Industrial applications of the IWA anaerobic digestion

Water Science and Technology Vol 47 No 12 pp 199–206 © 2003 IWA Publishing and the authors
model No. 1 (ADM1)
D.J. Batstone* and J. Keller**
* Department of Biotechnology, Lund University, Sweden (E-mail: damien.batstone@biotek.lu.se)
** Advanced Wastewater Management Centre, The University of Queensland, Australia (E-mail:
j.keller@awmc.uq.edu.au)

Abstract In this paper, the IWA anaerobic digestion model No. 1 (ADM1) is applied to two case studies
from contract work on industrial treatment plants. The first was the assessment of acid addition for pH
decrease and avoidance of calcium carbonate (CaCO3) precipitation in a paper mill fed UASB. The
simulation work found, with a high degree of confidence, that acid dosing was neither economical for pH
control, nor had any real effect on the CaCO3 levels present in the reactor. A specific calcium carbonate
precipitation equation was added to the ADM1 to undertake this study. The second case study was an
assessment of the benefits of thermophilic (as opposed to mesophilic operation) for reduced ammonia
inhibition, improved stability and gas production in a solids digester at a gelatine production facility. Here, it
was predicted that thermophilic operation could not attain either goal to a satisfactory extent. In addition to
demonstrating the application of the ADM1 to the two systems, we have also assessed the predictions
generated in the case studies in terms of quality and utility.
Keywords ADM1; application; calcium carbonate; model; precipitation

Introduction
In the last 30 years, there has been an increased level of comprehension of anaerobic
processes, and their representation in mathematical models. However, practical, industrial
use of the models have been relatively low. One of the reasons for this may be the wide
variety of models, and their often very specific nature. This is unfortunate, as a well defined
model is highly effective for assessment, optimisation, and design of existing and proposed
processes and process modifications. One of the objectives of the IWA Task Group for
Anaerobic Digestion Modelling was to increase accessibility to and application of simula-
tions by consultants and practitioners. This was achieved by presenting a more generic
model with consistent nomenclature, units and equation structure. The result of the task
group’s work, The Scientific and Technical Report (STR) describing the IWA anaerobic
digestion model No. 1 (ADM1) has been published by the IWA (Batstone et al., 2002).
Other reasons for the limited use of modelling and simulation in non-academic arenas
include the lack of implementation in widely used software packages, lack of clearly
detailed methods for implementation and application of the models, low numbers of case
studies, and insufficient demonstration of the commercial and intangible benefits available
from modelling. In this paper we address some of these limitations by applying the ADM1
to two industrial case studies that we used as part of contract projects with industry clients.
In each application, use of mathematical modelling allowed savings of cash and in-kind
resources as compared to laboratory, pilot or full-scale studies.
The first case study is on an upflow anaerobic sludge blanket (UASB) system treating
recycling paper mill wastewater. As is common in such paper mills, the influent calcium
concentrations are relatively high (0.4–0.7 kg·m–3 Ca2+), which leads to excessive precipi-
tation of calcium carbonate (CaCO3) in granules and pipes. In this particular reactor,
this has already caused cementing of the sludge bed once, which cost over $A200,000 and 199
 considerable additional labour costs to scour the pipes and reactor, and import new seed
sludge.
The system has a mixed acidification reactor (1,040 m3), and a methanogenic
UASB reactor (1,805 m3), with a loading rate of 5.7–6.6 kg COD·m–3 (on UASB only)
and 1,000 m3·d–1 feed flow (excluding recycle). Recycle is 67% of total flow and is
around the methanogenic reactor only. The client wished to know whether it was
economically effective to dose with HCl (or other acid) to: (a) decrease pH, and thus
D.J. Batstone and J. Keller

(b) decrease the reactor CaCO3 inventory. We addressed this by a series of steady-state
simulations.
The second system is an anaerobic solids digester fed with solids and concentrated liq-
uid streams from gelatine processing, which is high in proteins and fats. The reactor has 700
m3 volume with an average loading of 2.7 kg COD·m–3·d–1. The feed composition, on a
COD basis, is split between particulate proteins (57%), particulate fats (13%) and soluble
proteins (30%, based on flow-based average particulate COD, soluble COD, and O&G
analysis). Feed is semi-batch, once per day over two hours. In-reactor ammonia concentra-
tion is 0.3–0.4 M with a pH of 7.6, giving a free ammonia concentration of 7–13 mM, which
is well above the 50% inhibitory level of 1.8 mM as described in the STR (Batstone et al.,
2002). Free ammonia inhibition is causing instability in reactor operation, with fluctuations
in both pH and ammonia concentrations, and associated high costs of acid dosing, and high
effluent organic acids, which are causing problems in downstream treatment units.
Ammonia inhibition is less severe in thermophilic systems, and hydrolysis constants are
generally higher (Siegrist et al., 2002). Therefore one option considered to achieve better
solids solublisation and relief of ammonia inhibition is operation at thermophilic tempera-
tures (55°C), rather than mesophilic (35°C) temperatures. The client wished to know
whether operation at thermophilic conditions was likely to: (a) alleviate free ammonia
inhibition and thus allow more stable operation, and (b) allow better performance by
increased hydrolysis rates and increased uptake rates. Because of the semi-batch operation,
it was important to model the dynamics, and we used the model to assess probable reactor
stability and performance on that basis.

