9th International Conference on Urban Drainage Modelling
Belgrade 2012
Verifying a stormwater biofiltration model
Anna Lintern1 , Edoardo Daly2, Ana Deletic3
1
Centre for Water Sensitive Cities, Department of Civil Engineering, Monash University, Clayton 3800,
Victoria, Australia, anna.lintern@monash.edu
2
Centre for Water Sensitive Cities, Department of Civil Engineering, Monash University, Clayton 3800,
Victoria, Australia, edoardo.daly@monash.edu
3
Centre for Water Sensitive Cities, Department of Civil Engineering, Monash University, Clayton 3800,
Victoria, Australia, ana.deletic@monash.edu
ABSTRACT
Biofilters are a technology being implemented to manage stormwater by reducing
stormwater runoff volumes, peak flows, and pollutant levels. In light of their
increasing popularity, a model for assessing the performance of stormwater
biofilters was previously developed by the authors. This model contains a
hydrologic and a water quality component. The objective of this study was to test
the hydrologic component against data obtained over an 8-month period between
2010 and 2011 from an operating field system, the Monash University carpark
biofilter. We found that the model is able to replicate some of the observed peak
flows, but is unable to replicate low flows. The model slightly underestimated the
system’s hydraulic performance, with the modelled total outflow volumes over the
8-month period being 12 to 32% less than the measured volumes. Further work is
required to refine the model by accounting for the spatial and temporal variability
of hydraulic conductivity, and by incorporating soil-specific field capacity
parameters. However, despite these limitations, the model is a useful design tool as
it is able to provide a conservative estimate of biofilter outflow volumes and
relatively accurate estimate of peak flow reductions in a quick and easy way.
KEYWORDS
Biofilter, design, model, soil moisture, water balance
1
INTRODUCTION
Urban growth has led to an increase in impervious surface area, which has resulted in greater flows of
urban stormwater into receiving waters (Paul and Meyer, 2001; Roy et al., 2008). Stormwater flows
are a problem in both rural and urban environments, not only leading to flooding but also contributing
to the deteriorating health of rivers, lakes and bays (Walsh et al., 2004). Increased stormwater flows
can lead to geomorphic changes in receiving waters, such as scouring and erosion and furthermore,
these increased flows carry pollutants, such as nutrients, sediments and heavy metals, into receiving
natural waters (Paul et al., 2001; Roy et al., 2008).
1
Biofiltration systems (also known as bioretention systems or rain gardens) are one tool used to
mitigate these effects of urban stormwater, by both reducing the magnitude and improving the quality
of stormwater. These systems can attenuate peak stormwater flows by detaining stormwater inflows on
the surface of the biofilter (cross section shown in Figure 1(a)), allowing the water to gradually filter
vertically through the filter media and either exfiltrate into the surrounding soil if the system is
unlined, or to be carried by the outflow pipe to receiving waters. Polluted stormwater is also treated
during this filtering process. Additionally, a fraction of the water detained in the filter media is lost
through evapotranspiration. Systems can also include a submerged zone, where a permanently
saturated layer is maintained in the soil media to improve the removal of nutrients and enhance
vegetation water availability during dry seasons (Zinger et al., 2007; Blecken et al., 2009; Zhang et al.,
2011).
a
b
Figure 1. Schematic representation of a biofiltration system (a) and the water flow module buildingblocks (b) (Lintern et al., submitted).
Studies have shown that biofilters are a promising stormwater management tool. A field study of
biofiltration systems found that they are able to reduce total stormwater volumes by up to 30% and
peak flows by up to 85% (Hatt et al., 2009). They have also been found to be effective in the removal
of Total Suspended Solids (TSS), with recorded removal rates of over 90% (Bratieres et al., 2008;
Davis et al., 2009; Trowsdale and Simcock, 2011). Although the treatment of nutrients is more
variable, high removal rates have been observed for biofilters of certain designs. For example,
recorded Total Phosphorus (TP) removal rates from vegetated biofilters of over 75% were recorded by
Davis (2007) and Zhang et al. (2011) found Total Nitrogen (TN) removal rates of approximately 90%
by a vegetated system with a submerged zone.
In a previous study, a lumped model was developed for assessing the performance of stormwater
biofiltration systems in the removal of TSS, TP and TN loads (Lintern et al., submitted). This model
has two components, a hydrological component that assesses the flow of water through the
biofiltration system, and a water quality component that evaluates the removal of the pollutants. The
aim of this study is to test the hydrological component using an operating field system, the Monash
2
University carpark biofilter. This is a necessary step in ensuring that the model can be used by
stormwater practitioners to effectively design biofilters.
2
METHODOLOGY
2.1
Model description
The hydrologic stormwater biofiltration model is presented in detail in Lintern et al. (submitted). Thus,
only a brief description of the model and the relevant equations are presented below.
As shown in Figure 1 (b), the hydrological component models the water balance of a biofilter using a
bucket system with three stores: for ponding on the surface of the biofilter, for the filter media, and for
the submerged zone (if there is one present). Given the inflow volume, the model evaluates the soil
moisture and the outflows from the system (overflow of untreated stormwater, evapotranspiration,
outflow of treated stormwater and exfiltration into surrounding soil). This model can be applied to
systems with or without submerged zones.
The first bucket represents the surface of the biofilter. The water depth or ponding depth (hp) on the
surface of the biofilter is represented using Equation 1 as
A
dh p
dt
 RA  Qin  Qo  Q pf ,
(1)
where, A is the surface area of the biofilter, R is the rainfall directly onto the biofilter surface, Qin is the
inflow rate of stormwater, Qo is the overflow (applicable only when hp is greater than the weir height,
hw) and Qpf is the infiltration down into the filter medium. Qo and Qpf can be evaluated as
Qo  Cw Lw hp  hw  ,
3/ 2
(2)
and
Q pf


