Reduction of phosphorus (P) inputs to surface waters may decrease
eutrophication. Some researchers have proposed ﬁltering dissolved
P in runoﬀ with P-sorptive byproducts in structures placed
in hydrologically active areas with high soil P concentrations.
e objectives of this study were to construct and monitor a P
removal structure in a suburban watershed and test the ability of
empirically developed ﬂow-through equations to predict structure
performance. Steel slag was used as the P sorption material in the
P removal structure. Water samples were collected before and
after the structure using automatic samples and analyzed for total
dissolved P. During the ﬁrst 5 mo of structure operation, 25% of
all dissolved P was removed from rainfall and irrigation events.
Phosphorus was removed more eﬃciently during low ﬂow rate
irrigation events with a high retention time than during high ﬂow
rate rainfall events with a low retention time. e six largest ﬂow
events occurred during storm ﬂow and accounted for 75% of the P
entering the structure and 54% of the P removed by the structure.
Flow-through equations developed for predicting structure
performance produced reasonable estimates of structure “lifetime”
(16.8 mo). However, the equations overpredicted cumulative P
removal. is was likely due to diﬀerences in pH, total Ca and Fe,
and alkalinity between the slag used in the structure and the slag
used for model development. is suggests the need for an overall
model that can predict structure performance based on individual
Trapping Phosphorus in Runo with a Phosphorus Removal Structure
Chad J. Penn,* Joshua M. McGrath, Elliott Rounds, Garey Fox, and Derek Heeren
R (P) loading to surface
waters can help to prevent eutrophication. Previous
studies have suggested the use of certain industrial by-
products as P sorption materials (PSMs) for reducing P solu-
bility in high-P soils (Leader et al., 2008; Makris and Harris,
2006; Rhoton and Bigham, 2005). Although the addition of
PSMs to high-P soils has been shown to reduce water-soluble
P and therefore losses of dissolved P in runoﬀ (Gallimore et
al., 1999), such reductions in P solubility can be temporary
(Penn and Bryant, 2006). In addition, such an approach does
not truly remove P from the watershed; P pools within the soil
solid phase are simply shifted to less soluble forms.
A potential modiﬁcation to this approach is a P removal
structure. Such structures can be ﬁlled with PSMs and can be
strategically placed in “hot spots” or drainage ditches where
runoﬀ with elevated concentrations of dissolved P regularly
occurs (Penn et al., 2010). e P removal structure is designed
to intercept runoﬀ or subsurface drainage and channels ﬂow
through contained PSMs. After the PSMs become saturated
with P, they can be replaced with new PSMs; using this
approach, P can be eﬀectively removed from the watershed.
Some potential guidelines, theory, and approach for P removal
structure design are presented in Penn et al. (2010). Similarly,
previous studies have used various PSMs for removing P from
wastewaters (Koiv et al., 2010; Cucarella and Renman, 2009;
Wei et al., 2008) and subsurface drainage (McDowell et al.,
2008). A material that has shown tremendous promise as a
PSM in column studies is steel slag (Drizo et al., 2008, 2006,
2002), which is a by-product of the steel industry.
In a previous study, Penn and McGrath (2011) constructed
a pilot scale pond ﬁlter that used electric arc furnace steel slag
as the PSM. e authors developed empirical equations based
on laboratory ﬂow-through experiments that predicted struc-
ture performance as a function of retention time (RT) (i.e.,
the time required for one pore volume to pass through the
structure) and inﬂow P concentration. At a RT of 10 min, the
pond ﬁlter removed 34% of the all P pumped into it (172 mg
kg-1 of PSM) at the point of P saturation (i.e., the point at
which P was no longer removed from passing water). e ﬂow-
through equations reasonably predicted structure performance
(P removal and longevity), whereas the Langmuir equation
Abbreviations: DI, deionized; ICP–AES, inductively coupled plasma atomic
emission spectroscopy; PSM, phosphorus-sorbing material; PVC, polyvinyl
chloride; RhWT, rhodamine; RT, retention time.
Dep. of Plant and Soil Science, Oklahoma State Univ., 367 Agricultural Hall,
Stillwater, OK, 47078-1020. Assigned to Associate Editor Gerwin F. Koopmans.
Copyright © 2011 by the American Society of Agronomy, Crop Science Society
of America, and Soil Science Society of America. All rights reserved. No part of
this periodical may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopying, recording, or any information
storage and retrieval system, without permission in writing from the publisher.
J. Environ. Qual.
Posted online 1 Nov. 2011.
Received 14 Feb. 2011.
*Corresponding author (firstname.lastname@example.org).
© ASA, CSSA, SSSA
5585 Guilford Rd., Madison, WI 53711 USA
Journal of Environmental Quality SPECIAL SUBMISSIONS
Journal of Environmental Quality • Volume 41 • X–X 2012
developed from a batch isotherm experiment with the same
PSM material failed.
Other studies have shown potential for the development
of P removal structures. Penn et al. (2007) constructed a P
removal structure in a drainage ditch located on the Eastern
Shore of Maryland. is structure was ﬁlled with 226 kg of
acid mine drainage residual, and the PSM was able to remove
99% of the P, Zn, and Cu that ﬂowed into it during a 24-h
rainfall event that produced 30 cm of precipitation. However,
the structure soon thereafter failed as a result of ﬂow becom-
ing restricted through it (i.e., clogging). Agrawal et al. (2011)
tested a cartridge ﬁltration system on a golf course green sub-
surface drainage system for removing P and several pesticides
using a mixture of slag, zeolite, cement kiln dust, silica sand,
and coconut shell–activated carbon. Although the system was
eﬀective for removing certain pesticides, it was ineﬀective at
removing P, likely due to the small amount of slag used in the
ﬁltration system (3.5 L).
ere are no published studies on monitoring of a P
removal structure. erefore, the objectives of this study were
to construct and monitor a P removal structure in a subur-
ban watershed and to test the ability of previously constructed
ﬂow-through equations for predicting structure performance.