Modelling anaerobic systems for industrial application

The tasks required in modelling and simulation can be divided into structural implementa-
tion and specific application. The first area includes formulation of the equations, imple-
mentation of these into a solver, and verification of the model (i.e. correction of errors). The
second step involves configuration of the process within the solver, formulation of the
inputs, and sourcing of parameters (either by estimation from the process, those provided
with the implementation, or other sources such as the STR). Also required may be minor
modifications or extensions to the model structure to account for key processes not includ-
ed in the “standard” model, such as CaCO3 precipitation was added into the ADM1 here.
Model implementation is an extensive task, and it is generally more economic to take
already implemented models, linked to a specific package. Application however, is rela-
tively rapid, and this paper concentrates on this area. Aquasim 2.0 (Reichart, 1994;
http://www.aquasim.eawag.ch) was used for simulations, and the model files in .aqu for-
mat are available from the authors on request.
The single stage model of an anaerobic digester implemented by the task group in
Aquasim 2.0 (as used here) has three vessels: reactor, headspace, and outlet, with a recycle
(non-advective) stream around the reactor to allow for extended sludge retention times.
There is a diffusive link between the reactor and the headspace for gas transfer, and an
advective link between the reactor and outlet.
200
Case study 1: recycled paper-mill wastewater fed to UASB
Data for reactor operation from January–July 2000 was used to calculate flow and time
based averages. Since only steady state predictions were required, process dynamics were
ignored.

(i) Implementation of vessels and hydraulics: To included the recycle around the UASB,
an advective link was added between the outlet and null (no vessel; i.e. sink), with a recycle

D.J. Batstone and J. Keller

to the reactor inlet. The sludge retention time (tres_X) was set to 25 days (with the recycle
stream, this is effectively 75 days). Temperature was set to the process average of 308 K
(35°C).

(ii) Inputs: The time based averages of 1,072 m3·d–1 (reactor feed flow), 2,070 m3·d–1
(recycle flow) and 8.52 kg COD (soluble)·m–3 (all assumed to be monosaccharides) were
used. Without precipitation, the pH was fitted at steady state to 7.4 by addition of 0.093 M
cations, accounting for the amount of strong base in the feed. With precipitation, and
calcium in the feed of 0.011 M (from time based average measurement), 0.09 M cations
(excluding Ca2+) were required for a steady state pH of 7.50.

(iii) Parameters: The suggested parameter set from the STR was used.
To predict calcium inventory, calcium carbonate precipitation was also modelled. This
required three additional state variables to be defined, namely CO32–, Ca2+ and CaCO3
(SCO3, SCa and XCaCO3) and kinetic rate equations for HCO3–/CO32– acid base reactions and
CaCO3 precipitation to be added. Both reactions were modelled using simple first order
kinetics. These are shown in Table 1 in Peterson matrix format:
The following equilibrium coefficients were used at 298 K (From Stumm and
Morgan, 1996, Van Langerak and Hamelers, 1997; ∆H0 values to calculate temperature
dependence):

for HCO3–+OH– → CO32–+H2O pKa = 10.3 ∆H0 = –14,850 J pKa = 10.38 @ 308 K
for CO32–+Ca2+ → CaCO3 pKSO = 8.4 ∆H0 = 12,970 J pKSO = 8.33 @ 308 K

Calcium carbonate was given an additional residence time of 6 days (this triples given the
70% recycle) to fit the steady state calcium concentration in the solids of 0.125 M, as meas-
ured in the granules, and the average in-reactor TS of 48 kg·m–3. Daily feed flowrates, and
COD measurements were also read into Aquasim as real list variables, with linear interpo-
lation between discrete measurements to allow validation of the models against measured
gas flow.

