Beruflich Dokumente
Kultur Dokumente
Ecological Engineering
journal homepage: www.elsevier.com/locate/ecoleng
A R T I C L E I N F O A B S T R A C T
Article history: During the last two decades a couple of models were developed for constructed wetlands with differing
Received 20 March 2014 purposes. Meanwhile the usage of this kind of tool is generally accepted, but the misuse of the models still
Received in revised form 3 October 2014 conrms the skepticism. Generally some groups of models can be distinguished: on one hand
Accepted 8 October 2014
mechanistic models try to display the complex and diffuse interaction of occurring processes, on the
Available online 14 January 2015
other hand the same kind of models are used to investigate single processes. New kinds of simplied
approaches well appreciated by engineers try to display system performances without going to deep
Keywords:
into details. All types of models are valuable some more for scientic usage, others more for
Constructed wetlands
Model
engineering. The given summary tries to support potential users in taking the right choice in model
Modelling selection. Big differences can be found in the model availabilities. Whereas some of the compared
Simulation software packages are purchasable without limitation, some others are only accessible on their platform
Review level, and some can be seen as exclusive property. From the experience of the authors it can be
Comparison summarized, that research groups starting modelling/simulation studies should be encouraged to use the
given knowledge before starting from scratch again.
2014 Elsevier B.V. All rights reserved.
1. Introduction
http://dx.doi.org/10.1016/j.ecoleng.2014.10.031
0925-8574/ 2014 Elsevier B.V. All rights reserved.
206 D. Meyer et al. / Ecological Engineering 80 (2015) 205213
of each category of pollutant is typically associated with a specic levels: the level of knowledge of the process considered, the
microbial functional group. Pioneering modelling research works complexity associated to the model and its spatial resolution, and the
have employed circumstantial evidence (black box approach) to resources necessary to the usage of the reviewed models.
compute these basic assumptions into models mostly for domestic In 2013, in the frame of the 5th International Symposium on
wastewater. Then mechanistic models have been developed trying Wetland Pollutant Dynamics and Control, WETPOL 2013, a wide
to consider the hydrodynamic and the biodegradation processes range of recent simulation and modelling studies were
occurring in CW systems at the same time for a larger range of presented. Most of the concepts initially applied to CWs for
sources of wastewater (industrial, diffuse, surface runoffs . . . ) pre-treated domestic wastewater have been developed, adapted
The need for CW models can be summarized as follow: and applied to more and more case studies including for (raw)
domestic wastewater, combined sewer overows (CSOs) and
- Describing the phenomena ongoing in a CW system (e.g., water intensied CWs. All publications from the WETPOL conference
ow, adsorption, O2 transfer). dealing with modelling and simulation were selected for this
- Using models as a tool to compare 2 similar systems and their review, and all main authors contributed to this paper.
behavior under different conditions (e.g., effect of loading The concepts selected in this paper mostly deal with three
characteristics, effect of plant species, effect of season, etc.). main approaches that can be ranked as follows:
- Predicting the performances of a given system (e.g., developing
design recommendations). a Biokinetic models: the most advanced models using saturated
- Answering what if? questions (e.g., over or under loads, etc.). water ow are those developed by Rousseau (2005) and
- Performing system control (especially in case of system Langergraber et al. (2009), both considering bio-kinetic models
intensication like the use of articial aeration, recycling etc.). that are based on the IWA Activated Sludge Models (ASMs)
(Henze et al., 2000). For modelling vertical ow CWs with
In this context an increasing number of publications dealing with intermittent loading, transient variably-saturated ow
CW modelling has been observed over the last 3 years (20112013) as models are required. These systems are highly dynamic, adding
it represents more than 38% of the number of papers ever published requirements to the complexity of the overall system. The most
containing the words constructed/treatment wetland and model/ advanced reaction models are implemented in the Wetland
modelling in the title (Science Direct research engine). Module of the HYDRUS software package (Langergraber and
Among the most cited review papers, most of the most imunek, 2005), based again on the mathematical formulation
common modelling approaches of CWs treating domestic of the ASMs (Henze et al., 2000).
wastewater, including biokinetic models and process models, b Process dedicated models, which rely on simple kinetics
have been reported in Rousseau et al., 2004; Marsili-Libelli and employed to model a single process related to the degrada-
Checchi, (2005); Langergraber (2008); Langergraber et al. tion/transfer of one compound or one family of compound (e.g.,
(2009) and more recently Kumar and Zhao (2011). COD, NTK, O2 etc.).
