PUBLIC WORKS MANAGEMENT

Cothren&etPOLICY / July 2002 FLOW CONSTRUCTED WETLAND DESIGNS

al. / SUBSURFACE

INVESTIGATION OF SUBSURFACE FLOW

CONSTRUCTED WETLAND DESIGNS

Gianna M. Cothren, Ph.D., GIANNA M. COTHREN

has been an assistant pro- University of New Orleans
fessor in civil and environ-
mental engineering, Univer- SHULIN CHEN
sity of New Orleans, since Washington State University
August 1996. She has an
interest in Geographic JOHN H. PARDUE
Information System (GIS) Louisiana State University
applications in water
resources; physical, chemi-
cal, and biological wetland Results of a bench-scale model investigation of the dispersion impact on
functions; and urban and Subsurface Flow Constructed Wetland (SFCW) designs are presented in this arti-
rural environmental hydrol- cle. The research involved determining the dispersion number as interstitial veloc-
ogy. She is currently ity, a function of aspect ratio and flow rate, was varied. The range of dispersion
researching the development numbers found was from 0.107 to 0.345. Results indicated that as the aspect ratio is
of GIS and an associated decreased below 2:1, increased dispersion occurs. Additionally, an increase in dis-
database for the Lake persion number with increasing flow rate for larger aspect ratios (6:1, 3:1, 2:1)
Pontchartrain Basin, urban was revealed. Interstitial velocity was shown to control dispersion number for
and agricultural runoff higher flow rates; however, aspect ratio controlled the dispersion number when the
modeling in the ratio was reduced below 2:1. This research provides a valuable tool intended for
Pontchartrain Basin, and estimating the dispersion impact when applying alternative models to the design of
development of water and SFCWs.
energy budgets using the
Pontchartrain Basin GIS.
Background
Shulin Chen has a Ph.D.
(1991) in agricultural and Wastewater treatment using subsurface flow constructed wetlands (SFCWs) has grown in
biological engineering from popularity because the escalating cost of conventional wastewater treatment has led to the
Cornell University and is a search for cost-effective, environmentally sound alternatives requiring less electrical energy
registered civil engineer. and less labor for operation (U.S. Environmental Protection Agency [USEPA], 1988). The
Currently, he is an associate potential of the constructed wetland environment as a device for treating wastewater has been
professor in the Department recognized for some time. It is effective in treating wastewater because it supports a diverse pop-
of Biological Systems Engi- ulation of bacteria that are important in the removal of BOD5 and offers quiescent conditions for
neering at Washington State sedimentation, adsorption, filtration, and ion exchange (USEPA, 1988). Constructed wetlands
University. Research and have also been built to treat other types of wastewater such as acid mine drainage, industrial
teaching interests are in wastewater, agricultural runoff, and effluent from livestock operations. This type of system con-
water quality and sists of an excavated cell containing rock media with the wastewater level maintained below the
wastewater treatment, media surface. SFCWs provide a natural form of wastewater treatment that removes organics
watershed management, and nutrients based on the attached growth of microorganisms on crushed rocks where biologi-
waste management, and cal reactions take place. The water level of a SFCW is maintained just below the surface of the
bioresource utilization. media on which it is built.
It is frequently assumed that the performance of SFCWs can be described with a first-order
plug flow design model (Conley, Dick, & Lion, 1991; Crites, 1994; Reed, Crites, &
Middlebrooks, 1995; Watson, Choate, & Steiner, 1990) that is solely dependent on the system’s
hydraulic residence time and BOD decay reaction rate. However, the actual flow regime is

PUBLIC WORKS MANAGEMENT & POLICY, Vol. 7 No. 1, July 2002 32-45
© 2002 Sage Publications