Case study 2: gelatine production wastewater fed to solids digester

Time and flow based averages from 11/96–6/99 were used for input flow and concentration
averages, respectively. Apart from changing the biochemical kinetic parameters, there was
Table 1 Additional rate equations and coefficients

Component → i 10b 10c 12a 12b Rate (ρj, M.d–1)

j Process ↓ SHCO3 SCO3 SCa XCaCO3

A10b Inorganic carbon 1 –1 kA/BCO2 (SCO3 · SH+ – Ka,HCO3SHCO3)

acid-base (2)
A12 Precipitation –1 –1 1 kPRECIP (SCa ˙ SCO3 – Kso,CaCO3)
Bicarbonate Carbonate Calcium ions Calcium Kinetic parameters:
(M) (M) (M) Carbonate (M) k values all set to 10–13 M–1 d–1 to
simulate equilibrium-driven processes 201
 no need to change other parameters between simulations from mesophilic to thermophilic
operation, as all the physicochemical parameters had automatic temperature correction via
the van’t Hoff equation.

(i) Implementation of vessels and hydraulics: Volumes and temperature (308 K) were
inputted to the Aquasim implementation of the ADM1. A temperature of 328 K (55°C) was
used to simulate thermophilic conditions.
D.J. Batstone and J. Keller

(ii) Inputs: The starting point for dynamic simulations was the system at steady state with
the time-average feed flow of 16.8 m3·d–1, and flow-based average particulate protein
(35 kg COD·m–3), particulate fat (8 kg COD·m–3), soluble protein (18.9 kg COD·m–3),
inorganic nitrogen (0.0028 M) and anion concentration of 0.17 M fitted to give the
observed average pH of 7.5. To simulate dynamic operation of the real reactor (which was
fed daily over two hours), a dynamic simulation of 16 days was made, with the input flow as
a real list variable, in a symmetric form with flows of 0 to 403.2 to 0 m3·d–1 over two hours,
every 24 hours. Repeatable response to the time-variable input flow was achieved after 4
days (from steady state initial conditions).

(iii) Parameters: The suggested parameters from the STR for mesophilic and thermophilic
solids digesters were used, with hydrolysis parameters taken from the mesophilic high-rate
parameter set, and interpolated to 328 K using an Arrhenius constant calculated from the
temperature dependence of the composite disintegration first order parameter. In addition,
the parameter set from Siegrist et al. (2002) was used to allow qualitative assessment of
uncertainty. As compared to the parameter set for thermophilic conditions from the STR,
that of Siegrist et al. (2002) has higher decay and yield parameters, but lower KS values, and
higher km values. A system modelled using the parameter set from Siegrist et al. (2002)
tends to be more stable with concentration and flow pulses at high loading rates, but less
stable at low loading rates compared to a model with the ADM1 suggested parameter set.

Outcomes
Case study 1: recycled paper-mill wastewater fed to UASB
Assessing whether acid dosing was economically viable for dosing could be arguably
answered without modelling, as it can be done by ion balancing (most inorganic carbon is
as HCO3 at a pH of 7.6). However, it was easier to do it by a series of simulations to steady
state, and modelling would be required if the pH was lower, or other physicochemical
factors (such as free ammonia inhibition) were influencing biochemical reactions. An
-additional stream with 9 M anions (i.e. 32% W/V HCl) was added, and a series of steady
state simulations performed without CaCO3 precipitation with different flows (–3 m3·d–1,
i.e., addition of base, to 4 m3·d–1) of HCl added. The results of these simulations are shown
in Figure 1. Also shown are the results repeated with CaCO3 precipitation, which shows the
slight effect of CO32– at higher pH values. This allowed us to conclude that pH reduction by
HCl addition was probably not economical, as 1 m3·d–1 at a cost of $A300·d–1 (price at time
of assessment) would be required for a reduction in pH of approximately 0.15 units.
To assess whether dosing was effective in reducing the calcium inventory, the simulations
were repeated on a model with CaCO3 precipitation and retention. The results are shown in
Figure 2, and indicate that dosing would have almost no influence on the CaCO3 inventory.
This is largely because Ca2+ rather than CO32– is limiting (i.e. <10:1) until a pH of 7 (3 m3·d–1
HCl), after which the inventory starts to decrease at a greater rate. From simulations, even at
twice the Ca2+ concentration in the influent, there is still only a negligible effect.
202 To validate the two models, experimental gas flow from January 2000 to June 2000 and
 D.J. Batstone and J. Keller
Figure 1 Influence of HCl addition on pH using model without (continuous line), and with (dashed line)
CaCO3 precipitation

Figure 2 Influence of HCl addition on CaCO3 inventory (continuous line), and ion ratio (dashed line)

the model predictions for this period are shown in Figure 3. Note that the results from the
model with CaCO3 precipitation are almost identical to those without. Acetate concentra-
tion prediction using the STR mesophilic suggested parameter set was quite poor (data not
shown). However, this could be corrected by changing the half saturation coefficient

Figure 3 Sampled (♦) gas flow and simulation gas flow without (continuous line) and with (dashed line)
CaCO3 precipitation 203
 (KS,ac) from 0.15 kg·m–3 to 2 kg·m–3. The high KS,ac may well be due to diffusion limita-
tions in the granules caused by the precipitated CaCO3.