 hp
A Df  n f 
,.
 min   A  Qin , A  k f , 1  s 

t
t
 








potentialinfiltration
filter media capacity
 availablewater

(3)
where Cw is the weir constant coefficient, Lw is the weir length, kf is the saturated hydraulic
conductivity of the filter media, s is the relative soil moisture content, Df is the filter medium depth and
nf is the filter media porosity.
The second bucket represents the filter medium. The soil moisture content in the filter medium is
represented by
A Df n f
ds
 Qpf  QET  Qcr  Q fd ,
dt
(4)
where QET, Qcr and Qfd are the evapotranspiration rate, the capillary rise from the submerged zone, and
the infiltration into the drainage layer from the filter medium respectively. These three fluxes are
modelled as
3
QET
s  sh

A  Ew
sh  s  sw

sw  sh

 s  s 
 
  A   E w  Emax  E w  * w  s w  s  s * ,
 s  s w 
 

A  Emax
s*  s  1


Qcr  A  Cr (s  s * )  (s fc  s) s *  s  s fc ,
Q fd  A k f s 
hp  D f
Df
(5)
(6)
,
(7)
where Ew is the rate of evaporation from bare soil, s* is the water stress point, or the threshold below
which vegetation begins to reduce water losses, sfc is the field capacity, sw is the wilting point, sh is the
hygroscopic point and Emax is the maximum evapotranspiration rate. Additionally, Cr is the capillary
rise constant from the submerged zone (which is assumed to be 4Emax/[2.5(sfc-s*)2]), and  is
infiltration coefficient of the soil.
The third bucket is the drainage layer. Here, to track the depth of water in the submerged zone (hsz) we
use
A nsz
dhsz
 Q fd  Qcr  Qds  Qout ,
dt
(8)
where nsz is the porosity of the submerged zone, Qds is the exfiltration rate (Equation 9) and Qout is the
piped outflow (Equation 10).
Qds  ks  A  Cs hsz P 
Qout 
(9)
A nsz
hsz  Dsz 
t
(10)
In Equation 9, ks is the saturated hydraulic conductivity of the submerged zone, Cs is a coefficient used
to account for water flow from the sides of the drainage layer and P is the perimeter of the drainage
layer. In Equation 10, DSZ is the maximum submerged zone depth.
Equations 1 through to 10 have been programmed into the software Model for Stormwater
Improvement Conceptualisation (MUSIC) version 5 (eWater, 2010).
2.2
Data collection
Both flow and water quality data were collected from the Monash University carpark biofilter, an
operating biofiltration system in Melbourne, Australia. This system treats runoff from a nearby carpark
(4500 m2 in area) and contains 3 lined cells. Each cell has a surface area of 15 m2 and works as
separate systems. Cell 1 has a sandy loam filter media and is vegetated with Carex appressa, Cell 2
has a sandy soil filter media and is vegetated with Melaleuca ericifolia, and Cell 3 has a sandy loam
filter media that is enriched with compost and contains C. appressa. The filter media of all cells is 500
mm deep, with an additional transition layer of medium sand and drainage layer of fine gravel, each
100 mm. In addition, Cells 2 and 3 contain a submerged zone of 200 mm depth containing a carbon
4
source. Although Cells 1 and 3 have been operating for 6 years, Cell 2 is only 3 years old. All three
cells have overflow weirs of average width of 0.25 m and height of 0.2 m.
Inflow and piped outflow data were collected every minute using a V notch weir and ultrasonic depth
sensors (Siemens Miltronics) in the three cells from September 2010 to April 2011. The hydraulic
conductivities of the cells were measured in April 2010, five months prior to the study period.
Constant head tests were used to measure the hydraulic conductivity, where the systems were flooded
for four days, with a constant pressure head. The outflows were measured during this period, and
Darcy’s Law was used to determine the hydraulic conductivity.
2.3
Data analysis
The three cells of the Monash University carpark biofilter were modelled in MUSIC by importing into
the program, the inflows observed between September 2010 and April 2011. Inflows and outflows
obtained on the 17th of October and between the 3rd and 16th of November were excluded from the data
set because the outflows were not recorded due to flood tests and weir calibration tests. It was assumed
that the porosity of the filter media and the submerged zone was 0.40. The hydraulic conductivity, k,
for the period from September 2010 to April 2011, was the only calibration parameter used in the
study. Evapotranspiration data already programmed into MUSIC for Melbourne was used and for the