32

Downloaded from pwm.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016

 Cothren et al. / SUBSURFACE FLOW CONSTRUCTED WETLAND DESIGNS 33

dependent on the physical dimensions and hydraulics of the system and lies somewhere John Pardue holds the
between plug flow and completely mixed flow. Elizabeth Howell Stewart
In the United States, many SFCWs cannot maintain the water level below the surface of the Professorship in the
media. This problem, referred to as surfacing or overflowing, is likely due to clogging of the Department of Civil & Envi-
media or improper hydraulic design (Kadlec, Bastiaens, & Urban, 1993). Another problem that ronmental Engineering at
may develop is channeling, in which preferred flow channels are developed, leading to the inef- Louisiana State University.
ficient use of the entire system. Many of the systems that were constructed when this treatment His research group investi-
technology was in its infancy were designed with a high aspect ratio (L:W) to ensure plug flow gates the fate and transport
(Reed et al., 1995). However, a high aspect ratio causes an increased resistance to flow, leading of organic chemicals in wet-
to an inadequate hydraulic gradient. Today, the generally accepted practice is to select an aspect land environments. He is an
ratio from 0.4:1 up to 3:1 (USEPA, 1993). As the acceptable design of the system is transposed associate professor in the
from a long, narrow channel (high aspect ratio) to a short, wide channel (low aspect ratio), the Department of Civil and
interstitial velocity decreases, which helps prevent the wastewater, limited by resistance to flow, Environmental Engineering,
from surfacing or channeling. However, this decrease in interstitial velocity can cause a signifi- where he has served in sev-
cant change in dispersion, which is neglected in the plug flow model. One reason that larger eral capacities, most
aspect ratios were used in earlier designs is because of the simplicity and familiarity of plug flow recently as the associate
reactors. Many of the existing systems were designed with an aspect ratio greater than 10:1 to chair for research.
ensure plug flow (Conley et al., 1991; Reed et al., 1995; Watson et al., 1990). Because lower
aspect ratios are now recommended (USEPA, 1993), it is likely that the plug flow model no lon-
ger adequately describes the system performance.
Because SFCWs are now commonly designed with low aspect ratios, plug flow is not
ensured. Therefore, it is important to understand the effect of lowering aspect ratio on the sys-
tem performance when the design is based on the plug flow assumption. Although it is difficult
to estimate dispersion for any particular case, an estimate based on the fundamental variables
affecting the dispersion number is required. These are the interstitial velocity (which is depend-
ent on flow rate, porosity, and cross-sectional area) and the pore geometry (which is dependent
on media characteristics such as the average grain diameter and permeability). In addition to
these fundamental variables, the dispersion number will also fluctuate with a series of uncon-
trollable variables specific to a particular SFCW situation. For example, differences in plant
root growth and die off, change in hydraulic conductivity with time, inconsistencies in construc-
tion, unsteady flow, and animal habitation are all parameters that cause uncertainty in the dis-
persion number. Thus, the model developed here is limited to micro-scale longitudinal disper-
sion, in that it does not account for the effects of plants, solids accumulation, construction
inconsistencies, varying media, and so forth. Nonetheless, if new advances are to be made in the
design of SFCWs, an estimate of dispersion will be required to model the system more
accurately.

Methodology

SMALL-SCALE SFCW MODEL TRACER STUDIES One reason that larger

aspect ratios were used
A small-scale SFCW system was designed and constructed in the laboratory to perform in earlier designs is
tracer experiments for a range of interstitial velocities using various aspect ratio and flow rate
because of the simplicity
scenarios. The 1.8 m × 2.4 m system was designed so that tracer experiments for five different
aspect ratios could be performed. A movable wall was placed at the required system width,
and familiarity of plug
bolted on the two upper corners, and structurally supported at the floor of the system. Then the flow reactors.
containment liner was draped over the wall, and the 5 cm to 8 cm limestone rocks were added.
Peristaltic pumps were set to pump the desired flow rate of clean water from a constant head
tank to the header pipe of the system. This header pipe consisted of attachable PVC units with a
1/16 in. slit cut along the centerline of the pipe. Each unit was adjusted along the width of the
unit until the water was flowing uniformly across the width of the system. An ISCO 3200 sam-
pler was programmed to collect 24 samples during a nonuniform time period, which collected
samples at a smaller time step when the peak concentration was expected. The static head was

Downloaded from pwm.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016

34 PUBLIC WORKS MANAGEMENT & POLICY / July 2002
At the start of the
experiments, a manual
sample was collected to
determine the
background level of
tracer.
Mean residence time
was obtained by
calculating the first
moment or centroid of
the curve.
controlled at 0.3 m with a PVC outlet elbow in the outlet control tank. Once the flow rate was set,
the sampler was programmed, the header pipe was adjusted, the tracer was injected into the
header pipe, and the preprogrammed sampler began taking samples during the time period.
Fluorescent dye tracing techniques were investigated, and it was determined that
Rhodamine-WT would be used because it has a low tendency to stain silt, dirt, organic, and
other suspended matter; is detectable at 1 ppb; and poses no known environmental or health haz-
ards (Crompton & Knowles, 1988). The volume used for each experiment is determined from
the calibration curve of the Sequoia-Turner Model 450 fluorometer using a standard solution
made from the stock solution of the Rhodamine-WT. For each of the aspect ratio arrangements,
the tracer injection volume was determined by multiplying the optimum detectable tracer con-
centration by the effective system volume (i.e., interstitial volume). At the start of the experi-
ments, a manual sample was collected to determine the background level of tracer. After the
experiment was completed, the samples were collected from the sampler and transferred to 100
ml glass vials. These samples were analyzed for absorbency with a Sequoia-Turner Model 450
Fluorometer, and then used to calculate tracer concentration. This procedure was repeated for
five different flow rates with each aspect ratio setup.
Variables of the scale model were cross-sectional area and flow rate. The aspect ratios (Ar)
were set at 6:1, 3:1, 2:1, 1:1, and 3:4. The width was changed to 0.3 m, 0.6 m, 0.9 m, 1.8 m, and
2.4 m, whereas the length was held constant at 1.8 m to achieve the preceding aspect ratios.
Depth was held constant at 0.3 m. Several flow rates were used, up to the maximum limited by
the Reynold’s number that must be maintained below 10. Flow rate (Q) and hydraulic residence
times (HRT) for each aspect ratio setup that was performed in the scale model is shown in Table
1. Flow rates were selected based on realistic hydraulic residence time values of field systems
for each experimental setup. The water tracing procedure used to characterize the nonideal flow
was a stimulus-response technique such that the system was disturbed and the response was
evaluated. Pulse (delta function) tracer studies in which the tracer was injected into the inlet and
samples were collected at the effluent (through-the-wall method) during a period of time were
performed at existing SFCWs and in the small-scale SFCW. The effluent tracer concentration
(response) was normalized and plotted against time, which results in a c-curve that was ana-
lyzed for its distribution characteristics, mean residence time, and variance. Dividing the efflu-
ent concentration values by the area under the c-curve performed normalization. This tracer
response curve directly reflects the residence time distribution of the tracer. Mean residence
time was obtained by calculating the first moment or centroid of the curve. The variance (or
square of the spread) of the distribution was calculated from the second moment and was used to
determine the dimensionless dispersion number for each specific aspect ratio and flow
condition.
Analysis of the c-curve distribution for each experiment was the next step in determining the
dispersion number. Pulse tracer studies were performed for varying interstitial velocity condi-
tions in the SFCW small-scale system. The tracer concentration was then plotted against time,
which resulted in a c-curve that was analyzed for mean and variance. For the given c-curve, the
mean (t) is given by
t=
∑ t C ∆t .
i i i
(1)
∑ C ∆t i i
The variance (σ2) is given by
∑ t C ∆t
2
i i i
σ 2
= − t 2, (2)
∑ C ∆t i i
where variance has units of (time)2 and represents the spread of the distribution. The variance of
this curve was used to determine the dimensionless dispersion number (D/vL) for the specific
Downloaded from pwm.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
 Cothren et al. / SUBSURFACE FLOW CONSTRUCTED WETLAND DESIGNS 35