Case study 2: gelatine production wastewater fed to solids digester

The main variable that indicates system stability under dynamic conditions is acetate
concentrations, as in this reactor, acetate is the first to rise under overload conditions or free
ammonia inhibition. The daily response of the model at stable operation with parameters
D.J. Batstone and J. Keller

from the mesophilic, and thermophilic suggested parameter set from the STR, and the
thermophilic data set from Siegrist et al. (2002) are shown in Figure 4.
Note that there is no significant difference in acetate concentrations when using the ther-
mophilic as compared to the mesophilic parameter set from the STR. This is because any
advantages in decreased free ammonia inhibition are offset by: (a) the increased free
ammonia concentration (from 0.013 to 0.05 M) due to decreased Ka,NH4, (b) increased
KS,ac, and (c) increased pH. The model with parameters from Siegrist et al. (2002) has an
elevated acetate level, but same overall degree of fluctuation (~0.2 kg·m–3), which indi-
cates a similar stability to the pulse addition of the feed. Note that the peak is damped,
which indicates the influence of a high yield and decay rate.
To assess performance, and thus predict whether conversion is significantly improved,
the methane production rate is shown in Figure 5. Methane production was used instead of
total biogas production, as the CO2 and H2O components increase by a large amount
for thermophilic operation, neither of which are desirable components in the gas phase and
do not directly relate to increased conversions, but rather changes in physicochemical
parameters.
The simulations in Figure 5 indicate that initial uptake of the substrates is more rapid
under mesophilic conditions, which we assessed from other state variables as being due to
the lower KS values and decay constants. The two thermophilic simulations are consistent
with each other, as both show damping consistent with higher decay coefficients. Slightly
better total integrated gas flow, and therefore corresponding increased COD conversion,
was achieved under thermophilic conditions (8%), but not enough to justify the additional
expense involved in heating the influent. The relatively small increase was largely due to
the almost complete hydrolysis achieved in the reactor. The mesophilic parameters have
been validated (Siegrist and Batstone, 2001) against measured gas flow over two years, and
this prediction is therefore quite accurate.

Figure 4 Simulated daily profiles of acetate under mesophilic (continuous line), thermophilic (dotted line),
204 and with parameters from Siegrist et al. (2002) (dashed line)
 D.J. Batstone and J. Keller
Figure 5 Simulated methane production daily profile under mesophilic (continuous line), thermophilic (dot-
ted line), and with parameters from Siegrist et al. (2002) (dashed line)

Quality and value of predictions

Quality of predictions
We consider four classes of prediction:
1. High accuracy quantitative; the simulations are accurate ±10%
2. Medium accuracy quantitative; the simulations are accurate ±30%
3. Qualitative; the predictions demonstrate the overall qualitative response of the system
to case-study variables to a significant degree.
4. Valueless; confidence in the inputs, outputs or parameters is so poor, that no conclusion
can be made from the simulations.
For case study 1, using simulations, we have predicted that addition of HCl: (i) allows a
pH drop of 0.15 units per m3 32%W/V HCl, and (ii) that this will have no significant influ-
ence on reactor CaCO3 inventory. We assess both these predictions as being in (a), high
accuracy quantitative. For the first prediction, this is because the modelling work was,
essentially, ion balancing, and therefore mainly depends on pH and volumetric flow. In the
second case, we can confidently predict that HCl addition (in the range considered here)
will cause an impact of less than 10%, as the predicted inventory decrease with 4 m3
HCl·d–1 was 1.6%, and the system was relatively insensitive to influent Ca2+ concentration.
For case study 2, we have predicted that changing to thermophilic operation will have: (i) no
significant impact on ammonia inhibition or reactor stability, and (ii) will not allow higher per-
formance (methane production) via increased hydrolysis. We assess prediction (i) as qualita-
tive, as it is highly sensitive to KS and KI,NH3 parameters, which vary between systems and the
long-term and temperature effects of free ammonia inhibition are not very well studied.
However, the value of the prediction is enhanced, as it is supported by simulation with an inde-
pendent parameter set (Siegrist et al., 2002), and the modelling allowed analysis of the reasons
for negligible impact (i.e., decreased Ka,NH4 and increased KS values), the first of which is accu-
rate (due to its thermodynamic basis). We assess prediction (ii) as medium accuracy quantita-
tive, as it is largely a mass balance, which indicates that most COD is hydrolysed and converted
to methane already at mesophilic conditions and therefore only a limited increase in conversion
is possible under thermophilic operation. It is important to note that the original mesophilic
parameters were validated against experimental gas flow (i.e. Siegrist and Batstone, 2001).