Although the previous review efforts were performed to establish c Design support models such as GPS-X, enabling to model the
a non-exhaustive list of all the modelling approaches successfully system according to an initial Residence Time Distribution
developed (including their main results), this review paper aims at analysis; or RSF_Sim for the specic context of combined sewer
describing a sort of guideline to help users to choose the most overow treatment.
appropriated modelling approach adapted to their needs at different
Table 1
Reviewed modelling and simulation studies overview.
After a description on how the three main modelling take into account the carrying capacity of the system and the
approaches are implemented and computed, this paper will presence of accumulated inert solids on the growth rate of bacteria
describe the main reasons to choose one modelling strategy rather (Sams and Garca, 2014).
than another, nally the paper will compare the pros and cons of The biokinetic equation set has to be coupled with water ow,
each modelling strategy. solute and heat transport models to simulate the operation of CWs.
This is done in HYDRUS by solving the Richards equation
2. Material and methods describing capillary-driven ow in variably-saturated porous
media and equations describing convectivedispersive transport
2.1. Description of models and simulation studies of solute and heat and diffusion in the gaseous phase. Another sub-
model of HYDRUS is the root model describing water and solute
Modelling begins with ideas and concepts for simplied
uptake and oxygen release (imunek et al., 2012).
descriptions of the real world. Publications usually deal with Regarding BIO_PORE, the ux of water is described by
new mathematical descriptions (models) and/or with model combining the Darcy's equation with a mass-balance equation
applications (simulation). On conferences like Wetland Pollutant to determine the exact location of the water table at each time. This
Dynamics and Control (WETPOL) both are often put together in one strategy was adopted to avoid the growth of bacteria in the
session, but for a comparative review a separation is required. In a unsaturated areas of the bed. All these models were spatially
rst step of comparison, all participating publications were discretized in a two-dimensional domain.
assorted in three differing groups: The rst group contains Among the ve studies carried out using the two described
applications of generally agreed biokinetic reaction models. The mechanistic models, Morvannou et al. (2014) as well as Plfy and
second group is dedicated to models focusing on single processes Langergraber (2013) employed CW2D in HYDRUS, Rizzo et al.
or phenomena within the diffuse and complex interactions of real (2014) as well as Plfy and Langergraber (2014) employed CWM1
processes in CWs. The third group was chosen for a new generation in HYDRUS, and Sams and Garca (2013a,b); Sams and Garca
of modelling approaches aiming more at engineering than on (2013a,b) presented CWM1 in BIO_PORE.
science. Plfy and Langergraber (2013) aimed at designing a HFCW for
Table 1 gives an overview of the studies contributing to the polishing treatment of the outlet of a sequencing batch reactor
given review, including abbreviations. Short descriptions of the (SBR). Despite it was used to simulate a HFCW, CW2D was preferred
used models and the according studies presented during WETPOL over CWM1 as it includes phosphorus removal processes. Further-
2013 (or published in journals previously) can be found thereafter. more if the organic load is low, the lter will remain in anoxic
conditions justifying the choice to be put on CW2D.
2.1.1. Group 1: CWM1 and CW2D in HYDRUS and BIO_PORE Morvannou et al. (2014) applied CW2D to a French-type VFCW
The HYDRUS Wetland Module and BIO_PORE are general (Molle et al., 2005). Particular emphasis was given to the
mechanistic models and both use biokinetic models inspired by modelling of ammonium removal. Therefore linear adsorption
the Activated Sludge Models (ASMs) for simulating biological was taken into account with a value changing according to the
transformation and degradation processes and microbial dynamics organic matter content. Furthermore one biokinetic parameter
in subsurface ow constructed wetlands: the Constructed Wetland was estimated based on a solid respirometric method (Morvan-
Model 2D (CW2D) (Langergraber and imunek, 2005) and the nou et al., 2011). The model was calibrated over a two-week
Constructed Wetland Model No. 1 (CWM1) (Langergraber et al., dataset including continuous measurements of inlet and outlet
2009). Both CWM1 and CW2D are implemented in the HYDRUS pollutant concentrations.
software (Langergraber and imunek, 2011) while a modied Plfy and Langergraber (2014) presented simulations of water-
version of CWM1 is implemented in the COMSOL MultiphysicsTM saturated batch operated mesocosms. The aim was to verify the
platform for the BIO_PORE model (Sams and Garca, 2013a). biokinetic sub-model only, the CWM1 implementation in HYDRUS.