Table 1: Small-Scale Model Experimental Flow Rates and Hydraulic Residence Times (HRTs)

Aspect Ratio
Q (ml/min) HRT (h) 6:1 3:1 2:1 1:1 3:4

Q1 48 82 164 218
Q2 36 36 72 109 218 291
Q3 24 55 109 164 327 436
Q4 12 109 218 327 654 872
Q5 8 1,308
Q6 6 218 436 654 1,308
Q6 3 436 1,308

NOTE: Q = flow rate.

media and flow condition, where D is the dispersion coefficient, v is the interstitial velocity, and
L is the reactor length (Levenspiel, 1972). For large amounts of dispersion, the shape of the
curve changes significantly with time and is nonsymmetrical with an extended tail. The
dimensionless parameter D/vL can be calculated from the variance by the following equation
(Levenspiel, 1972):
2
 D  D
σ θ = 2   + 8  ,
2
 vL   vL  (3)

where

σ2
σθ =
2
. (4)
t2

For pulse tracer input, as noted in Levenspiel (1972), in 1959 Aris has shown that the second
term can be dropped without error, resulting in the equation:

 D
σθ = 2  .
2
 vL  (5)

The longitudinal dispersion number was calculated according to the method outlined in
Levenspiel (1972). For each experiment, three replicate absorbance values were recorded for
each of the data points and averaged. This value was then converted to a tracer concentration
value with the equation describing the calibration curve. The nonuniform time series of when
the samples were collected and their corresponding tracer concentration values were placed on a
spreadsheet for further calculation, and the corresponding c-curves were generated. The first
step was to compute the first moment of the area (or centroid), which represents the hydraulic
residence time, using Equation 1. Next, the second moment, or the variance (see Equation 2),
was computed. These two values were substituted into Equation 5 and the dimensionless disper-
sion number was computed. The interstitial velocity was estimated by dividing the effective sys-
tem volume (interstitial volume) by the flow rate. Dividing the dimensionless dispersion num-
ber by this value and the system length gave the longitudinal dispersion coefficient with units of
time per length squared. The mass recovery of tracer was calculated by:

Vol = QΣC∆t, (6)

where Vol is the recovered tracer volume, Q is the flow rate, and C is the recorded tracer concen-
tration for each time step, ∆t. The percentage recovery, which was obtained by dividing the
recovered tracer volume by the injected tracer volume, was maintained within 90% to 115%.