other parameters, default values programmed into MUSIC were used.
Two objective functions were used to calibrate the hydraulic conductivity of each cell, and the effect
of the choice of objective function on the modelled results was compared. These were the coefficient
of efficiency (Nash and Sutcliffe, 1970),
(12)
and a more unbiased objective function that favours lower flows,
.
(13)
In the previous objective functions, E is the coefficient of efficiency, i is the timestep, n is the total
number of timesteps, Mi is the measured outflow at timestep i, Pi is the modelled outflow at timestep i,
and is the average of all modelled outflows.
The model was run at a 6 minute timestep in MUSIC in order to capture the short term variability in
flows. To validate the hydrological component of the model, the MUSIC outputs (flows) were
compared to the data collected at the Monash University carpark biofilter and outflow volumes were
also evaluated and compared.
3
3.1
RESULTS AND DISCUSSION
Calibration of hydraulic conductivity
Table 1 shows the measured and calibrated hydraulic conductivity values. Cell 3 had identical
hydraulic conductivity values for the two objective functions. However, for the other two cells, there is
a significant discrepancy between the hydraulic conductivity calibrated using the two objective
functions. This is due to the fact that whilst E favours high flows, φ favours lower flows (McCarthy et
al., 2011). This is evident in the model runs shown in Figure 2.
5
The hydraulic conductivity calibrated using the unbiased function was within ±22% of the measured
values for Cells 1 and 2, and ±28% for Cell 3. This is within the range of accuracy expected for field
measurements of hydraulic conductivity (Le Coustumer et al., in press). However the calibrated
hydraulic conductivity values determined using the coefficient of efficiency objective function, were
far apart from the measured; the relative differences were in the range -22% to +87% of the measured
values.
Table 1 : Measured hydraulic
objective functions.
k measured in
April 2010
123 mm/hr
Cell 1
144 mm/hr
Cell 2
77 mm/hr
Cell 3
conductivity and hydraulic conductivity calibrated using the two
Coefficient of efficiency
k
E
230 mm/hr
0.62
200 mm/hr
0.84
60 mm/hr
0.65
Unbiased function
k
φ
100 mm/hr
2.16
170 mm/hr
0.33
60 mm/hr
0.37
Figure 2: Measured flows and modelled flows for Cells 1 to 3 for 12/01/2011 to 15/01/2011 using
hydraulic conductivity values calibrated using the two objective functions. Modelled (E) represents
6
modelled flows when k is calibrated using the coefficient of efficiency, and Modelled (φ) represents
modelled flows when k is calibrated using the unbiased function (note that for Cell 3, the two modelled
flows are identical and the lines overlap).
3.2
Model performance
Figure 2 indicates that the model is not always able to accurately predict peak outflows in the
hydrograph, particularly for more conductive soils of Cells 1 and 2. There are instances where the peak
flows are overestimated, as well as instances where they are underestimated. As discussed previously,
when the hydraulic conductivity is calibrated using the unbiased function, these discrepancies in peak
outflows are more significant. In addition, it is evident that the modelled biofilter drains too quickly,
leading to differences in the measured and modelled durations of the flow events. The total outflow
volumes (Figure 3) show that these over- and underestimations occur throughout the whole study
period and not just the 3 days shown in Figure 2.
As shown in Figure 3, the total outflow volumes are also modelled more closely when the hydraulic
conductivity is calibrated using the coefficient of efficiency. Total outflow volumes modelled using
the hydraulic conductivity calibrated using the unbiased function significantly underestimates the
measured outflow volumes for all three cells. The inability of the unbiased function to match the peak
flows is most likely the main cause of this difference.
Regardless of the hydraulic conductivity entered into the model, the model seems to perform the best
for Cell 1, with the model substantially underestimating outflow volumes for Cells 2 and 3, both of
which contain a submerged zone. Figure 4 indicates that for Cell 1, the modelled outflow volumes for
each event match the measured outflow volumes well, with the majority of outflow volumes,