CWM1 and CW2D include the biochemical degradation and Using data from controlled greenhouse column experiments, the
transformation processes for organic matter and nitrogen and effects linked to water ow, heat transport and other processes
additionally for phosphorus in CW2D and sulfur in CWM1 with the occurring in an uncontrolled environment were excluded. The
main objective to predict efuent concentrations. CW2D is suitable numerical results were calibrated on experimental data by
to the VFCWs modelling as it considers aerobic and anoxic biokinetic iteratively adjusting the biokinetic parameters. The results were
processes. In addition, CWM1 considers the anaerobic biokinetic evaluated by a new goodness of t analysis method, the deection
processes and therefore it is appropriate for both HF and VF CWs. analysis.
The CW2D requires a total of 46 parameters to solve the Based on HFCW experimental results (Galvo and Matos, 2012),
equations describing 9 different processes. For each element of the Rizzo et al. (2014) explored the capability of the HYDRUS Wetland
model, 12 variables are calculated: 9 for the liquid phase and 3 for Module to model the response of HFCW to unsteady load. In
the biomass which is immobile and therefore exclusively addition to the biokinetic processes in CWM1, Rizzo et al. (2014)
associated with the solid phase (heterotrophic and two autotro- included ammonium adsorption using a Langmuir isotherm.
phic bacteria groups). Qualitative comparisons were carried out between simulated
The CWM1 is more complex as it also takes anaerobic reactions and measured efuent concentrations of COD and NH4-N by
into account. It considers 17 processes and 16 components calculating the mean percentage error.
(8 soluble and 8 particulate out of which 6 represent bacteria The BIO_PORE model was calibrated using data from the rst
groups). It requires a total of 65 parameters. operation year of a pilot-scale HFCW (Sams and Garca, 2013a).
As a distinctive feature from the original CWM1s formulation, Then the simulation was extended over a 3-years period in order to
slowly biodegradable and inert particulate COD are divided into qualitatively describe the bacteria communities distribution and
aqueous and solid phases in BIO_PORE which allows describing dynamics (Sams and Garca, 2013b). In Sams and Garca (2014),
their retention and re-suspension. Moreover, the unlimited and the necessity of limiting the growth of bacterial communities
unrealistic exponential growth of biomass affecting most of the when using ASM-based models is justied and a theory to describe
available CW models based on the ASM formulation is avoided in the most basic functioning patterns of HFCWs, namely The
BIO_PORE model by including two logistic functions in the growth Cartridge Theory, is presented.
expressions of all functional bacterial groups. These two functions
208 D. Meyer et al. / Ecological Engineering 80 (2015) 205213
2.1.2. Group 2: process-dedicated models lter were taken into consideration, namely, dispersion, sedimen-
Besides displaying process interactions, mechanistic models tation and adsorption of particles. Sedimentation process can be
can also be used to gain in-depth understanding of a particular divided into three major mechanisms: settling and deposition,
phenomenon. Some researchers carried out specic simulations aggregation of SS via coagulation, and mass exchange between SS
for this purpose: Petitjean et al. (2011, 2012) and Forquet et al. and the dissolved phase. Because of its relatively slow rate the
(2009a,b) investigated the inuence of the air phase on latter can be treated separately and is not taken into account in the
hydrodynamics (Diph_M), Morvannou et al. (2013) focused on current model.
non-equilibrium ow in French-type VFCW with the dual-porosity
model (DPM), Claveau-Mallet et al. (2012, 2014) studied the 2.1.3. Group 3: design support models
phosphorus removal by precipitation in steel slag lters (P- RTD/GPS-X employs a two-step model coupling hydraulic and
hydroslag) and Sani et al. (2013) investigated particulate transport kinetic modelling on xed bed aerated biolm reactors. Despite it
(WangScholz-Model). relies on the ASM1 model for biokinetic reactions which clearly is a
The new model called Diph_M was used to describe the mechanistic model, we choose to classify it among design support
interplay between gaseous oxygen and bacteria during the models because of the chosen hydraulic model. It is based on
feeding of a second-stage vertical-ow sand lter. Under Residence Time Distribution analysis (tracer injection and detec-
partly-saturated conditions, feeding involving ponding leads to tion) carried out on different steps of biolm development in the
water ow disturbance by the air phase. In order to overcome this reactor. Based on the shape of the breakthrough curves, a hydraulic
issue, Forquet et al. (2009a,b) implemented a 1D numerical model model is chosen (among models like Plug Flow, Tank in Series, Tank
in MATLAB which solves simultaneously the Richards equation in Series with exchange) regarding saturation conditions and
for both the air and the water phases. Aside from improving the biolm development (Zeng et al., 2013a). This rst step aims at
hydrodynamic description of the lter, this model permits an determining hydraulic parameters like number of tanks in series,
accurate estimation of the gaseous uxes within the lter. type of tanks (plug ow or perfectly mixed), average retention
Petitjean et al. (2011, 2012) coupled a hydrodynamic set of time, dead volume, recycle ratio and actual volume of the reactor.
equations solving reactive transport according to the These parameters are then implemented in GPS-X software to
ASM1 model. simulate biodegradation processes (Zeng et al., 2013b). The
The dual-porosity model DPM was developed to simulate the biokinetic model chosen in GPS-X is a 2-step nitrogen model
preferential ows occurring in gravel lters. A dual-porosity based on ASM in a biolm reactor with biolm diffusion fraction
hydraulic model (imunek and van Genuchten, 2008) was chosen for substrates (Vigne et al., 2010). It is then calibrated with
in Hydrus-1D (imunek et al., 2013) with a mass-exchange experimental kinetic parameters from different heights of the
coefcient proportional to the water content. The model assumes xed bed biolm reactor. Finally this model simulates TOC removal
that the water phase is divided between a mobile part (in our case and several NH4-N conversion processes taking into account
the macroscopic pores) and an immobile part (water contained in biolm development, oxygen concentration, water saturation level
the organic matter matrix and also some capillary water). The ow and height of the biolter.
only occurs in the mobile part but exchanges are possible between Meyer and Dittmer (2014) presented the model RSF_Sim
the two water parts. The model was tted versus a tracer test which was created in order to estimate long-term treatment
experiment carried out on a French rst-stage VFCW. performance of constructed wetlands for combined sewer
The conceptual model used in P-hydroslag focuses on tertiary overow treatment (CSO-CWs), also known as Retention Soil
treatment with enhanced requirements of phosphorus removal. It Filters (RSFs). The development of RSF_Sim (Meyer and Dittmer,
is based on slag dissolution and hydroxyapatite (Ca5(PO4)3OH) 2014) was based on a fundamental research study (Dittmer and
crystallization rates. It included three steps: (i) dissolution of slag Schmitt, 2011), but also on experiences in the application of
as Calcium oxide (CaO), (ii) phosphorus precipitation and crystal HYDRUS-2D/CW2D (Dittmer et al., 2005; Henrichs et al., 2007,
growth in alkaline conditions, (iii) evolution of slag lter 2009; Meyer at al., 2006; Plfy et al., 2014). CW2D was generally
properties (slag dissolution rate, porosity and hydraulic conduc- applicable to simulate laboratory RSF column operation, but
tivity) related to slag exhaustion and to the presence of crystals extended dry periods exceeded the ability of reasonable nutrient/
(Claveau-Mallet et al., 2012). The conceptual model was biomass balances due to the missing description of particulate
mathematically expressed in a 1D geochemical model including organic matter degradation.
equilibrium and rates (Claveau-Mallet et al., 2014). The software RSF_Sim works with three complete stirred tanks in vertical
PHREEQC was used for numerical simulations. The slag dissolu- series: [1] the retention layer provides the water storage on top of
tion equations were determined from a simple experimental the process layer (ponding), [2] the process layer describes the
batch test. Numerical simulations of a 1D lter were run and the sand/gravel layer (saturated during feeding, drained afterwards) in
evolution of phosphorus concentration at the outlet of the system which treatment occurs, [3] the drainage layer improves the
was computed. volume balances (German model version), but can also be used as a
The Civil Engineering Research Group at The University of water reservoir under permanent saturation (French model
Salford used a 1-D model for particulate transport in a VFCW version) (Meyer et al., 2013).
implemented in COMSOL MultiphysicsTM. It has initially been Up to date, descriptions of treatment performances for COD
introduced by Sani et al. (2013). Wastewater contains waste and NH4-N are kept basically simple. The total COD needs to be
particles of different size and composition. As these particles build separated into two fractions: particulate COD becomes reduced
up within the lters, the efciency of subsequent removal is by ltration (down to a background concentration), dissolved
reduced due to the decrease in pore space. Some particles are COD is reduced by a treatment efciency factor, which can vary
electro-statically attracted and stack onto each other, often due to inuencing factors like temperature, outow limitation
forming dendrites. These dendrites block the pore space, thus rates or the duration of previous dry periods (most important).
increasing particle interception. The proposed WangScholz The retention of NH4-N during the lter passage is calculated
Model assumes a uniform water velocity throughout a saturated with a steady-state two-stage linear isotherm, nitrication
column, which was experimentally estimated using Darcys law. is considered only after re-aeration by 1st order kinetics
The model aims at estimating particle settling in the porous media. (Meyer et al., 2014).
Three major mechanisms that affect the particle transport in the
D. Meyer et al. / Ecological Engineering 80 (2015) 205213 209
2.2. Model comparison presented at the 2013 WETPOL symposium can be mainly found
in the category of PDE models at the exception of RSF_Sim which is
In this section, the aim was not to provide performance basically tank-in-series description.
comparison based on benchmark tests but rather to provide
guidance so that a new user may choose the appropriate model for 3.2. Fields and limits of application
simulating constructed wetland behavior. We proposed a three-
step method based on (i) the type of constructed wetland the user 3.2.1. Group 1: the HYDRUS Wetland Module and BIO_PORE
is interested in, (ii) the objectives of the study, and (iii) the On one hand the HYDRUS Wetland Module might be too
resources dedicated to modelling. complex to be directly used for design purposes (Table 2), meaning
In the rst step, we associated the models with different types that the money and effort invested into it cannot be easily justied.
of constructed wetlands, themselves classied in two categories: In some countries standards are limiting the design to be based on
saturated or variably-saturated, similarly to what was suggested by rules of thumbs, e.g. the Austrian standard, NORM (2009),
Fonder and Headley (2013). making the design-purpose use of numerical models with such
The second step allows classifying the models based on the user complexity impossible. For these reasons the HYDRUS Wetland
goals. The rst criterion is whether the study aims at understand- Module can be primarily suggested for research purposes and for
ing the processes or at designing new systems. It also gives an idea creating simplied models based on them. There will always be
about the complexity associated with the model and its spatial options to test new elds of application on the edge or outside of
resolution. The second criterion is whether short or long-term the original scopes. For instance, Rizzo et al. (2014) have
simulation will be performed. demonstrated that the HYDRUS Wetland Module with CWM1 is
Finally, the user must take into account the resources necessary able to adequately simulate performances of laboratory horizontal
to the usage of the reviewed models. They are expressed in terms of ow CWs loaded with articial wastewater that are subjected to
the user training level, the price of the model software or of the unsteady loads using data from Galvao and Matos (2012). Also the
modelling platform, and the level of development of the model and studies using CW2D for simulation of CSO-CWs (Dittmer et al.,
its associated documentation (since it requires additional efforts to 2005; Meyer et al., 2006; Henrichs et al., 2007, 2009) and for
work with an uncompleted or poorly documented model). simulating nitrate dynamics in a rural headwater catchment
(Smethurst et al., 2013) exceeded the original scope of the model.
3. Results of model comparison On the other hand, Plfy and Langergraber (2014) demonstrated a
case in which the HYDRUS Wetland Module with CW2D facilitated
3.1. Classication the design of a HF CW for polishing the efuent of a sequencing batch
reactor with 700 m3 per day peak capacity. This suggests that the
Models and simulation studies participating to this review scale of the project and the characteristics of the wastewater might
show a broad bandwidth of aims and scopes. In order to summarize justify the use of complex tools similar to CW2D or CWM1.
the ndings of comparisons it seems to be important to classify the Sams and Garca (2014) pointed out possible weak points of
tools and their applications. Modelling approaches range from a the equations describing biokinetics in ASM-based models
simple regression to a partial differential equations model (PDE). (including CWM1 and CW2D) and showed that the BIO_PORE
Description of these has already been discussed in previous papers model also includes the possibility to simulate the accumulation of
(Kumar and Zhao, 2011; Langergraber et al., 2009). Models inert organics in long term scenarios.