Downloaded from pwm.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016

36 PUBLIC WORKS MANAGEMENT & POLICY / July 2002
Two field sites were
selected to perform
tracer studies for
comparison to the
small-scale model tracer
studies.
During the period of the
tracer study, all
wastewater flow from
the lagoons was diverted
to the third cell in the
parallel system.
FIELD SFCW TRACER STUDIES
Two field sites were selected to perform tracer studies for comparison to the small-scale
model tracer studies. The two field sites were the Baton Rouge Zoo (BRZ) SFCW in Baton
Rouge, Louisiana, and the Hansen’s Center for Disease Control Center (CRV) SFCW in
Carville, Louisiana. The BRZ system is constructed on a 2.2-acre site in the interior of the park.
The previously existing force main serves as the wastewater influent to a primary settling tank,
which flows into a 50,000 gallon flow equalization basin. Then it flows through a SFCW, an
ultraviolet disinfecting unit, and an effluent pump station; down a rock cascade to raise the dis-
solved oxygen level; and to a holding pond, where it is released to a canal leading to Cypress
Bayou. The design of this SFCW was based on a 24-hour detention time, a 0.6-m depth, 40%
void space, and a maximum uniform flow of 500,000 gpd. An aspect ratio of 9:1 was selected for
this system; however, it was later decreased to 5:1 due to surfacing problems by bypassing a
large section of the system (Barry, 1991). With this system, it is estimated that the zoo saves
$200,000 annually in sewer fees (Barry, 1991).
A tracer study was performed at the BRZ SFCW under normal operating conditions. The
objective of the tracer study was to determine the longitudinal dispersion number for the BRZ
SFCW and evaluate the plug flow system design assumption. The area under investigation had a
length of 113 m, a width of 22 m, and a depth of 0.6 m. The system was filled with 5 cm to 8 cm
limestone with a porosity of 0.446. Flow rate was measured in a 24-in. Parshall Flume with a
model 3200 ISCO bubble flow meter. The average flow rate was 537 L/m and the estimated
interstitial velocity was 6 m/hr. One liter of Rhodamine-WT dye was injected at the siphon inlet
of the system. An ISCO 3200 automatic sampler was programmed to collect samples during a
nonuniform time series that covered the length of the study.
The theoretical time to peak concentration, based on the initial measured flow rate of 625 L/
m and the estimated system volume, was calculated to be 10.4 hours. However, the actual peak
reached the sampling point at 24 hours because the actual flow rate averaged 537 L/m. Manual
samples were collected in addition to the automatic samples during the time that the actual peak
concentration occurred so that the data would be representative of the tracer exit distribution.
The CRV SFCW has been studied by other researchers in the past (Zacharitz & Fuller, 1993).
However, the system did undergo drastic changes that completely rearranged the hydraulics of
the system since that study. The CRV SFCW site has undergone a complete overhaul to circum-
vent water surfacing problems. The original 146 m long channel was divided into three parallel
flow cells with an aspect ratio of 3:1 of approximately 15 m in width by 46 m in length. The ini-
tial pea gravel media was excavated and replaced with 5 cm to 8 cm limestone. Also, the system
was not replanted because it performed so well without the plants. The original daily design
flow rate of 150,000 gpd was changed to 400,000 gpd for all three cells (R. Kelly, personal com-
munication, 1995). The SFCWs annual operation and maintenance cost is estimated at $650 per
year, which is about 6% of the total operation and maintenance costs of the entire treatment sys-
tem (USEPA, 2000).
The objective of this tracer study was to determine the longitudinal dispersion number for
one cell of the CRV SFCW and evaluate the plug flow system design assumption under normal
operating conditions. During the period of the tracer study, all wastewater flow from the lagoons
was diverted to the third cell in the parallel system. The flow rate was measured through a 45-
degree V-notch weir with a model 3200 ISCO bubble flow meter. The initial average flow rate
was 83 L/m, and the estimated theoretical hydraulic residence time and interstitial velocity were
41 hours and 1.0 m/hr, respectively. The Rhodamine-WT tracer (200 ml) was injected at the out-
let of the second lagoon. An ISCO 3200 automatic sampler was programmed to collect samples
during a nonuniform time series that covered the length of the study. Each sample was trans-
ferred from the automatic sampler to labeled 2 ml glass sample bottles and analyzed in the lab.
Downloaded from pwm.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
 Cothren et al. / SUBSURFACE FLOW CONSTRUCTED WETLAND DESIGNS 37

Table 2: Longitudinal Dispersion Numbers for Each Experiment

Aspect Ratio
HRT (h) 6:1 3:1 2:1 1:1 3:4

48 0.125 0.283 0.275

36 0.149 0.154 0.126 0.161 0.232
24 0.196 0.173 0.156 0.229 0.227
12 0.107 0.183 0.137 0.200 0.226
8 0.236
6 0.183 0.345 0.300 0.219
3 0.290 0.267

NOTE: HRT = hydraulic residence time.

Results and Discussion

EXPERIMENTAL SMALL-SCALE
MODEL DISPERSION NUMBERS

The dimensionless dispersion number for each of the 25 small-scale model tracer experi-
ments was calculated with Equation 5. The estimated dispersion numbers ranged from 0.107 to
0.345 for the various arrangements of aspect ratios and flow rates. These results are shown in
Table 2.

LONGITUDINAL DISPERSION AND

ASPECT RATIO RELATIONSHIP

Not all flow rates were repeatable for each aspect ratio, because decreasing the aspect ratio
results in an increase in effective volume, resulting in an unrealistically long hydraulic resi-
dence time. To determine if the aspect ratio was a controlling factor on the dispersion number,
the experimental flow rates of all the experiments were grouped into the following flow rate
ranges: 35 to 100 ml/min, 109 ml/min, 163 to 218 ml/min, 327 to 436 ml/min, 600 to 872 ml/
min, and 1,300 ml/min. In Figure 1, the aspect ratio was the independent variable and the longi-
tudinal dispersion number was the dependent variable for a specified flow rate range. A first-
order linear trend line was fit to the data and the corresponding coefficient of determination (r 2 )
was given (see Table 3). The r 2 value provides a measure of the goodness of fit of the regression
equation to the experimental data.
For the lowest flow rate range, a slight increase in longitudinal dispersion was detected with
an increase in aspect ratio. This was not the expected result, because the flow rate was held con-
stant and the system volume was reduced when the aspect ratio was increased, effectively
increasing the interstitial velocity. However, this same phenomena also occurred when consid-
ering the relationship between flow rate and the dispersion number.
No relationship between longitudinal dispersion and aspect ratio existed for the flow rates in
the range of 109 ml/min to 436 ml/min because the coefficient of determination values are so For the highest flow
small. For the highest flow rates, an increase in longitudinal dispersion with increasing aspect rates, an increase in
ratio was again detected. In general, as aspect ratio was increased, consequently reducing the longitudinal dispersion
system volume and increasing the interstitial velocity, the dispersion number had a tendency to with increasing aspect
increase. Preferred flow channels may develop when the interstitial velocity reaches a certain ratio was again
magnitude and some of the tracer may become trapped in dead zones, causing some of the tracer detected.
to reach the sampling point quickly and some after a delay. The overall result would be a greater
spread on the c-curve and a higher dispersion number.