Materials and Methods
e P removal structure was placed at the outlet of a 320-ha
suburban watershed in Stillwater, Oklahoma. e watershed
land use consisted of approximately 35, 50, and 15% residen-
tal, undeveloped, and gof course, respectively. Two irrigated
golf greens were located within 130 to 150 m from the struc-
ture. e greens were regularly irrigated by golf course per-
sonnel as necessary, and this irrigation produced runoﬀ that
reached the P removal structure. e structure was located
in a drainage ditch immediately on the downstream side of a
drainage culvert (Fig. 1) where all water exited the watershed
via a concrete trapezoidal bar ditch maintained by the city
of Stillwater. e bar ditch drained directly into Stillwater
Creek. Some runoﬀ entered the structure by ﬂowing along
the side of the culvert into the structure inlet (Fig. 1).
Structure Construction and Runo Sampling
e P removal structure was 2.4 m wide × 3 m long × 0.2 m
deep and was constructed using 0.63-cm-thick carbon steel with
all joints welded to be water tight. e structure was welded in
situ along with two 3-m steel support pipes (5 cm diameter).
e bottom of the structure was set to a 3% slope toward the
outlet. irteen inlet pipes (5 cm diameter) were welded into
the front plate of the structure, and then each pipe was adapted
to polyvinyl chloride (PVC) pipe of the same diameter inside the
structure. Each PVC pipe was 2.3 m long and perforated (four
rows of 0.635-mm diameter holes at 5 cm apart) to evenly dis-
tribute inﬂow water throughout the surface of the structure. e
perforated distribution manifold is not visible in Fig. 1 because
the pipes are buried immediately below the surface. A 10-cm-
diameter steel drainage pipe was welded at the bottom center of
the downstream side of the structure; this pipe was adapted to a
15.2-cm-diameter PVC pipe ﬁtted with a shutoﬀ valve. All steel
was treated with two coats of primer and paint.
Two Isco 6712 (Teledyne Isco Inc., Lincoln, NE) automatic
samplers were housed on site in a small plastic building to take
runoﬀ samples at the structure inlet and outlet (drainage pipe)
during ﬂow events. In addition, the automatic sampler for the
outﬂow side of the structure was ﬁtted with an Isco 730 ﬂow
module (“bubbler”), which was connected to a 15.2-cm-ﬂow
oriﬁce insert placed in the structure drainage pipe (outﬂow).
e 730 ﬂow module was programmed to take a ﬂow rate mea-
surement every minute. e automatic sampler for the outﬂow
water was programmed as the “primary” and began sampling
when ﬂow was detected; the inlet sampler was programmed as
the “slave” to the outﬂow sampler and therefore was triggered to
sample at the same time as the outﬂow sampler. Discrete (not
composited) samples (800 mL) were taken using two programs;
from 0 to 34.5 L min-1 samples were taken every 30 min, and
at ﬂow rates >34.5 L min-1 samples were taken every 45 min.
Regarding potential “overﬂow” runoﬀ events, an Isco 2112
ultrasonic probe was ﬁtted near the downstream side of the
structure to monitor the depth of water on top of the structure.
e Isco 2112 could provide the ﬂow rate of untreated overﬂow
water during events that exceeded the capacity of the structure.
erefore, outlet ﬂow volume plus overﬂow volume equals total
ditch ﬂow volume.
Electric arc furnace steel slag was obtained from a steel mill
in Ft. Smith, Arkansas (Tube City IMS). Slag was sieved at
a nearby gravel quarry to achieve a size of 6.35 to 11 mm in
diameter. Previous experiments showed that the nonsieved slag
had a limited saturated hydraulic conductivity (Penn et al.,
2011). Approximately 2712 kg of the sieved slag was placed in
the P removal structure on 10 July 2010.
A rhodamine WT (RhWT) dye test was conducted to quan-
tify hydraulic RT in the structure. A constant water ﬂow rate
was discharged into a pool of water at the inlet of the struc-
ture (Fig. 1) for approximately 1 h to achieve steady state
ﬂow before initiating the dye test. e dye was injected into
the inﬂow solution and monitored in the inﬂow and outﬂow
Fig. 1. Picture of the phosphorus (P) removal structure with runo
inlets, drain for treated water, and overow weir. The P sorption mate-
rial in the structure is 2712 kg of 6.3- to 11-mm-diameter steel slag.
over time. e dye test was simulated using CXTFIT (ver-
sion 2.1) (Toride et al., 1999), a model used extensively for
solving the one-dimensional convective–dispersion equation
for solute transport through soils (e.g., Baumann et al., 2002;
Lee et al., 2002). Fate and transport parameters in the model,
such as pore velocity, hydrodynamic dispersion, and retar-
dation coeﬃcient, were optimized to the observed RhWT
concentrations. is process is also known as “inverse estima-
tion” of model parameters, as opposed to forward modeling,
where parameters are input and concentrations are predicted.
From these fate and transport parameters, various character-
istics of the ﬂow and contaminant transport system can be
measured, such as the RT and Peclet number. Physical and
chemical equilibrium of RhWT was assumed. e input
boundary condition for the dye was modeled in CXTFIT
as multiple pulse inputs based on measured inﬂow concen-
trations. CXTFIT used a nonlinear least-squares parameter
optimization method to derive the dye transport parameters
(i.e., velocity and dispersion coeﬃcient) that best predicted
the outﬂow RhWT concentrations. e inversely estimated
velocity from CXTFIT was used to estimate the average RT
of the dye in the structure.