Table 2
Model comparison overview.
Similar to the HYDRUS Wetland Module, the BIO_PORE can be explained by the fact that the model does not account for
software (Sams and Garca, 2013a) is dedicated to (a) simulate biological growth and decay of plant matter. Nevertheless, the
the hydraulics and hydrodynamics of wetlands(Garca et al. 2004), model is also useful to distinguish between physical and
(b) simulate the contribution of plants to the treatment of water, biochemical process inuences.
(c) predict efuent pollutant concentrations in HFCWs, (d) study The diphasic model Diph_M is a research oriented type of
the location and dynamics of bacterial communities in HFCWs, and mechanistic model. It is focusing on single process understanding
e) improve the understanding of wetlands. BIO_PORE was not (2nd stage kinds of VFCWs), without any aim of sizing or even
developed for design purposes, but it supports understanding of treatment yields estimation. It was validated separately for ow
the functioning. A special emphasis is given on the prediction of and transport phenomena using experimental data. Results gave
failures, which correspond to the time in which bacterial better insights on oxygen transport inside VFCWs (Forquet et al.,
communities are pushed that much towards the outlet by the 2009a,b; Petitjean et al., 2012). Apart from that, the benets of
accumulation of inert solids that they are not able to consume the development can be found in a long-term perspective when this
desirable proportion of inuent pollutants (cartridge theory, specic part will be implemented into more general mechanistic
Sams and Garca, 2014). The main weakness of BIO_PORE can models. Right now it is used to evaluate the prominence of
be found in the missing link between biolm growth and changes processes as a sound tool for the study of systems in which strongly
in hydraulics, because it still does not include clogging. Not more interdependent phenomena are occurring.
than 3 to 5 years (after which clogging is generally more evident) The P-hydroslag model shall be used to (a) predict P
can be simulated with certain guarantees. concentration evolution at the outlet of a slag lter, (b) predict
Although it is not yet implemented in BIO_PORE, Sams (2014) concentration evolution of other compounds of interest (pH, Ca,
presents a mathematical formulation to simulate the effects of etc), (c) predict the longevity [d] and the retention capacity [mg P/
clogging in the hydraulics and hydrodynamics of CWs, which also g slag] of a given slag lter, and (d) evaluate the inuence of
allows simulating the runoff generated as a result of severe clogging. operating parameters as the type of slag, hydraulic retention, type
In the application of biokinetic reaction models like CWM1 or of voids and the composition of wastewater on the performance
CW2D, a general limitation is the lack of proper data to verify and longevity of the system. Limits of this rst approach can be
simulation results. For instance, Rizzo et al. (2014) highlighted that found in the fact that the actual model does not consider porosity
acetate was found as the main efuent COD component during peak reduction of the lter related to precipitate accumulation. It also
inuent events. These results could not be validated due to the lack of does not consider the limiting effect of crystal accumulation on
experimental data for COD fractionation. However, although the slag dissolution. Accordingly the actual model probably over-
experiments by Galvao and Matos (2012) have not been designed estimates longevity and retention capacity.
with a modelling exercise in mind, Rizzo et al. (2014) could simulate In order to reach a high number of users, the P-hydroslag the
their results. This study also showed that there is a need for planning model is reasonably complex as it involves simplications which
experiments in a different way (e.g. sampling frequency, parameters are: (i) the wide variability of slag composition resulting in
analyzed) when dynamic models should be applied. different reactivity is simplied by a CaO dissolution reaction with
a rate decreasing over time, (ii) precipitation rates are simplied to
basis expressions with good experimental tting, and (iii) the
3.2.2. Group 2: process-dedicated models
initial model considered only hydroxyapatite, but its accuracy was
In contrast to black box approaches such as multiple
greatly improved when monetite and its transformation to
regression modelling, principal component analysis, redundancy
hydroxyapatite were added (Claveau-Mallet et al., 2014).
analysis, and the self-organizing map modelling (Dong et al., 2012),
The dual-porosity model (DPM) uses a larger number of
process-dedicated models offer a better understanding of the
hydraulic parameters than the simple hydraulic model (equilib-
underlying science within a wetland. For example, within the
rium model) in HYDRUS-1D, which makes it more complex to use.