Downloaded from pwm.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016

38 PUBLIC WORKS MANAGEMENT & POLICY / July 2002

0.4
1300
0.35

Longitudinal Dispersion
600-872
0.3 327-436

0.25 163-218
109
0.2
35 -100
0.15
0.1
0.05
0
0 2 4 6 8
Aspect Ratio

Figure 1: Dispersion Number and Aspect Ratio for Each Flow Rate Range

Table 3: Dispersion Number and Aspect Ratio Regression Results

2
Q (ml/min) Regression Equation r

35-100 y = 0.0101x + 0.1128 0.51

109 y = –0.0088x + 0.1673 0.29
162-218 y = –0.0123x + 0.2369 0.17
327-436 y = 0.0176x + 0.2006 0.23
600-872 y = 0.0589x + 0.1749 0.61
1,300 y = 0.0306x + 0.2019 0.73

NOTE: Q = flow rate.

LONGITUDINAL DISPERSION AND

INTERSTITIAL VELOCITY RELATIONSHIP

The interstitial velocity is a function of the flow rate and aspect ratio as well as the porosity
and depth of the system. However, the porosity of the media was assumed constant because
clean water was used, and the depth was maintained at 0.3 m throughout the experiments. The
interstitial velocity, determined by dividing the flow rate by the cross-sectional area, was the
independent variable and the longitudinal dispersion number was the dependent variable. A
first-order linear trend line was fit to the data and resulting coefficients of determination were
calculated.
Because the interstitial velocity is a function of the flow rate and the effective cross-sectional
area of the system, the results are similar to the relationship between longitudinal dispersion and
flow rate when the aspect ratio is held constant. Table 4 gives the regression equations and corre-
sponding coefficients of determination associated with the longitudinal dispersion and intersti-
tial velocity relationship for each aspect ratio. Table 5 and Figure 2 show the results of similar
data between the longitudinal dispersion and flow rate relationship.
In general, when the aspect ratio is large (6:1, 3:1, and 2:1), an increasing longitudinal dis-
persion with increasing interstitial velocity exists, indicating the development of preferred flow
channels and dead zones at interstitial velocities of this magnitude. However, the trend was not
evident if the higher velocities were not considered. It is evident from the dispersion and inter-
stitial velocity relationship that the interstitial velocity does not control the dispersion number
after the aspect ratio is decreased beyond 2:1. When only the first three aspect ratios (6:1, 3:1,
and 2:1) are considered, an exponentially increasing relationship between longitudinal disper-
sion and interstitial velocity exists. The equation defining this relationship is

Downloaded from pwm.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016

 Cothren et al. / SUBSURFACE FLOW CONSTRUCTED WETLAND DESIGNS 39

Table 4: Dispersion Number and Interstitial Velocity Regression Results

Aspect Ratio Regression Equation r2

6:1 y = 0.072x + 0.1277 0.64

3:1 y = 0.2273x + 0.1047 0.93
2:1 y = 0.0866x + 0.1266 0.66
1:1 y = –0.0199x + 0.2286 0.02
3:4 y = –0.0214x + 0.2507 0.09
6:1 to 3:4 y = –0.0561x + 0.1773 0.23

Table 5: Dispersion Number and Flow Rate Regression Results

2
Aspect Ratio Regression Equation r

6:1 y = 0.0003x + 0.1277 0.64

3:1 y = 0.0005x + 0.1047 0.93
2:1 y = 0.0001x + 0.1266 0.66
1:1 y = 0.00002x + 0.2286 0.02
3:4 y = –0.00001x + 0.2507 0.09

0.4
Aspect Ratio = 3:4
0.35 Aspect Ratio = 1:1
Longitudinal Dispersion

0.3 Aspect Ratio = 2:1

0.25 Aspect Ratio = 3:1

Aspect Ratio = 6:1
0.2
0.15
0.1
0.05
0
0 200 400 600 800 1000 1200 1400
Flow Rate (ml/min)

Figure 2: Dispersion Number and Flow Rate

y = a + be(–x), (7)

where the variables a and b are curve-fitting coefficients with a = 0.319 and b = 0.212 and with a
coefficient determination of 0.6 (see Figure 3).
A relationship between interstitial velocity and longitudinal dispersion is not recognized
until some significant velocity is reached. At a normalized velocity of 3 units, the experimental A relationship between
velocity at aspect ratios of 6:1 and 3:1 is approximately 0.5 ft/hr. At aspect ratios of 2:1 and 3:4, interstitial velocity and
the significant velocity is approximately 0.4 ft/hr. It is concluded that the significant velocity longitudinal dispersion
where a relationship between interstitial velocity and the dispersion number begins to develop is not recognized until
ranges from 0.4 to 0.5 ft/hr. some significant velocity
is reached.

