193.
Schilling, K.E., 2005. Relation of Baseflow to Row Crop Intensity in
Iowa. Agriculture, Ecosystems and Environment 105:433-438.
Schilling, K.E. and R.D. Libra, 2000. The Relationship of Nitrate
Concentrations in Streams to Row Crop Land Use in Iowa. Jour-
nal of Environmental Quality 29:1846-1851.
Schilling, K.E. and R.D. Libra, 2003. Increased Baseflow in Iowa
Over the Second Half of the 20th Century. Journal of the Ameri-
can Water Resources Association 39(4):851-860.
Schilling, K.E. and D.S. Lutz, 2004. Relation of Nitrate Concentra-
tions and Baseflow in the Raccoon River, Iowa. Journal of the
American Water Resources Association. 40(4):889-900.
Schilling, K.E. and C.F. Wolter, 2001. Contribution of Baseflow to
Nonpoint Source Pollution Loads in an Agricultural Watershed.
Ground Water 39:49-58.
Schilling, K.E. and Y.K. Zhang, 2004. Contribution of Baseflow to
Nitrate-Nitrogen Export in a Large Agricultural Watershed,
USA. Journal of Hydrology 295:305-316.
Sloto, R.A. and M.Y Crouse, 1996. HYSEP: A Computer Program
for Streamflow Hydrograph Separation and Analysis. U.S. Geo-
logical Survey Water Resources Investigation Report 96-4040,
U.S. Geological Survey, Lemoyne, Pennsylvania.
Smakhtin, V.U., 2001. Low Flow Hydrology: A Review. Journal of
Hydrology 240:147-186.
Tomer, M.D., D.W. Meeks, D.B. Jaynes, and J.L. Hatfield, 2004.
Evaluation of Nitrate-Nitrogen Fluxes From a Tile-Drained
Watershed in Central Iowa. Journal of Environmental Quality
32:642-653.
USGS (U.S. Geological Survey), 1999. National Elevation Data
(feet). Available at [Link]
IA_State/Elevation/NED/ned_ft.zip. Accessed in July 2004.
USGS (U.S. Geological Survey), 2004. NWISWeb Data for the
Nation. Available at [Link] Accessed in Jan-
uary 2004.
Vogel, R.M. and C.N. Kroll, 1992. Regional Geohydrologic-Geomor-
phic Relationships for the Estimation of Low-Flow Statistics.
Water Resources Research 28:2451-2458.
Zecharias, Y.B. and W. Brutsaert, 1988. The Influence of Basin Mor-
phology on Groundwater Outflow. Water Resources Research
24:1645-1650.

JAWRA 1346 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION