Available online at www.sciencedirect.

com

Bioresource Technology 99 (2008) 21322140

Analysis, evaluation, and optimization of kinetic parameters

for performance appraisal and design of UASB reactors
Puspendu Bhunia, M.M. Ghangrekar

Environmental Engineering and Management Section, Department of Civil Engineering, Indian Institute of Technology, Kharagpur 721 302, India
Received 30 January 2007; received in revised form 23 May 2007; accepted 23 May 2007
Available online 20 July 2007

Abstract
Studies have been undertaken to explore the applicability of dierent kinetic models for the performance appraisal of upow anaerobic sludge blanket (UASB) reactors treating wastewater in the range of 3004000 mg COD/l. Three kinetic models namely, Monod,
Grau second-order, and Haldane model are considered for the analysis. Both linear and nonlinear regressions have been performed
to examine the best-t among the kinetic models. In this process, ve error analysis methods have been used to analyze the data. Apart
from optimization of kinetic coecients with minimization of associated errors, prediction of euent COD has also been undertaken to
verify the applicability of kinetic models. In both the cases, Grau second-order model is found to be the best class of t for wide range of
data sets in UASB reactor.
 2007 Elsevier Ltd. All rights reserved.
Keywords: Error analysis; Grau model; Haldane model; Monod model; UASB reactor

1. Introduction
The upow anaerobic sludge blanket (UASB) reactor is
a well accepted anaerobic reactor types for treatment of
sewage as well as low and high strength industrial wastewaters. The success of high chemical oxygen demand (COD)
removal has led to its increasing applicability in the treatment of domestic wastewaters, especially under Mediterranean climatic conditions (Castillo et al., 1999). The
removal of organic matters is reasonably good; especially
for biodegradable organic carbon, making supplementary
treatment unnecessary in many instances (Schellinkhout
and Osorio, 1994).
An understanding of the process kinetics is vital in the
design, development and operation of UASB reactors.
Based on the biochemistry and microbiology of anaerobic
*

Corresponding author. Tel.: +91 3222 283440; fax: +91 3222 282254.
E-mail addresses: puspendu@civil.iitkgp.ernet.in, puspendubhunia@
yahoo.co.uk (P. Bhunia), ghangrekar@civil.iitkgp.ernet.in, ghangrekar@
iitkgp.ac.in (M.M. Ghangrekar).
0960-8524/$ - see front matter  2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.biortech.2007.05.053

process, kinetics provides a judicious basis for process analysis, control, and design. In addition to quantitative
description of the substrate utilization rates, process kinetics also deals with operational and environmental factors
aecting these rates.
Bacterial growth kinetics are based on two fundamental
relationships, i.e., growth rate and substrate utilization
rate. Various kinetic models are reported for anaerobic
processes (Castillo et al., 1999; Isik and Sponza, 2005;
Lokshina et al., 2001; Pavlostathis and Giraldo-Gomez,
1991) predominantly based on Monods equation or its
modications. However, the ample of information obligatory for many of these models, limits their eld application
globally. The kinetic method of designing a biological
treatment system is expected to be superior over empirical
methods, wherein a design is based on few empirical factors. The values of kinetic coecients are to be estimated
by means of regression analysis of experimental data generated from a lab-scale and/or pilot-scale study.
A review of literature reveals that most of the kinetic
models are nonlinear in nature (Ong, 1990). Thus, one

P. Bhunia, M.M. Ghangrekar / Bioresource Technology 99 (2008) 21322140

2133

Nomenclature
Q
Vb
Xo
Xe
X
So
Se
hc
l
lm

ow rate of inuent wastewater (l d1)

volume of sludge bed (l)
biomass concentration of inuent wastewater
(mg l1)
biomass concentration of euent wastewater
(mg l1)
biomass concentration in reactor (mg l1)
inuent substrate concentration (mg l1)
euent substrate concentration (mg l1)
solid retention time (d)
specic growth rate (d1)
maximum specic growth rate (d1)