Downloaded from pwm.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016

40 PUBLIC WORKS MANAGEMENT & POLICY / July 2002
Dispersion numbers
were normally
distributed within each
group of velocities;
therefore, analysis of
variance was utilized.
Rank 1 Eqn 21 y=a+be^(-x)
r^2=0.57905024 DF Adj r^2=0.50889195 FitStdErr=0.049048769 Fstat=17.882545
a=0.31853429
b=-0.21154048
0.4
0.35
0.3
D/vL 0.25
0.2
0.15
0.1
0.05
0 0.5 1 1.5 2
v(ft/hr)
Figure 3: Longitudinal Dispersion and Interstitial Velocity Relationship for the Bench-Scale System Aspect
Ratios of 6:1, 3:1, and 2:1
TRACER RECOVERY IN THE BENCH-SCALE MODEL
Tracer recovery was computed using Equation 6. This equation is based on an approximation
of the area under the c-curve; therefore, computational error is unavoidable. Because the change
in time is computed by a forward difference, the recovered concentration is overestimated at the
highest concentrations until the peak concentration is reached. Also, if the equation represent-
ing the calibration of the fluorometer is a high estimate based on the calibration data, the recov-
ered concentration will result in a high estimate. The tracer recovery calculated for most of the
bench-scale experiments ranged from 90% to 115%.
STATISTICAL ANALYSES
One-way analysis of variance was performed on groups of the dispersion numbers to deter-
mine if a statistically significant relationship exists between dispersion and aspect ratio and dis-
persion and interstitial velocity. No significant relationship was found between dispersion and
aspect ratio when all data were included in the analysis (see Table 6). The conclusion of the test
was that the differences in the median values among the treatment groups were not great enough
to exclude the possibility that the differences were due to random sampling variability at the
36% level.
Downloaded from pwm.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
 Cothren et al. / SUBSURFACE FLOW CONSTRUCTED WETLAND DESIGNS 41

Table 6: Kruskal-Wallis One-Way Analysis of Variance on Ranks

Group (aspect ratio) n Median 25% 75%

6 5 0.183 0.138 0.219

3 4 0.178 0.163 0.264
2 6 0.146 0.126 0.266
1 5 0.219 0.190 0.245
0.75 5 0.236 0.230 0.254

NOTE: H = 4.35, p = .361.

Table 7: One-Way Analysis of Variance With Grouped Velocities

Group (v) n M SD SEM df SS MS

0.03 to 0.09 10 0.192 0.0563 0.0178 2 0.0288 0.01438

0.09 to 0.15 5 0.174 0.0545 0.0244 19 0.0582 0.00306
0.15 to 0.4 7 0.263 0.0545 0.0206 21 0.0870

NOTE: F = 4.69, p = .022.

By contrast, a significant relationship was found between interstitial velocity and dispersion
when the velocities were grouped into the following categories: 0.03 to 0.09 m/hr, 0.09 to 0.15
m/hr, and 0.15 to 0.4 m/hr. Dispersion numbers were normally distributed within each group of
velocities; therefore, analysis of variance was utilized. The analysis of variance (see Table 7)
revealed that interstitial velocity was explaining a significant portion of the variability of the dis-
persion numbers. The differences in the mean values among the treatment groups were greater
than what would be expected by chance. There was a statistically significant difference (p =
.022).
The Student-Newman-Keuls method was used to identify differences between the treat-
ments (see Table 8). The q test statistic in this method is the number of rank sums spanned in the
comparison. Traditionally, it is concluded that there are significant differences when p < .05.
Results indicated that the 0.15 to 0.4 m/hr treatment was significantly different than the lower
two groupings of velocity. However, the 0.03 to 0.09 and 0.09 to 0.03 groups were not statisti-
cally different from each other. The analysis of variance provided statistical evidence that
higher velocities increased dispersion in the tracer experiments.
Aspect ratio did explain a significant amount of variability when the high velocity data (v >
0.15 m/hr) were excluded from the analysis (see Table 9). The differences in the mean values
among the treatment groups were greater than what would be expected by chance (p = .003).
Comparisons of means using the Student-Newman-Keuls method indicated that the 1:1 and
3:4 aspect ratio dispersion numbers were significantly greater than the higher aspect ratios (6:1,
3:1, and 2:1) but were not significantly different from each other (see Table 10). If a group was
found not to be significantly different from another group, all groups with ranks in between the
rank sums of those two groups are also assumed not significant, resulting in no test performed
for those groups. These analyses of variance provided statistical evidence that higher flow rates
and lower aspect ratios both act to increase dispersion in constructed wetlands.

BRZ TRACER STUDY RESULTS The actual peak

concentration occurred
The tracer experiment at the BRZ SFCW revealed a dimensionless dispersion number of at 24 hours, which was
0.078 with an interstitial velocity of 6 m/hr and system length of 113 m (see Figure 4). With 14 hours later than the
these results, the dispersion coefficient is calculated to be 53 m2/hr. The theoretical hydraulic theoretical hydraulic
residence time based on the initial flow rate of 624 L/m and the effective system volume was residence for this size
10.4 hours. However, visible evidence of the peak concentration passing the sampling point was system.