Analysis of Water Samples and Slag
All water samples were collected within 12 h of a runoﬀ
event, ﬁltered through a 0.45-µm membrane, and refriger-
ated. Samples were analyzed within 3 d for P, copper (Cu),
zinc (Zn), chromium (Cr), and boron (B) by inductively cou-
pled plasma–atomic emission spectroscopy (ICP–AES). A pH
probe was used to measure pH in all samples. Alkalinity was
determined by automatic titration (TitriLab 865; Radiometer
Analytical, Villeurbanne Cedex, France) to pH 4.5.
All analyses of steel slag used in the P removal structure
were conducted in triplicate. Slag pH was determined with a
pH meter using a solid/deionized (DI) water ratio of 1:5 (w/v).
Alkalinity was determined as previously described using 2 g
of material suspended in 20 mL of DI water. Slag was ground
before analysis of total elements by the EPA 3051 nitric acid
digestion method (USEPA, 1997). Digestion solutions were
analyzed for Ca, Mg, S, Fe, and Al by ICP–AES. Samples were
also extracted with DI water at a 1:10 (w/v) solid/solution ratio
for 1 h, followed by ﬁltration with a 0.45-µm ﬁlter and analysis
for Ca, Mg, S, Fe, and Al by ICP–AES.
A standard batch isotherm was conducted for the slag using
2 g of sample and 16 h equilibration (shaking) in 30-mL
solutions of 0, 1, 10, 25, 50, and 100 mg P L-1. Phosphorus
solutions were made using KH2PO4, and the matrix solution
consisted of 5.6, 132, 110, 10, and 17 mg L-1 of Mg, Ca, S,
Na, and K, respectively, adjusted to a pH of 7. Reagent-grade
magnesium sulfate, calcium sulfate, sodium chloride, and
potassium chloride were used to make the matrix. is matrix
was chosen because it was found to be representative of agricul-
tural runoﬀ measured in a previous study (Penn et al., 2007).
After equilibration, solutions were centrifuged for 15 min and
ﬁltered through a 0.45-mm ﬁlter before P analysis by ICP–AES.
Phosphorus sorption was quantiﬁed by the diﬀerence
between P concentrations added and the ﬁnal equilibrated con-
centrations. ese values were applied to a nonlinear Langmuir
using the following equation:
where S is the sorbed P concentration (mg kg-1), Smax is the
maximum sorption capacity of the soil (mg kg-1), K is the
Langmuir binding strength coeﬃcient (L mg-1), and C is the
equilibrium concentration (mg L-1). e best ﬁt model param-
eters for the nonlinear equation were obtained by ﬁnding the
combinations of parameters that provided the best ﬁt to the
observed data. is was done by using an Excel spreadsheet
as prepared and described by Bolster and Hornberger (2007).
is program was designed to provide K and Smax values in
addition to the “goodness-of-ﬁt” indicator, model eﬃciency
(E). An E value of 1 indicates a perfect ﬁt of the data, and E <
0 indicates that taking the average of all measured P sorption
values in the isotherm would give a better prediction than the
model (Bolster and Hornberger, 2007).
Flow and sampling data were synchronized with Flow Link
software (Teledyne Isco Inc., Lincoln, NE) when downloaded
directly from the automatic samplers. Because ﬂow rate mea-
surements were taken every minute, the discrete runoﬀ volume
produced at any given minute can be determined by:
Discrete runoﬀ volume = ﬂow rate * 1 
where discrete runoﬀ volume is expressed in liters and ﬂow
rate in L min-1. Discrete runoﬀ volume was calculated at every
minute for each ﬂow event. erefore, the total runoﬀ volume
produced for a given time period could be determined by
the sum of all discrete runoﬀ volumes over that time period.
Weighted average ﬂow rate (L min-1) was calculated as:
total runoff volume
Weighted average flow rate
total runoff time
where total runoﬀ volume and time are in units of liters and
minutes, respectively. Phosphorus loading to the structure
between each sampling point was calculated by integrating P
concentrations with respect to ﬂow volume. e sum of all P
loads for each sampling point interval represents the total P
load for an event. is value is used to calculate ﬂow-weighted
P concentrations (mg L-1):
Flow-weighted P concentration
total flow volume
where P load and total volume are in units of milligrams and
liters, respectively. After P loads were determined for inﬂow
and outﬂow (treated) water, the P removal (mg) could be cal-
culated as a mass balance:
P removed = inlet P load – outﬂow P load 
where inlet and outflow P load are expressed as milligrams.
Retention time (in minutes) of the structure at different
flow rates was also estimated as described in Penn and
total structure pore space
flow rate at outlet
Journal of Environmental Quality • Volume 41 • X–X 2012
where total structure pore space and ﬂow rate at outlet are in
units of liters and L min-1, respectively. Total pore space (574 L)
was calculated based the total mass of material (2712 kg), bulk
density (1.8 g cm3), and porosity (38%).
Prediction of Field Results Using an Empirical Model
A series of empirical ﬂow-through equations developed by Penn
and McGrath (2011) was used to compare ﬁeld results of the
P removal structure with the predicted amount of P removed.