WangScholz Model, the particle sedimentation process and its
Moreover, among those parameters, some are difcult or even
effect on clogging can be modeled predominantly based on SS
impossible to measure. Only a calibration carried out from
accumulation inside the CWs.
outow data combined with tracer tests as well as measurements
Clogging processes within constructed wetlands are complex.
of water contents and pressure heads within the lter allows their
They are difcult to be quantied by representative values without
determination. In DPM, no ow in the immobile phase is
destroying the wetland lter or interfering with its operation.
considered. Only exchanges between mobile and immobile
Values obtained from the literature have been used for parameters
phases are taken into account. The dual-permeability model
where no measurements were available. The experimental CWs do
simulates ows in both phases, but it involves a larger number of
not operate in continuous inow mode, because they are drained
hydraulic parameters to be calibrated, which further complicates
and relled at xed intervals. For simplication, a pseudo
their determination. On the other hand, as the water ow through
convection velocity of wastewater is assumed to represent the
a rst-stage of French VFCW is mainly gravity-driven (in the
average rate of the water change in the whole year.
macro pores), it may be preferable to turn towards models where
The WangScholz Model has been applied to simulate the SS
preferential ows are described by the Kinematic Wave equation
sedimentation processes within experimental CWs. The relatively
(Larsbo and Jarvis, 2003). These models are tailored to soils with
good results (Sani et al., 2013) demonstrate that the four
macro pores or heavy clay soils with distinct shrinkage cracks.
mechanisms (i.e., diffusion, convection, sedimentation and
pebble surface adsorption) may play the main role in the SS
transport in vertical-ow wetland systems. It can be seen from 3.2.3. Group 3: design support models
the characterized modelling parameters that the pebble surface The RTD/GPS-X model (Soric et al., 2013) allows (a) predicting C
adsorption takes the most prominent effect on SS ltering or and N concentrations at different depths of the biolters, (b)
interception. An average approach ignoring the varied particle predicting actual and dead volumes evolution i.e. hydraulic
sizes and the discontinuous inow works effectively for behavior taking into account biolm development and saturation
modelling the water treatment using small scale vertical-ow conditions, and (c) evaluating the inuence of hydraulic loading
wetland systems. However, the WangScholz Model consistently rate on TOC and TN removal. By this, the model aims to be used to
under-predicts the SS concentrations within the litter zones. This improve wetland design and operation. So far, RTD/GPS-X has been
D. Meyer et al. / Ecological Engineering 80 (2015) 205213 211
the low level of complexity can be seen as a main quality. In a simple and free numerical tool (PHREEQC). Currently the
this way, an increased complexity would counter the model application for any kind of alkaline slag is under approval. In future,
purpose. On the other hand, a more detailed background based the calibration on controlled slag lters and later on real lters is
on stochastical data analysisnot affecting the usabilitywould scheduled in preparation to implementation of P-hydroslag into 2D
give benets to the application. lter models.
RSF_Sim was developed for the German type of CSO-CWs.
4.2. Current elds of model improvement and extended application Currently it is under validation for the French system (Meyer et al.,
2013). Especially the role of a permanently saturated water layer
The latest verication of the CWM1 implementation in the needs to be investigated in real operation rst, and implemented
HYDRUS Wetland Module (Plfy and Langergraber, 2013) needed into the model afterwards. The feeding with raw CSO as a further
three parameters to be adjusted for each experimental column: main difference might have an inuence on the treatment process
(1) initial bacterial concentrations, (2) initial adsorbed NH4-N description. For both of the CSO-CW systems a horizontal
concentrations, and (3) cRoot, the parameter determining discretization of the lter media layer is needed to identify the
root re-aeration rate. Bacterium concentrations have a great effects and importance of shortcut ows. In general, a collection of
effect on wetland performance. A method which can measure the available input data is in progress in order to increase the reliability
concentration of active bacteria in microcosms and link the of performance predictions.
activity to groups identical to those found in process-based In RTD/GPS-X it is scheduled to determine the biolm thickness
numerical models could help to decrease uncertainty around with more accuracy. Also kinetic parameters of N removal
biokinetic parameter values and changes in bacterium commu- processes (anoxic), gas emission of the reactor for mass balance
nities. accuracy, and nally non-soluble substrate and inert particles shall
Results from Rizzo et al. (2014) suggest that the HYDRUS be taken into account for full-scale application.