Downloaded from pwm.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016

42 PUBLIC WORKS MANAGEMENT & POLICY / July 2002

Table 8: Student-Newman-Keuls Pairwise Multiple Comparison

Group (v) q p < .05

0.15-0.4 vs. 0.09-0.15 3.846 Yes

0.15-0.4 vs. 0.03-0.09 3.638 Yes
0.03-0.09 vs. 0.09-0.15 0.839 No

Table 9: One-Way Analysis of Variance

Group (aspect ratio) n M SD SEM df SS MS

6 3 0.150 0.0446 0.02573 4 0.0333 0.00832

3 3 0.170 0.0148 0.00854 13 0.0158 0.00121
2 4 0.136 0.0144 0.00719 17 0.0491
1 4 0.221 0.0558 0.02791
0.75 4 0.245 0.0217 0.01086

NOTE: F = 6.85, p = .003.

Table 10: Student-Newman-Keuls Pairwise Multiple Comparison

Group (aspect ratio) q p < .05

0.75 vs. 2 6.249 Yes

0.75 vs. 6 5.016 Yes
0.75 vs. 3 3.992 Yes
0.75 vs. 1 1.384 No
1 vs. 2 4.865 Yes
1 vs. 6 3.735 No
1 vs. 3 2.710 No test
3 vs. 2 1.793 No
3 vs. 6 0.958 No test
6 vs. 2 0.769 No test

noted during a site visit after 20 hours. Manual samples were taken in between automatic sam-
ples so that the peak concentration would not be missed. The actual peak concentration occurred
at 24 hours, which was 14 hours later than the theoretical hydraulic residence for this size sys-
tem. The centroid of the c-curve, or calculated hydraulic residence time, occurred at 31 hours.
The peak concentration occurring earlier than the calculated hydraulic residence time is a direct
result of the variance (or dispersion) associated with the c-curve. An explanation for the peak
concentration occurring 14 hours later than the theoretical hydraulic residence time is that the
flow rate used to compute the hydraulic residence time overestimated the actual average flow
rate of 537 L/m. Also, the design depth of 0.6 m used to calculate the effective volume was likely
an underestimate of the actual construction depth of the system.
The calculated tracer recovery based on Equation 4, the estimated concentration, and the
estimated flow rate was 121%. A tracer recovery greater than 100% reveals a likely overestima-
tion of the flow rate during times when it was not recorded. At times when the flow rate was not
recorded during a sampling event, the previous flow rate value was used in recovery calculation.
Because the area under the c-curve (C × dt) is multiplied by each flow rate and summed to deter-
mine the tracer recovery, a high estimate of flow rate during a high concentration sample event
would lead to an overestimation on the tracer recovery.

Downloaded from pwm.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016

 Cothren et al. / SUBSURFACE FLOW CONSTRUCTED WETLAND DESIGNS 43

0.045

Normalized Rhodamine-WT
0.04
0.035
0.03

Concentration
0.025
0.02
0.015
0.01
0.005
0
0 20 40 60 80 100 120 140 160

Time (hrs)

Figure 4: Baton Rouge Zoo Tracer Study Results

Normalized Rhodamine-WT

0.06
0.05
Concentration

0.04
0.03
0.02
0.01
0
0 20 40 60 80 100 120 140 160

Time (hrs)

Figure 5: Carville Subsurface Flow Constructed Wetlands Tracer Study Results

CRV TRACER STUDY RESULTS

The dimensionless dispersion number for the CRV study was calculated to be 0.119, with an
interstitial velocity of 1.0 m/hr and a system length of 46 m resulting in a dispersion coefficient
of 5 m2/hr (see Figure 5). The theoretical hydraulic residence time was calculated to be 41 hours,
based on the initial recorded flow rate of 83 L/m and the system’s effective volume. Therefore,
the time increment was reduced at times near 41 hours. However, the peak concentration
occurred at 16 hours during a time when the sampler was programmed to take samples at a
higher time increment. The calculated hydraulic residence time from the c-curve (centroid) was
25 hours. Because the actual peak concentration occurred 25 hours earlier than expected based
on the theoretical hydraulic residence time, few samples were taken as the peak concentration
passed. A lower flow rate would normally result in a longer hydraulic residence time because
the effective system volume remains constant; however, the hydraulic residence time actually
increased as the average flow rate decreased. Possibly, the flow meter was calibrated incorrectly
or the equation used for the V-notch weir did not accurately estimate actual outlet flow condi-
tions. The tracer recovery was calculated to be 68%.