particularly those of a large magnitude falling within a 25% error range. There was less success
however in modelling smaller events, where volumes are less than approximately 0.6 m3. At these
smaller events, although there were outflows from the biofilter observed, the model predicted no
outflows. It appears therefore that the model is not as reliable in predicting small flow event volumes.
This appears to be the case for Cells 2 and 3 also (Figure 4). However, from a design perspective this
is most likely not problematic, as such small flow events are unlikely to have a great impact on
receiving waters.
Figure 3: Measured and modelled total outflow volumes for study period (September 2010 – April
2011). Dotted lines indicate measured outflow volumes, and solid lines indicate modelled outflow
volumes.
7
Figure 4: Modelled event volumes against measured event volumes for k calibrated using E and φ
(dotted lines represent 25% error ranges). Note that there is only one figure for Cell 3 because the
hydraulic conductivity calibrated using E and φ were the same.
One cause of the error in the model is most likely the simplification that biofilters are a one
dimensional system. It is assumed that there is flow in only one direction (vertical) and that the inflows
are distributed evenly across the surface of the biofilter. However, in reality it is expected that there
would be both vertical and horizontal flows through the system, and that there would be greater flows
through the filter media near the inlet compared with the rest of the system.
Similarly, the assumption that biofilter properties such as hydraulic conductivity are constant not only
spatially, but also temporally is also expected to lead to errors. It has been found in previous studies
that the hydraulic conductivity of a biofiltration system can have high spatial and temporal variability
(Le Coustumer et al., 2009). Greater accuracy may be achieved if the spatial and temporal variability
of the hydraulic conductivity is taken into account.
8
The error in the model, particularly in Cells 2 and 3 could also be attributed to the fact that MUSIC
makes the assumption that the field capacity (sfc) of the three cells is 0.37 and 0.61 for systems with
and without submerged zones, and the water stress point (s*) is 0.22 and 0.37 respectively. These
parameters were determined using data obtained in a laboratory study of biofilter columns vegetated
with C. appressa and with a sandy loam filter media (Daly et al., 2009; Lintern et al., submitted).
Although Cell 1 had these properties (sandy loam filter media with C. appressa), Cells 2 and 3 did not.
It could be necessary to calibrate sfc and s* for a range of soil types to achieve greater model accuracy,
because theoretically, these parameters are dependent on soil texture (Arnell, 1999).
4
CONCLUSIONS
This study aimed to test a water flow model of biofiltration systems, presented in a previous study
(Lintern et al., submitted) by comparing the modelled results with observed data collected at the
Monash University carpark biofilter between September 2010 and April 2011. One of the input
parameters, the hydraulic conductivity of the biofilter was calibrated using two objective functions and
it was found that the choice of objective function significantly affects model results, with the
coefficient of efficiency favouring peak flows, and the unbiased function favouring lower flows. This
highlights that the objective function used during calibration should be selected depending on the
objectives of the modelling.
When the model was verified, it was found that although it is able to predict some hydrographs
accurately, there are a considerable number of events where the model does not predict the flow rates
or the duration accurately. This resulted in underestimations of total outflow volumes of 12 to 32%
when the hydraulic conductivity was calibrated using the coefficient of efficiency and 30 to 35% when
the hydraulic conductivity was calibrated using the unbiased function. This model could be considered
as a conservative model, as it always under-predicted the treated volumes. This is most likely due to
the simplifications made in the model; namely, the assumption that hydraulic conductivity term is
constant over the study period. The fact that the model performed better for Cell 1 than for Cells 2 and