may anticipate a nonlinear regression technique would be

more competent for better evaluation of kinetic constants
imbedded in the models. Indeed, it is reported that with
asymptotically large samples, nonlinear regression is better in obtaining the optimal estimates of empirical constants associated with nonlinear models (Ong, 1990).
However, for limited sample size, linear regression is also
able to yield a set of good (though not optimal) estimates,
if the nonlinear model can be transformed into a proper
linear form.
A specic nonlinear model can be transformed into different forms of linearized equations. In principle; the estimates obtained from any of the linearized form of
nonlinear equations should yield identical results, if data
being used are free of errors. Since, errors are inevitable
in all biological experimental data; legitimacy of estimates
derived from linearized equations will depend on the form
of equations being used. Of late, the superiority and applicability of dierent kinetic models are being assessed based
on magnitude of the correlation coecient (R2) for linear
regression. The kinetic models giving an R2 value closest
to unity are considered best-t. However, transformations
of nonlinear equations to its linear forms implicitly alter
their error structure, may infringe the error variance and
normality assumptions of standard least squares.
In this article, three kinetic models, namely, Monod
model, Grau second-order substrate removal model, and
Haldane model have been scrutinized. These models are
tested for their capability to demonstrate the substrate
removal and microorganisms growth rate kinetics of
UASB reactor in treating wastewater with COD ranging
from 300 to 4000 mg l1. For the best-t model, error analysis is executed to evaluate the adequacy and accuracy of
the model equations. Error functions are the tools to solve
nonlinear regression analysis. In this study, ve dierent
error functions have been examined and the kinetic parameters are critically assessed by minimizing individual error
function with the aid of solver add-in with Microsofts
spreadsheet, Excel 2003 (Microsoft Corporation).

Kd
Ks
Y
Ks2
a, b
hH
Ki
n
p
y

endogenous decay coecient (d1)

half-velocity saturation constant (mg l1)
Cell yield coecient (mg VSS mg COD1)
Grau substrate removal rate constant (d1)
kinetic constants (see Eq. (14))
hydraulic retention time (d)
inhibition constant (mg l1) (see Eq. (17))
number of data points
number of kinetic constants
dependent variable of the respective kinetic
equation and sux calc = calculated, and
exp = experimental

2. Methods
Two identical UASB reactors (R1 and R2), made of
Plexiglas having eective volume 12.57 l, internal diameter
100 mm, and eective height 1.6 m were used in the study.
Synthetic feed having sucrose as carbon source was used in
this study. Synthetic wastewater used in this study was
mainly consists of 0.89 g sucrose, 1.5 g NaHCO3, 0.318 g
NH4Cl, 0.064 g MgSO4, 0.035 g K2HPO4, and 0.009 g
KH2PO4 per g of COD. Trace metals (Fe, Ni, Mn, Zn,
Co, Cu, and Mo) were added as per the composition suggested by Ghangrekar et al. (2005). The reactors were
seeded with sludge collected from bottom of a septic tank.
Analysis of parameters such as pH, COD, alkalinity, suspended solids (SS), and volatile suspended solids (VSS)
was accomplished as per APHA standard methods
(APHA, 1998). For euents from the reactors, COD was
determined for settled samples. Amount of sludge present
in the reactor was determined by measuring VSS proles
over height of the reactor at various stages under each
loading condition. Together with the data pertaining to
washout of sludge through euent and manual sludge
wastage, an integral picture was composed to nd out solid
residence time (SRT) in the reactor. The COD of synthetic
wastewater used for these experiments was between
300 mg l1 and 600 mg l1. Hydraulic retention time
(HRT) ranging from 8 to 4 h was used under dierent loading rates. All experiments were performed at ambient room
temperature, between 22 and 28 C. In each experimental
run, performance of the reactor was evaluated under constant organic loading rate (OLR).
3. Model descriptions
In all the models, equations are expressed considering
UASB reactor as two continuous ow completely mixed
reactors in series, namely sludge bed zone and sludge blanket zone. The substrate removal in sludge blanket zone is
neglected, i.e., substrate concentration at sludge blanket