Value of predictions
It can be argued that in both these cases, the simulation work was used to eliminate, rather
than validate or develop process changes. This is nevertheless of significant value, since the 205
 modelling work was more rapid, and much cheaper than laboratory or pilot-scale tests and
the predictions were not at all obvious. In fact in the first case, construction work for an HCl
addition system had already started based on some expectated benefits from this. The appli-
cation and benefits of simulation work has also been shown for process development, vali-
dation, and analysis (e.g. Van Langerak et al., 1997).
As seen in this paper, the quality of predictions involving the physicochemical system
can generally be assessed as being higher than for the biological system. This is not because
D.J. Batstone and J. Keller

the physicochemical interactions are less complex, but rather because the parameters
involved (especially coefficients) have a fundamental, thermodynamic basis, and therefore
their values at different conditions can generally be predicted quite well. However, caution
in some cases is still required for prediction of certain physico-chemical parameters, in par-
ticular for precipitation as they can be strongly influenced by other factors (solids, oversat-
uration, inhibitors, non-ideal behaviour of ions, etc.) in the reactor system (see e.g. van
Langerak et al., 1999).

Conclusions
This paper demonstrates the application of the newly developed ADM1 to two full-scale,
industrial case studies to help in the evaluation of possible process or operational changes.
In both cases, it was possible to predict with good accuracy the likely effect of the proposed
changes and then make an informed decision whether these changes were justified or not.
Even the rejection of the proposed changes in both situations was a highly valuable out-
come, as the model was able to prevent costly and time-consuming process modifications
that had a small chance of success. The simulation work to come to these conclusions could
be done in a short time and with minimal costs.
This is a typical and desirable application of modelling and simulation in general, and of
the ADM1 in particular, for direct benefits in the industry. The paper also illustrates that a
high accuracy of all model predictions is not required in such cases, as only a limited
number of simulation outputs are of relevance, and the accuracy of these can be estimated
quite well with some practical considerations. It also showed, that using the model allowed
a much better understanding of the governing processes in these full-scale reactors, and
therefore making a good assessment of the possible impacts of operational modifications.

Acknowledgements
We thank Visy Paper Limited, and Leiner Davis Gelatine for operational data.

References
Batstone, D.J., Keller, J., Angelidaki, I., Kalyuzhnyi, S.V., Pavlostathis, S.G., Rozzi, A., Sanders, W.T.M.,
Siegrist, H. and Vavilin, V.A. (2002). Anaerobic Digestion Model No. 1 (ADM1), IWA Task Group for
Mathematical Modelling of Anaerobic Digestion Processes. IWA Publishing, London.
Reichart, P. (1994). AQUASIM – A tool for simulation and data analysis of aquatic systems. Wat. Sci. Tech.,
30(2), 21–30.
Siegrist, H. and Batstone, D.J. (2001). Free ammonia and pH inhibition of acetotrophic methanogenesis at
mesophilic and thermophilic conditions. in A.F.M. van Velsen and W.H. Verstraete, editors, Preprints of
9th World Congress: Anaerobic Digestion 2001, Antwerp, Belgium, pages 395–400.
Siegrist, H., Vogt, D., Garcia-Heras, J. L. and Gujer, W. (2002). Mathematical model for meso and
thermophilic anaerobic sewage sludge digestion. Environ. Sci. Tech., 36, 1113–1123.
Stumm, W. and Morgan, J.J. (1996). Aquatic Chemistry: Chemical Equilibria and Rates in Natural Waters,
3 edition. John Wiley and Sons, New York.
van Langerak, E.P.A., Hamelers, H.V.M. and Lettinga, G. (1997). Influent calcium removal by
crystallization reusing anaerobic effluent alkalinity. Wat. Sci. Tech., 36(6–7), 341–348.
van Langerak, E.P.A., Beekmans, M.M.H., Beun, J.J., Hamelers, H.V.M. and Lettinga, G. (1999). Influence
of phosphate and iron on the extent of calcium carbonate precipitation during anaerobic digestion.
J. Chem. Tech. Biotech., 74, 1030–1036.
206