Downloaded from pwm.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016

44 PUBLIC WORKS MANAGEMENT & POLICY / July 2002
Tracer studies were
performed in the small-
scale SFCW with
hydraulic residence
times comparable to
field systems.
Research Implications
The results of this research offer a tool that can be used to estimate the relationship between
the dispersion number and interstitial velocity as a function of flow rate and aspect ratio, which
leads to an initial estimate of the dispersion coefficient for a system to be designed based on an
alternative to the plug flow model. A preliminary system size requirement comparison was
made between systems designed based on the plug flow equation and the convective-dispersive
equation. As an example with a hypothetical design situation, the plug flow equation is used to
obtain the system’s cell design length of 39.8 m with a 30.2 m/d interstitial velocity and a 31.7 hr
hydraulic residence time. Using these same parameters and Equation 7, the estimated disper-
sion coefficient is calculated to be 175 m2/d for the cell. This estimate is used with the Wang and
Chen (1995) solution to the convective-dispersive equation to calculate a total required length of
54.5 m, which is 14.7 m longer than the length of a system designed to operate under plug flow
conditions.
Conclusions
A SFCW small-scale model was designed and constructed for the purpose of determining the
relationship between the longitudinal dispersion number and the interstitial velocity as a func-
tion of the flow rate and aspect ratio. Tracer studies were performed in the small-scale SFCW
with hydraulic residence times comparable to field systems. Each of the experiments was per-
formed with a different flow rate and aspect ratio combination, and the dispersion numbers for
each condition were determined. The dimensionless dispersion numbers calculated from the
bench-scale model tracer experiments ranged from 0.107 to 0.345. These values are comparable
to the range of 0.11 to 0.38 (Kadlec et al., 1993).
Empirical relationships between the longitudinal dispersion number, flow rate, aspect ratio,
and interstitial velocity were developed. The results indicated an increase in dispersion number
with increasing flow rate for larger aspect ratios (6:1, 3:1, and 2:1). Also, as the aspect ratio was
decreased, allowing a greater available area for preferred flow channels and dead water zones,
the dispersion number increased. Because the interstitial velocity is a function of both the flow
rate and the aspect ratio, interstitial velocity controls the dispersion number only at higher
aspect ratios. The significant interstitial velocity, where a relationship between velocity and the
dispersion number begins to develop, ranges from 0.4 to 0.5 ft/hr in the small-scale system.
However, when the aspect ratio is reduced beyond 2:1, the aspect ratio controls the dispersion
number. The statistical evidence of the research results indicated that as the aspect ratio
decreases (becomes shorter and wider), the dispersion number increases. And in fact, results of
the BRZ study (aspect ratio = 5:1) and CRV study (aspect ratio = 3:1) show an increase in disper-
sion number (0.078 to 0.119) as the aspect ratio is decreased from 5:1 to 3:1. However, the statis-
tical analyses also indicated that dispersion increases as the flow rate is increased, yet the disper-
sion number decreased as the flow rate per unit width and depth of the CRV system and BRZ
system increased from 3.8 ft/hr to 19.8 ft/hr, respectively. Therefore, the increasing dispersion
number with decreasing aspect ratio was validated in the field but the increasing dispersion with
increasing flow rate was not.
This natural treatment technology has been around for some time, but knowledge of its com-
plex operation is not yet advanced enough to provide adequate predictive models. The aim of
this study is to provide a means for using more detailed predictive models as this technology
gains national regulatory acceptance. Because plug flow clearly does not represent the flow
condition occurring in this type of system, SFCWs should be designed based on a partially
mixed model. This research provides a valuable tool for applying alternative models to the
design of SFCWs. By recognizing that the parameter controlling the dispersion number occurs
at some significant velocity, it is possible to determine which of the parameters, flow rate or
aspect ratio, will control the SFCW hydraulic design.
Downloaded from pwm.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016
 Cothren et al. / SUBSURFACE FLOW CONSTRUCTED WETLAND DESIGNS 45

References

Barry, S. J. (1991). Greater Baton Rouge Zoo uses innovative technology to save dollars. Baton Rouge, LA: Simons J.
Barry & Associates, Consulting Engineers.
Conley, L. M., Dick, R. L., & Lion, L. W. (1991). An assessment of the root zone method of wastewater treatment. Jour-
nal of Water Pollution Control Federation, 63, 239-247.
Crites, R. W. (1994). Design criteria and practice for constructed wetlands. Water Science Technology, 29(4), 1-6.
Crompton & Knowles (1988). Intracid rhodamine WT liquid (Technical Data Bulletin, IPD No. 022). Charlotte, NC:
Industrial Products Division.
Kadlec, R. H., Bastiaens, W., & Urban, D. T. (1993). Hydrological design of free surface treatment wetlands, con-
structed wetlands for water quality improvement. CRC Press.
Levenspiel, O. (1972). Chemical reaction engineering (2nd ed.). New York: John Wiley.
Reed, S. C., Crites, R. W., & Middlebrooks, E. J. (1995). Natural systems for wastewater treatment (2nd ed.). New York:
McGraw-Hill.
U.S. Environmental Protection Agency. (1988). Design manual: Constructed wetlands and aquatic plant systems for
municipal wastewater treatment (EPA/625/1-88/022). Cincinnati, OH: Author.
U.S. Environmental Protection Agency. (1993). Guidance for design and construction of a subsurface flow constructed
wetland. Water Management Division, Municipal Facilities Branch, Technical Section. Cincinnati, OH: Author.
U.S. Environmental Protection Agency. (2000). Constructed wetlands treatment of municipal wastewaters, Office of
Research and Development, CN, Ohio (625R-99/010). Cincinnati, OH: Author.
Wang, G.-T., & Chen, S. (1995). A new model describing convective-dispersive phenomena by using the mixing-cell
concept. New York: Elsevier Science.
Watson, J. T., Choate, K. D, & Steiner, G. R. (1990). Performance of constructed wetland treatment systems at Benton,
Hardin, and Pemboke, Kentucky, during the early vegetation establishment phase (Constructed Wetlands in Water
Pollution Control Proceedings). New York: Pergamon.
Zacharitz, W. H., & Fuller, J. W. (1993). Performance of an artificial wetlands filter treating facultative lagoon effluent
at Carville, Louisiana. Water Environment Research, 65(1), 46-51.

Downloaded from pwm.sagepub.com at PENNSYLVANIA STATE UNIV on May 17, 2016