Although details of the general use of these empirical equa-
tions appear in a companion paper (Stoner et al., 2012), we
provide a brief description here. e following equations were
originally developed by Penn and McGrath (2011) to predict
the amount of discrete P removal (% P removal) with P loading
to sieved slag (x in mg P kg-1) using an exponential equation:
Discrete P removal (%) = bemx 
where b is the Y intercept and m is the slope coeﬃcient for this
relationship. Because this is an exponential decay equation, m
is always negative. e following equations (signiﬁcant at P <
0.01; R2 = 0.68 and 0.48 for Eq.  and , respectively) are
used to estimate the b and m parameters for Eq.  as a func-
tion of RT and inﬂow P concentration (Penn and McGrath,
log-m = (0.08506RT) - (0.07416Cin) - 2.53493 
log b = (0.06541RT) - (0.00864Cin) + 1.60631 
where Cin is the inﬂow P concentration (mg L-1). As described
in greater detail in Stoner et al. (2012) and Penn and McGrath
(2011), these equations were developed from a series of labora-
tory ﬂow-through cell experiments in which a known mass of
slag was exposed to a ﬂowing P solution at ﬁve diﬀerent RTs
and ﬁve diﬀerent inﬂow P concentrations. When parameters
m and b are inserted into Eq. , the result is a predicted P
removal curve speciﬁc to the inﬂow P concentration and RT
conditions that were input into Eq.  and . Integration
of the predicted P removal curve (Eq. ) yields a prediction
of cumulative P removal (%) at any given level of P added (x;
Cumulative P removed
Phosphorus removal approaches zero (1%) as described by the
equation for the predicted P removal curve (Eq. ) when the P
inﬂow concentration ≈ P outﬂow concentration (i.e., the point
at which the PSM is “spent”). Insertion of 1% for cumulative P
removed into Eq.  and subsequent rearrangement to solve
for x results in an estimate of the maximum amount of P that
can be delivered to the P removal structure before the PSM is
spent. Such a rearrangement results in the following equation:
Maximum P added b
Insertion of the maximum amount of P that can be added to
the P removal structure as determined from Eq.  into Eq.
 results in the total amount of P predicted to be removed by
the PSM under the conditions (i.e., RT and inﬂow P concen-
tration) used for the ﬂow-through equations (Eq.  and )
used to produce the predicted P removal curve.
Results and Discussion
Phosphorus Removal Structure: Flow
Results from the dye test indicated that when a ﬂow rate of
57.1 L min-1 was applied to the structure, the average RT was
9.3 min as estimated by CXTFIT (R2 = 0.97 between measured
and predicted dye outﬂow concentrations). is RT is similar
to the calculated value of 10 min estimated by Eq. .
During the 5-mo period in which all runoﬀ was monitored,
there were 54 total runoﬀ events. Twenty of the events were
rainfall, and 34 were due to irrigation of nearby golf course
greens (Table 1). Over that time period, the rainfall totaled
24.6 cm; the largest rainfall event was 4 cm on 8 Sept. 2010.
e P removal structure was able to treat all water delivered to
it, as evidenced by the fact that no water crested the overﬂow
weir, which was continuously monitored with an ultrasonic
probe. During the largest rainfall event, the maximum ﬂow
rate through the structure was 506 L min-1.
As expected, rainfall events produced higher ﬂow rates
through the structure than irrigation events from nearby golf
greens, which translated into a lower average RT for the rain-
fall runoﬀ events (Table 1). All runoﬀ samples were analyzed
for total dissolved P, and several random irrigation and storm
runoﬀ samples were analyzed for dissolved reactive P (i.e.,
orthophosphate). Because the entire area immediately drain-
ing into the structure was well covered with grass, there was no
sediment in the samples, and thus >90% of the total dissolved
P was orthophosphate. e overall ﬂow-weighted average total
dissolved P concentration in runoﬀ delivered to the P removal
structure (0.50 mg L-1) is comparable to other studies, includ-
ing those conducted on agricultural land. Harmel et al. (2004)
showed that several agricultural subwatersheds consisting of
cultivated crops or pasture that received 0 to 358 kg P ha-1 yr-1
Table 1. Summary of the suburban phosphorus removal structure performance over the rst 5 mo of operation.
Rainfall runo events Irrigation runo events All runo events
Number of runo events 20 34 54
Maximum ow rate, L min-1506 47 506
Weighted average ow rate, L min-130.3 11.5 29.8
Weighted average retention time, min 18.9 50 19.3
Maximum runo P concentration, mg L-11.61 0.97 1.61
Flow-weighted runo P concentration, mg L-10.59 0.44 0.50
Total P input to structure, mg kg-192.1 10.7 102.8
Total P removed by structure, mg kg-119.3 6.6 25.9
produced average dissolved P concentrations of 0.09
to 2.29 mg L-1. Among 35 agricultural catchments
monitored over 4 yr in Ireland, runoﬀ-dissolved
P concentrations ranged from 0.01 to 0.70 mg L-1
(Daly et al., 2002). A golf course in Texas produced
an average dissolved P concentration of 0.13 mg L-1
over 5 yr (King et al., 2007).
Figure 2 shows hydrographs and corresponding
inﬂow total dissolved P concentrations for typical
runoﬀ events from rainfall and irrigation. Not only
did rainfall runoﬀ events produce higher dissolved P
concentrations than irrigation runoﬀ events (Table
1), but rainfall runoﬀ events also tended to produce
increasing P concentrations with ﬂow rate into the
P removal structure. is suggests that hydrologi-
cal connectivity increased among certain portions of
the watershed as soils became saturated with mois-
ture and runoﬀ increased, allowing runoﬀ from these
“variable source” areas (Sharpley et al., 2008) in the
watershed to reach the outlet, which is the ditch P
removal structure. Similarly, Pionke et al. (1999)
found that dissolved P concentrations delivered from
an agricultural watershed increased with ﬂow rate. In
our case, we speculate that high-P soils contribute P
to the structure only during large events when they
become “connected” and such runoﬀ is able to reach
the outlet. Because the irrigation events that occurred
throughout the monitoring period were from the
same location, runoﬀ produced from such events
typically displayed relatively steady runoﬀ P concen-
trations delivered to the structure between 0.3 and
0.5 mg L-1 (Fig. 2).