3 could be due to the fact that the parameters in the model (sfc and s*) were found empirically for
systems with soil media that is different to that of Cells 2 and 3. Further work is required to modify the
model to take into account both temporal and spatial variability in hydraulic conductivity, and also to
have a range of values for sfc and s* that could be applied to systems with a range of soil media.
However, as a design tool, regardless of its simplifications, this model is useful because it is
conservative, easy to use and informs decisions in the conceptual design stage of a biofiltration
system.
5
ACKNOWLEDGEMENTS
This work was funded by the Victorian Government’s Science, Technology and Innovation grant
scheme, and a number of industry partners. The authors would also like to acknowledge the work of
Andrew Radion and Rory O’Gallagher in the collection of the hydraulic conductivity measurements at
the carpark biofilter.
9
6
REFERENCES
Arnell, N.W. (1999). A simple water balance model for the simulation of streamflow over a large
geographic domain. Journal of Hydrology 217(3-4), 314-355.
Blecken, G.-T., Zinger, Y., Deletic, A., Fletcher, T.D. and Viklander, M. (2009). Impact of a
submerged zone and a carbon source on heavy metal removal in stormwater biofilters.
Ecological Engineering 35(5), 769-778.
Bratieres, K., Fletcher, T.D., Deletic, A. and Zinger, Y. (2008). Nutrient and sediment removal by
stormwater biofilters: A large-scale design optimisation study. Water Research 42(14), 39303940.
Daly, E., Zinger, Y., Deletic, A. and Fletcher, T. (2009). A possible mechanism for soil moisture
biomodality in humid-land environments. Geophysical Research Letters 36, L07402.
Davis, A.P. (2007). Field performance of bioretention: water quality. Environmental Engineering
Science 24(8), 1048-1064.
Davis, A.P., Hunt, W.F., Traver, R.G. and Clar, M. (2009). Bioretention Technology: Overview of
Current Practice and Future Needs. Journal of Environmental Engineering 135(3), 109-117.
Hatt, B.E., Fletcher, T.D. and Deletic, A. (2009). Hydrologic and pollutant removal performance of
stormwater biofiltration systems at the field scale. Journal of Hydrology 365(3-4), 310-321.
Le Coustumer, S., Fletcher, T.D., Deletic, A., Barraud, S. and Lewis, J.F. (2009). Hydraulic
performance of biofilter systems for stormwater management: Influences of design and
operation. Journal of Hydrology 376, 16-23.
Le Coustumer, S., Fletcher, T.D., Deletic, A., Barraud, S. and Poelsma, P.J. (in press). The influence
of design parameters on clogging of stormwater biofilters: a large-scale column study. Water
Research.
Lintern, A., Daly, E., Hatt, B.E., Fletcher, T.D. and Deletic, A. (submitted). Presentation of a lumped
stormwater biofiltration model - Part II: testing of the hydrologic and treatment model.
McCarthy, D.T., Deletic, A., Mitchell, V.G. and Diaper, C. (2011). Development and testing of a
model for Micro-Organism Prediction in Urban Stormwater (MOPUS). Journal of Hydrology
209, 236-247.
Nash, J.E. and Sutcliffe, J.V. (1970). River flow forecasting through conceptual models part I - A
discussion of principles. Journal of Hydrology 10, 282-290.
Paul, M.J. and Meyer, J.L. (2001). Streams in the urban landscape. Annu. Rev. Ecol. Syst. 32, 333-365.
Roy, A.H., Wenger, S.J., Fletcher, T.D., Walsh, C.J., Ladson, A.R., Shuster, W.D., Thurston, H.W.
and Brown, R.R. (2008). Impediments and solutions to sustainable, watershed-scale urban
stormwater management: Lessons from Australia and the United States. Environmental
Management 42(2), 344-359.
Trowsdale, S.A. and Simcock, R. (2011). Urban stormwater treatment using bioretention. Journal of
Hydrology 397(3-4), 167-174.
Walsh, C.J., Leonard, A.W., Ladson, A.R. and Fletcher, T.D. (2004). Urban stormwater and the
ecology of streams, Melbourne Water Studies Centre, CRC for Freshwater Ecology, CRC for
Catchment Hydrology.
Zhang, Z., Rengel, Z., Liaghati, T., Antoniette, T. and Meney, K. (2011). Influence of plant species
and submerged zone with carbon addition on nutrient removal in stormwater biofilter.
Ecological Engineering 27, 1833-1841.
10
Zinger, Y., Fletcher, T.D., Deletic, A., Blecken, G.T. and Viklander, M. (2007): Optimisation of the
nitrogen retention capacity of stormwater biofiltration systems. In: Novatech 2007, 6th
International Conference on Sustainable Techniques and Strategies in Urban Water
Management, Lyon, France, June 25-28, 2007.
11