2134

P. Bhunia, M.M. Ghangrekar / Bioresource Technology 99 (2008) 21322140

S = Se (euent substrate concentration). The three phase

separator (gasliquidsolid) provided at the top of reactor
is considered to play a role of sludge retention. This minimizes loss of sludge through euent. Biochemical conversion of organic matter, if any, in the three phase
separator is neglected. Overall kinetic coecients have
been determined at steady-state conditions using the data
sets summarized in Table 1.
3.1. Application of Monod kinetic

or referred as mean cell residence time (hc) and calculated

from Eq. (3).
hc

dX
Q
Q

 Xo 
 X e l  X  Kd  X
dt
Vb
Vb
dS
Q
Q
lX

 So 
 Se 
dt V b
Vb
Y

1
2

The ratio of total biomass in the reactor to biomass wasted

per given time correspond to the average time called SRT

The association between specic growth rate and rate limiting substrate concentration can be expressed by Monod
Eq. (4) as
l

For an UASB reactor without biomass recycle, the rate

of change of biomass and substrate in the system can be
expressed respectively as Eqs. (1) and (2):

VbX
Q  Xe

lm  S e
K s Se

The constant lm indicates maximum growth rate of microorganisms when the substrate is being used at its maximum
rate, and Ks indicates the level of substrate concentration at
one-half the maximum specic substrate utilization rate. If
it is presumed that biomass concentration of inuent
wastewater, Xo is negligible and at steady-state conditions
(when dX
0, and dS
0), then the equations derived from
dt
dt
Eqs. (1)(4) are as

Table 1
Details of dierent performance parameters under steady-state conditions
Sl. no.

Inuent COD
(mg l1)

Euent COD
(mg l1)

Sludge bed
volume (l)

X (g l1)

Inuent ow
rate (l d1)

SRT
(d)

HRT
(h)

Reference

1
2
3
4
5
6
7
8
9
10
11
12
13
14

602.08
603.74
622.22
312.16
318.31
574.83
463.55
313.13
308
448.52
448.52
580.15
482.63
778.25

49.42
48.19
79.43
57.71
105.42
83.33
76.7
57.69
75.14
56.18
58.35
34.33
28.8
62.18

4.32
7.85
5.49
2.36
3.53
4
4
4
4.16
5.1
5.1
4.16
7.85
4.32

35.68
31.12
32.33
17.3
13.2
37.98
32.07
26.23
32.21
23.4
31.4
38.3
29.35
39.95

37.7
75.4
75.4
37.7
37.7
37.7
50.9
75.4
75.4
50.9
50.9
62.83
94.25
56.55

213.17
130.2
89.13
261.6
228.04
106.7
113.73
49.77
89.84
273.58
297.93
94.98
76.15
150.86

8
4
4
8
8
8
6
4
4
6
6
4.8
3.2
5.33

Present
experiment

15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34

913
1012
948
932
1023
3130
3214
3034
3024
3009
4054
3435
988
1019
3068
985
965
3253
1944
2068

58
60
55
29
82
210
256
282
241
235
938
439
59
43
294
35
64
164
94
111

2.5
2.5
2.5
2.5
2.5
2.3
2.6
2.7
2.83
3
2.7
2.9
2.7
4.5
4.5
8.5
10.5
9.5
7
6

26.63
26.02
24.55
25.95
24.48
30.28
29.78
40.49
42.69
40.45
38.82
45.78
28.91
32.275
22.68
35.82
35.975
44.315
43.26
36.96

6
8
10.322
12.533
14.953
7.345
10.03
14.392
18.113
19.592
19.917
19.672
12
12
12
36
36
36
24
24

215
159.5
115.5
104.5
79.5
91.05
51.12
47.95
40.68
40.49
22.28
33.26
85.59
145.82
44.76
86.3
143.73
76.93
143.5
135.88

16
12.12
9.3
7.7
6.4
13.07
9.57
6.67
5.305
4.9
4.82
4.88
8
16
16
8
16
16
12
12

Ghangrekar
(1998)