Phosphorus Removal Structure:
e sum of total dissolved P delivered to the structure over
the 5-mo period was 0.282 kg or 0.0047 kg ha-1; 88% of
this P delivery occurred during rainfall induced runoﬀ events
(Table 1). Among all dissolved P transported in runoﬀ to the
P removal structure, 75% of this was delivered during the six
largest rainfall events. Various authors have suggested that
large rainfall events export the majority of P from watersheds
(Sharpley et al., 2008; Udawatta et al., 2004; Pionke et al.,
1999; Pionke et al., 1997). For example, Pionke et al. (1997)
found that 70% of annual dissolved P loads were exported by
the seven largest storms.
During the 5 mo of monitoring, the P removal structure
sorbed 25.9 mg P kg-1 slag, which was 25.2% of the total dis-
solved P delivered to it (Table 1). Of the 25.9 mg P kg-1 sorbed,
approximately 75 and 25% occurred during rainfall and irriga-
tion runoﬀ events, respectively. Phosphorus transported during
irrigation runoﬀ events was more eﬃciently removed by the
structure compared with rainfall runoﬀ events (i.e., 62 versus
21% P removal for irrigation and rainfall events, respectively)
(Table 1). e diﬀerence in P removal eﬃciency among rain-
fall and irrigation events is likely due to the fact that rainfall
runoﬀ events resulted in higher P concentrations and ﬂow
rates. Higher structure ﬂow rates during rainfall runoﬀ events
translated into a RT that was more than two times less than
irrigation events (Table 1). Regarding the impact of ﬂow rate
and RT on P removal by the structure, P removal on an event
basis was negatively correlated to the weighted average event
ﬂow rate (Fig. 3). Similarly, in a previous study (McDowell et
al., 2008) involving slag placed in subsurface drainage pipes, it
was noted that larger events resulted in less contact time with
the slag and lesser diﬀerences in dissolved P concentrations
relative to control drains.
Although the weighted average RT for all rainfall runoﬀ
events was 18.9 min, the RT for the six largest rainfall events
that delivered 75% of the P to the P removal structure was only
8.9 min. In addition, 54% of all the P removed by the structure
(14.1 mg kg-1) occurred over these six largest rainfall events.
Predicting Lifetime and Performance of the Structure
A predicted P removal curve estimated by the equations devel-
oped in Penn and McGrath (2011) for the electric arc furnace
steel slag is shown in Fig. 4. is curve (Eq. ) describes the
eﬀect of P loading to the PSM on discrete P removal. is curve
was produced by estimating its Y intercept (b) and its slope
coeﬃcient (m) with Eq.  and  in which RT and P inﬂow
concentration are used as inputs. For the RT of the runoﬀ in
the P removal structure, we used 8.9 min (i.e., the RT for the
Fig. 2. Typical hydrograph and corresponding inow total dissolved phosphorus
(P) concentrations to the ditch P removal structure from a rainfall-induced (a) and
irrigation-induced (b) runo event. The 3.73-cm rainfall/runo event shown in (a)
occurred on 17 Aug. 2010, and the irrigation/runo event occurred on 3 Aug. 2010.
Journal of Environmental Quality • Volume 41 • X–X 2012
six largest rainfall events that delivered 75% of the P to the P
removal structure), whereas the average ﬂow-weighted P inﬂow
concentration was set at 0.74 mg L-1. e predicted P removal
curve can be used to estimate the potential “lifetime” of the P
removal structure. When discrete P removal approaches nearly
zero (i.e., 1%), then the slag is eﬀectively “spent” and needs to
be replaced with fresh PSM because the P inﬂow concentra-
tion will nearly equal the outﬂow concentration. e structure
“lifetime” can be predicted using an estimate of P loading to
the structure per unit time and the predicted maximum P load-
ing to the P removal structure at the point in which the PSM
is “spent” (Eq. ). Using predicted values of the Y inter-
cept (b) and the slope coeﬃcient (m) from the ﬂow-through
equations (see above), a maximum cumulative loading of the
P removal structure amounting to 345 mg kg-1 was calculated
using Eq. . Based on the current P loading rate of the P
removal structure (i.e., 20.5 mg kg-1 mo-1), this would cor-
respond to a potential lifetime of 16.8 mo. e measured P
removal curve that was ﬁtted to the ﬁeld data of the
actual discrete P removal and P loading of the P removal
structure is shown in Fig. 4. Using the ﬁtted values of
the Y intercept (b) and the slope coeﬃcient (m), a maxi-
mum cumulative loading of 316 mg kg-1 was estimated
with Eq. , which corresponds to a structure lifetime
of 15.4 mo. us, the lifetime prediction of 16.8 mo
diﬀers by a factor of only 1.09 of the projected lifetime
using current structure performance data. In practice,
one may be inclined to remove the slag material before
P saturation if environmental thresholds such as total
maximum daily loads are exceeded. is estimate of ﬁlter
lifetime does not take into account processes of sorbed
P on the slag changing forms and allowing for more P
sorption sites to become available, as described in Drizo
et al. (2008). Apparently, such a factor did not have a sig-
niﬁcant impact on predicting ﬁlter lifetime due to near
agreement (16.8 vs. 15.4 mo). However, a slow P sorp-
tion mechanism as described by Drizo et al. (2008) that
was active would result in an underestimation of ﬁlter
lifetime by the predicted P removal curve. e steel slag used
in this study diﬀered from that of Drizo et al. (2008) in that it
was sieved to exclude ﬁne particles.
e predicted P removal curve shown in Fig. 4 can be inte-
grated to estimate the cumulative amount of P that the struc-
ture will remove as a function of P added (Eq. ). Figure 5
shows the predicted cumulative amount of P removed by the P
removal structure as a function of P loading. For comparison,
the measured values of the cumulative amount of P removed
from runoﬀ as a function of P loading of the P removal structure
are shown. e predicted cumulative P removal compared with
the measured values showed that the ﬂow-through equations
used to produce the predicted P removal curve overestimated
P removal. For example, after 5 mo and a total P input of 103
mg kg-1 to the P removal structure, the integrated predicted P
removal curve estimated 79 mg kg-1 of P sorption, whereas the
actual measured P sorption was 25.9 mg kg-1 (Table 1).