Wetland Module is a suitable tool to study HFCWs under time-
variable ows, opening new possible future applications. Among 5. Conclusions and recommendations
these, the stochastic behavior of HFCW outows is of great interest.
Indeed, the randomness of HF CW efuent is generally neglected During the last two decades a couple of models were
during the typical design procedure, and the HFCW size should be developed for CWs with differing purposes. Meanwhile the
highly overestimated to face real outow variability (Kadlec and usage of this kind of tool is generally accepted, but the misuse of
Wallace, 2009). In this context, HFCW efuent stochastics could be the models still conrms the skepticism. Generally three groups
simulated by HYDRUS-CWM1. For instance, a HFCW could be of models can be distinguished: (1) On one hand mechanistic
forced by stochastic inuent loads modeled as noisy signals (Ridol models try to display the complex and diffuse interaction of
et al., 2011), allowing to investigate if the randomness of efuent is occurring processes, (2) on the other hand the same kind of
driven by the variability of inows. Moreover, users could models are used to investigate single processes. New kinds of
investigate if the implementation of noisy signal technique could simplied approaches (3)well appreciated by engineers - try to
improve the HFCW design procedure. display system performances without going to deep into details.
Needs for further development of the HYDRUS wetland module All types are valuablesome more for scientic usage, others
can be found in a clogging model (particle transport and inuence more for engineeringand all support their users, but all require
on hydraulic properties), in a dual-porosity ow model coupled skills of the users.
with Wetland Module to simulate preferential ow, and nally in Big differences can be found in the model availabilities. Whereas
active and passive uptake functions of HYDRUS to be made some of the compared software packages are purchasable without
available in the HYDRUS Wetland Module. limitation, some others are only accessible on their platform level,
In BIO_PORE, the biokinetic models are considered to be the and some can be seen as secret property. From the experience of the
most important part, just as bacteria are the most important authors it can be summarized, that research groups starting
component on CWs. Therefore better CWs require better biokinetic modelling/simulation studies usually tend to either buy a software
models. In this regard Sams and Garca (2013a,b) believe that a package or to start from scratch with their own ideas. Adaptations of
single biokinetic model would help decrease model divergence existing approaches or tool sharing are uncommon. This leads to a
and concentrate efforts on a single platform, just as in ASMs waste of time, energy and moneyboth because modelling is time-
(Sams et al., 2014). In future, special emphasis will be put on consuming and older approaches start to disappear immediately
clogging, micro-scale bacterial models, and chemical reactions after their developers focus on other topics.
For Diph_M the input parameter estimation is currently under In order to nd the best simulation tool for specic needs, the
investigation. The model should be implemented into general CW authors would like to recommend the following procedure:
models when VFCW are operated with ponding. Future develop-
ments on mechanistic models should take the inuence of biomass 1. Dene tasks and challenges and do a review in order to nd
growth on hydraulics into account; similar to like it was described similar cases.
by Sams and Garca (2014) for HFCW. 2. Get into contact with the model developers and check the
The Wang-Scholz Model would benet from a long-term availability of approaches.
modelling study of a real wetland becoming more mature until it 3. Try to use older models and nd out their limits (Fig.1).
requires decommissioning, which might, however, take as long as 4. Create ideas for improvements and share them.
20 years. As an improvement of the model itself, the microbial
removal processes responsible for ammonia-nitrogen, nitrate- In addition to the accessibility of tools, simulation studies often
nitrogen and plant-originating SS reduction should be imple- suffer from a lack of data. When the use of dynamic models is
mented. However, access to long-term wetland system data is aimed at the experiments have to be planned in a way that data for
rather unrealistic considering that most research projects are model calibration and validation are available. Data requirements
based on up to three years of data (Dong et al., 2012). are different for different models but include careful planning of
The P-hydroslag model was originally based on dissolution and sampling frequency and parameters analyzed.
precipitation rates that lead to prediction of slag lter longevity and Additionally, scientic institutions as well as private enterprises
retention capacity, the methodology based on a simple batch test and usually publish their best results of research or technology
D. Meyer et al. / Ecological Engineering 80 (2015) 205213 213