P. Bhunia, M.M. Ghangrekar / Bioresource Technology 99 (2008) 21322140

Q  Y  hc  S o  S e
V b  1 K d  hc
K s  1 K d  hc
Se
hc  lm  K d  1

5
6

Eqs. (5) and (6) are nonlinear in nature; hence, it is indispensable to transform them to equivalent linearized forms
so that linear regression can be used for estimating the values of kinetic constants. Two dierent linearized equations
can be framed to obtain values Y and Kd, which are as
Q  S o  S e 1 1
1
  Kd
VbX
Y hc Y

and
1
Q  So  S e
Y
 Kd
hc
V b  X  hc

The other forms of linearized equation reported in the literature for estimation of lm and Ks are as follows:
X Vb
1 Ks 1
1


Q  S o  S e Y lm S e lm
Q  S o  S e  Y
Q  S o  S e  Y
lm  K s 
X Vb
X  V b  Se

15

The coecient b in Eq. (15) is close to one and generally

reects the impracticality of attaining a zero value of COD.
So
The substrate removal kinetic constant K s2 aX
. This indicates substrate removal rate for each unit of microorganism depending on second-order substrate removal
kinetics. If the values of Ks2 in terms of a is replaced in
Eq. (13) then Eq. (16) can be obtained as
Se
1

S o 1 haH

16

The Haldane kinetics follows the equation as stated in

Eq. (17) (Versyck et al., 1997):
lm
l
17
1 KS es KS ei
The constant Ki indicates the substrate concentration,
above which growth rate of microorganisms is less than
lm
indicating inhibition eect. At steady-state conditions,
2
after applying substrate balance in UASB reactor, Eq.
(18) can be obtained
2

10

V b :S e  X
S e
Se K s

Q  S o  S e K i  k k
k

11

where k lYm , the maximum specic substrate utilization

rate.

3.2. Grau second-order multi-component substrate removal

model
The general equation of a second-order kinetic model is
exemplied in Eq. (12) (Isik and Sponza, 2005; Grau et al.,
1975; Sandhya and Swaminathan, 2006).
 2
dS
Se
K s2  X 
12
dt
So
If Eq. (12) is integrated (boundary conditions: S = So to Se;
and t = 0 to hH) and then linearized, Eq. (13) will be
formed:
S o  hH
So
hH
So  Se
K s2  X

hH
a b  hH
E

3.3. Haldane model

To obtain the estimates of lm and Ks, linear regression is

commonly applied on the linearized equation derived from
Eq. (6). A review of literature reveals that Eq. (9) has been
widely used for estimating lm and Ks (Ong, 1990).
V b  Se  X
Y
Y  Ks

Q  S o  S e lm  S e
lm

2135

13

If the second term of the right part of Eq. (13) is acknowledged as a constant, Eq. (14) will be obtained (Grau et al.,
1975; Buyukkamaci and Filibeli, 2002)

18

3.4. Error analysis

Five dierent error functions have been used which are
popularly used in non-linear regression analysis. These
error functions were scrutinized and the kinetic parameters
were settled on by minimizing the respective errors.
1. The sum of the squares of the errors (SSE):
n
X
2
SSE
y calc  y exp i