At the point of P saturation when the PSM is “spent,” the
integrated predicted P removal curve estimated a cumulative
removal of 101 mg P kg-1, or 28% of the total P added to
the structure. is estimated value was obtained from the pre-
dicted P removal curve (Fig. 4), which was produced using Eq.
[7–11] with an input of 8.9 min RT and 0.74 mg L-1 inﬂow
(i.e., the conditions of the six largest rainfall events that deliv-
ered 75% of the P). Speciﬁcally, ﬂow-through Eq.  and 
predicted the P removal curve parameters (b and m) for Eq.
; the resulting predicted design curve (Fig. 4) was integrated
(Eq. ) (Fig. 5), which produced an estimate of maximum P
removal under the conditions of the design curve (i.e., inﬂow P
concentration and RT).
Apparently, the empirical ﬂow-through equations were
able to predict that P would be removed from runoﬀ by the
P removal structure as the P loading increased, but not to
the correct degree in which it was occurring. is is likely
due to the fact that the equations were unable to accurately
predict the Y intercept (b) of the design curve (via Eq. )
(Fig. 4). e maximum amount of P projected to be removed
by the structure (i.e., 0.065 g kg-1 determined from integra-
tion of the curve ﬁtted to measured ﬁeld data in Fig. 4) is low
Fig. 3. Phosphorus (P) removal eciency presented per event as impacted by
the ow rate of runo water passing through the ditch P removal structure.
*Signicant at the 0.05 probability level.
Fig. 4. Discrete phosphorus (P) removal as a function of cumula-
tive P added to the ditch runo P removal structure. Predicted P
removal (dashed line) estimated based on average retention time
and P concentration of the six largest rainfall events that delivered
75% of runo P load (average weighted retention time, 8.9 min; total
dissolved P concentration, 0.74 mg L−1) using Eq. [7–10]. Measured
discrete P removal (open circles and solid line) calculated on a per-
event basis. Error bars indicate a 95% condence interval for the
predicted P removal curve. *Signicant at the 0.05 probability level.
compared with other studies that have investigated the use
of electric arc furnace steel slag for P sorption (Drizo et al.,
2006; Drizo et al., 2002). For example, Drizo et al. (2002)
achieved 1.35 to 2.35 g P removed kg-1; however, their study
used a much higher RT (~8 h) compared with the RT of the
runoﬀ in the P removal structure in our study. In addition,
the large particle size fraction used in our study (i.e., 6.35–11
mm) compared with previous studies (Kostura et al., 2005;
Drizo et al., 2002) is not nearly as sorptive compared with the
ﬁner slag fraction (Stoner et al., 2012). However, the beneﬁt
of the large size fraction is higher hydraulic conductivity of
the structure, which reduces the “footprint” or area of the P
removal structure and allows more water to be treated com-
pared with a ﬁner-sized fraction.
Equations  and , which were used to estimate the Y
intercept (b) and the slope coeﬃcient (m) of the predicted
design curve in Fig. 4, were developed using slag with the same
size fraction collected from the same steel mill as slag used in
the P removal structure but was collected at a diﬀerent time
(about 8 mo apart). In other words, cumulative P removal
predictions from equations developed by Penn and McGrath
(2011) are speciﬁc to their particular slag material, and any
variation in slag properties would likely result in deviation
from the predictions. is could explain why integration of
the predicted P removal curve with sample-speciﬁc parameters
indicated in Eq.  and  from Penn and McGrath (2011)
overpredicted cumulative P removal as compared with mea-
sured values (Fig. 5). For example, the slag placed in the P
removal structure contained less alkalinity and less total Ca and
Fe, and had a lower pH compared with the slag used to develop
the ﬂow through equations of Penn and McGrath (2011). Slag
pH and alkalinity are integral to Ca phosphate precipitation
(Bowden et al., 2009; Kostura et al., 2005). e role of Ca
and Fe in P sorption by industrial by-products has been well
documented (Penn et al., 2011; Leader et al., 2008). Lesser
amounts of Ca and Fe would result in less Ca phosphate pre-
cipitation and P binding by Fe oxy/hydroxide minerals. e
Langmuir K value was also much less for the slag sample used
in this study compared with that used for development of ﬂow-
through equations (i.e., 0.00126 vs. 2.43 L mg-1, respectively,
from Penn and McGrath, 2011).
Other Water Quality Parameters
Average pH of inﬂow and outﬂow treated water was 7.7 and
9.2 (SE, 0.04 and 0.08, respectively). e increase in pH of the
treated water was expected due to the elevated pH of the PSM
tested in the laboratory (i.e., 9.4) (Table 2). However, alkalinity
of the treated water was similar to inﬂow water; average inﬂow
and outﬂow alkalinity was 77 and 81 mg CaCO3 L-1 (SE, 21 and
23, respectively). A minimum alkalinity of 20 mg L-1 is required
for ecosystems, and an alkalinity up to 400 mg L-1 has no impact
on human health (USEPA, 1986).
For all inﬂow and treated water, Zn, Cu, Cr, and Mn con-
centrations were all below detection limits (i.e., 0.01 mg L-1).
Average B concentrations were similar among inﬂow and out-
ﬂow treated waters (i.e., 0.14 and 0.15 mg L-1; SE, 0.003 and
0.005, respectively). However, these B concentrations are not
considered hazardous to aquatic life or B-sensitive agricultural
crops (USEPA, 1986).
During the ﬁrst 5 mo of operation, the P removal structure
trapped 25% of runoﬀ dissolved P. is could be improved
by using the smaller particle size fraction of the slag, which is
much more sorptive than the large fraction used in this study
(Stoner et al., 2012). However, the smaller-sized fraction
would reduce the hydraulic conductivity, thereby reducing
the amount of water that can be treated during a large runoﬀ
event. Alternatively, the ﬁlter dimensions could be adjusted to
allow for a higher RT. e ﬂow-through equations presented
in Penn and McGrath (2011) predicted a lifetime of 16.8 mo,
which is similar to the projected lifetime of 15.4 mo based on
current measurements. However, the ﬂow-through equations
overestimated current P removal (79 vs. 26 mg P kg-1) by the
P removal structure. Diﬀerences in P removal between pre-
dictions and measurements were likely a result of variability
in slag chemical properties among slag used in the P removal
structure and for development of ﬂow-through equations. is
emphasizes the need to develop a “universal” ﬂow-through
Fig. 5. Cumulative phosphorus (P) removal by the ditch P removal
structure over a 5-mo period as measured and predicted (dashed
line) using a series of ow-through based equations (Eq. [7–10]).
Predicted P removed estimated by integration of the curve pre-
sented in Fig. 4 using Eq. . Error bars indicate 95% condence
interval for the predicted P removal based on the standard error for
each model coecient.
Table 2. Chemical properties of the steel slag used in the suburban phosphorus removal structure.
Smax† K pH Alkalinity Total‡ Water soluble
Ca Mg S Fe Al Ca Mg S Fe Al
mg kg–1 L mg–1 mg CaCO3 kg–1 —————————————————————— mg kg–1 ——————————————————————
† Smax is the maximum sorption capacity of the soil. Langmuir isotherm Smax and K values were estimated using Eq. .
‡ Determined by EPA3051 digestion method.
§ Values in parentheses indicate standard error.
Journal of Environmental Quality • Volume 41 • X–X 2012
model that takes into account chemical characterization of
sorption materials in addition to RT and P concentrations.
Because 75% of all P delivered to the structure occurred over
the six largest rainfall events, P removal structures should be
designed for handling these events to maximize P removal.
Compared with other best management practices, poultry
litter transport programs and limitation of fertilizer P applica-
tions only prevent soil P from increasing further. is technol-
ogy can help to prevent P losses to surface waters in the short
term. In addition, the structure provides an easily quantiﬁed P
removal that not only can be removed from the watershed but
also may be useful to nutrient trading programs that are anal-
ogous to current carbon credit exchange programs (USEPA,
2001). Such programs apply a monetary value to P discharged
or transported from a site or prevented from being transported.
e authors thank the associate editor for his considerable time and
eﬀort invested in improving this manuscript.
Agrawal, S.G., K.G. King, J.F. Moore, P. Levinson, and J. McDonald. 2011.
Use of industrial byproducts to ﬁlter phosphorus and pesticides in golf
green drainage water. J. Environ. Qual. 40:1273–1280.
Baumann, T., S. Muller, and R. Niessner. 2002. Migration of dissolved heavy
metal compounds and PCP in the presence of colloids through a hetero-
geneous calcareous gravel and a homogeneous quartz sand: Pilot scale
experiments. Water Research 36:1213–1223.
Bolster, C.H., and G.M. Hornberger. 2007. On the use of linearized Lang-
muir equations. Soil Sci. Soc. Am. J. 71:1796–1806. doi:10.2136/
Bowden, L.I., A.P. Jarvis, P.L. Younger, and K.L. Johnson. 2009. Phosphorus
removal from waste waters using basic oxygen steel slag. Environ. Sci.
Technol. 43:2476–2481. doi:10.1021/es801626d
Cucarella, V., and G. Renman. 2009. Phosphorus sorption capacity of ﬁlter
materials used for on-site wastewater treatment determined in batch
experiments: A comparative study. J. Environ. Qual. 38:381–392.
Daly, K., P. Mills, B. Coulter, and M. McGarrigle. 2002. Modeling phospho-
rus concentrations in Irish rivers using land use, soil type, and soil phos-
phorus data. J. Environ. Qual. 31:590–599. doi:10.2134/jeq2002.0590
Drizo, A., J. Cummings, D. Weber, E. Twohig, G. Druschel, and B. Bourke.
2008. New evidence for rejuvenation of phosphorus retention capacity
in EAF steel slag. Environ. Sci. Technol. 42:6191–6197. doi:10.1021/
Drizo, A., C. Forget, R.P. Chapuis, and Y. Comeau. 2006. Phosphorus removal
by electric arc furnace steel slag and serpentininte. Water Res. 40:1547–
Drizo, A., Y. Comeau, C. Forget, and R.P. Chapuis. 2002. Phosphorus satura-
tion potential: A parameter for estimating the longevity of constructed
wetland systems. Environ. Sci. Technol. 36:4642–4648. doi:10.1021/
Gallimore, L.E., N.T. Basta, D.E. Storm, M.E. Payton, R.H. Huhnke, and
M.D. Smolen. 1999. Water treatment residual to reduce nutrients in
surface runoﬀ from agricultural land. J. Environ. Qual. 28:1474–1478.
Harmel, R.D., H.A. Torbert, B.E. Haggard, R. Haney, and M. Dozier. 2004.
Water quality impacts of converting to a poultry litter fertilization strat-
egy. J. Environ. Qual. 33:2229–2242. doi:10.2134/jeq2004.2229
King, K.W., J.C. Balogh, K.L. Hughes, and R.D. Harmel. 2007. Nutrient
load generated by storm event runoﬀ from a golf course watershed. J.
Environ. Qual. 36:1021–1030. doi:10.2134/jeq2006.0387
Koiv, M., M. Liira, U. Mander, R. Motlep, C. Vohla, and K. Kirsimae. 2010.
Phosphorus removal using Ca-rich hydrated oil shale ash as ﬁlter mate-
rial: e eﬀect of diﬀerent phosphorus loadings and wastewater com-
positions. Water Res. 44:5232–5239. doi:10.1016/j.watres.2010.06.044
Kostura, B., H. Kulveitova, and J. Lesko. 2005. Blast furnace slags as sor-
bents of phosphate from water solutions. Water Res. 39:1795–1802.
Leader, J.W., E.J. Dunne, and K.R. Reddy. 2008. Phosphorus sorbing materi-
als: Sorption dynamics and physicochemical characteristics. J. Environ.
Qual. 37:174–181. doi:10.2134/jeq2007.0148
Lee, J., R. Horton, and D.B. Jaynes. 2002. e feasibility of shallow time
domain reﬂectometry probes to describe solute transport through un-
disturbed soil cores. Soil Sci. Soc. Am. J. 66:53–57. doi:10.2136/
Makris, K.C., and W.G. Harris. 2006. Time dependency and irreversibil-
ity of water desorption by drinking-water treatment residuals: Implica-
tions for sorption mechanisms. J. Colloid Interface Sci. 294:151–154.
McDowell, R.W., A.N. Sharpley, and W. Bourke. 2008. Treatment of drain-
age water with industrial by-products to prevent phosphorus loss from
tile-drained land. J. Environ. Qual. 37:1575–1582. doi:10.2134/
Penn, C.J., and R. Bryant. 2006. Application of phosphorus sorbing materials
to streamside cattle loaﬁng areas. J. Soil Water Conserv. 61:303–310.
Penn, C.J., R.B. Bryant, M.A. Callahan, and J.M. McGrath. 2011. Use of
industrial byproducts to sorb and retain phosphorus. Commun. Soil Sci.
Plant Anal. 42:633–644. doi:10.1080/00103624.2011.550374
Penn, C.J., R.B. Bryant, P.A. Kleinman, and A. Allen. 2007. Sequestering
dissolved phosphorus from ditch drainage water. J. Soil Water Conserv.
Penn, C.J., and J.M. McGrath. 2011. Predicting phosphorus sorption onto steel
slag using a ﬂow-through approach with application to a pilot scale sys-
tem. J. Water Resour. Prot. 3:235–244. doi:10.4236/jwarp.2011.34030
Penn, C.J., J.M. McGrath, and R.B. Bryant. 2010. Ditch drainage manage-
ment for water quality improvement. p. 151–173. In M.T. Moore (ed.)
Agricultural drainage ditches: Mitigation wetlands for the 21st century.
Research Signpost, Kerala, India.
Pionke, H.B., W.J. Gburek, R.R. Schnabel, A.N. Sharpley, and G.F. Elwing-
er. 1999. Seasonal ﬂow, nutrient concentrations and loading patterns
in stream ﬂow draining an agricultural hill-land watershed. J. Hydrol.
Pionke, H.B., W.J. Gburek, A.N. Sharpley, and J.A. Zollweg. 1997. Hydro-
logic and chemical controls on P loss from catchments. p. 225–242. In
H. Tunney, O.T. Carton, P.C. Brookes, and A.E. Johnston (ed.) Phos-
phorus loss from soil to water. CAB International Publishers, New York.
Rhoton, F.E., and J.M. Bigham. 2005. Phosphate adsorption by ferrihydrite-
amended soils. Soil Sci. Soc. Am. J. 34:890–896.
Sharpley, A.N., P.J.A. Kleinman, A.L. Heathwaite, W.J. Gburek, J.L. Weld,
and G.J. Folmar. 2008. Integrating contributing areas and index-
ing phosphorus loss from agricultural watersheds. J. Environ. Qual.
Stoner, D., C.J. Penn, J.M. McGrath, and J.G. Warren. 2012. Phosphorus
removal with by-products in a ﬂow-through setting. J. Environ. Qual.
(in press). doi:10.2134/jeq2011.0049
Toride, N., F.J. Leij, and M.. van Genuchten. 1999. e CXTFIT code
for estimating transport parameters from laboratory or ﬁeld tracer ex-
periments, version 2.1. Research Rep. 137. USDA–ARS, U.S. Salinity
Laboratory, Riverside, CA.
Udawatta, R.P., P.P. Motavalli, and H.E. Garrett. 2004. Phosphorus loss and
runoﬀ characteristics in three adjacent agricultural watersheds with clay-
pan soils. J. Environ. Qual. 33:1709–1719. doi:10.2134/jeq2004.1709
USEPA. 1986. Quality criteria for water. EPA Rep. 440/5-86-001. USEPA,
Oﬃce of Water Regulations and Standards, Washington, DC.
USEPA. 1997. Method 3051a: Micro-wave assisted acid dissolution of sedi-
ments, sludges, soils, and oils. 2nd ed. U.S. Gov. Print. Oﬃce, Wash-
USEPA. 2001. Chesapeake Bay program nutrient trading fundamental prin-
ciples and guidlines. EPA 903-B-01-001. CBP/TRS 254/01. USEPA
Report Collection, Regional Center for Environmental Information,
USEPA Region 3, Philadelphia, PA.
Wei, X., R.C. Viadero, Jr., and S. Bhojappa. 2008. Phosphorus removal by
acid mine drainage sludge from secondary eﬄuents of municipal waste-
water treatment plants. Water Res. 42:3275–3284. doi:10.1016/j.