19

i1

2. The sum of the absolute errors (SAE):

n
X
SAE
jy calc  y exp ji

20

i1

3. The average relative error (ARE):




n 
100 X
y exp  y calc 
ARE


n i1  y exp 

21

S o  hH
b  hH a
So  Se

14

(So  Se)/So expresses the substrate removal eciency and

is symbolized as E. Therefore, Eq. (14) can be written as

4. The hybrid fractional error function (HYBRID) (Allen

et al., 2004):
"
#
2
n
y exp  y calc
100 X
HYBRID
22
n  p i1
y exp
i

P. Bhunia, M.M. Ghangrekar / Bioresource Technology 99 (2008) 21322140

5. Marquardts percent standard deviation (MPSD) (Allen

et al., 2004):
v
!2
u
n
u 1 X
y exp  y calc
t
MPSD 100
23
n  p i1
y exp
i

Since, each of the error criteria generate dierent sets of

kinetic parameters, an overall optimum set becomes dicult to recognize. To facilitate comparison of the error values, the results of each set were normalized and combined.
The values of the errors acquired for each error function
(for each set of kinetic constants) were divided by the maximum errors for the respective error function to get normalized errors, called sum of normalized errors (SNE).
The SNE for each model was calculated by (Allen et al.,
2004):
(a) Selecting one error function at a time and determining the kinetic parameters that minimize that error
function to produce the kinetic parameter set for that
error function.
(b) Determining the error values for all other error functions for that kinetic parameter set.
(c) Calculating all other parameter sets for each error
function and their associated error function values.
(d) Selecting each error measure in turn and dividing the
errors determined for that error function to get the
normalized errors; and
(e) Summing all these normalized errors for each parameter set.

slope of the straight line using Eq. (7) are 0.083 mg VSS mg
COD1 and 0.006 d1, respectively as illustrated in
Fig. 1(a). The values of lm and Ks were determined from
Fig. 1(b) (using Eq. (9)) as 0.058 d1 and 226.1 mg l1,
respectively. The value of sludge yield Y is higher than that
reported for acetoclastic methanogens (0.010.05 mg
VSS mg COD1) and lower than acidogens (0.140.17 mg
VSS mg COD1) (Pavlostathis and Giraldo-Gomez,
1991). This is due to determination of overall yield for
the mixed culture. The magnitude of endogenous decay
coecient observed is well within the range reported for
acetoclastic methanogens (0.0040.037 d1) (Pavlostathis
and Giraldo-Gomez, 1991). The obtained lm value is at
lower side of the wide range (0.0410.912 d1) reported
for mixed and pure cultures of acetoclastic methanogens
at temperature 3537 C (Lokshina et al., 2001; Pavlostathis and Giraldo-Gomez, 1991; Fukuzaki and Nagai,
1990). This may be due to the lower temperature (22
28 C) and mixed culture used in the kinetic coecient
determination.
To examine the deviation of kinetic coecients due to
linearization, the kinetic data sets were analyzed with linear

0.7
y = 11.999x + 0.0725
2
R = 0.7646

0.6
Q(So-Se)/(VbX)

2136

This method permits a direct comparison of the scaled

errors through which the kinetic constants which would
provide the closest t to the measured data can be
recognized.

0.5
0.4
0.3
0.2
0.1
0
0

0.01

0.02

0.03

0.04

0.05

1/SRT

In order to obtain overall kinetic coecients for the

models, analyses have been carried out with the results
obtained from the present study as well as those reported
by Ghangrekar (1998) at steady-state conditions (Table
1). The kinetic coecients thus obtained from 34 experiments on UASB reactor under dierent operating conditions oer a meaningful comparison and greater insight
into applicability of the kinetic models in designing UASB
reactors.

b 1800
Monod Model

1600

Haldane Model

1400
(VbSeX)/(Q(So-Se))

4. Results and discussion

1200
1000
800
600

Monod Model: y = 1.4335x + 322.99

2

R = 0.7747

400

4.1. Monod model

Thirty four steady-state performance data sets were used
to establish the kinetic parameters required for applying
the model. On the account of poor correlation (R2 < 0.3)
with the data sets (Table 1), Eqs. (8), (10), and (11) have
not been considered for the estimates of kinetic constants.
The Y and Kd values calculated from the intercept and the

Haldane Model: y = -0.0005x + 1.8204x + 304.02

2
R = 0.7436

200
0
0

200

400

600

800

1000

Se

Fig. 1. Determination of Monod and Haldane kinetics: (a) Y and Kd of

Monod model; (b) lm and Ks of Monod model, and kinetic coecients of
Haldane model.

P. Bhunia, M.M. Ghangrekar / Bioresource Technology 99 (2008) 21322140

as well as nonlinear forms of the equation. The nonlinear

form of Monod equations are as
Q  S o  S e
X  lm  S e
obtained from Eq: 9

Vb
Y  Ks S e
Substituting
Then,

X  lm  S e
y1
Y  K s S e

QS o S e
Vb

24

y 1 and

1
l  Se
m
 K d obtained from Eq: 7 and Eq: 8
hc K s S e
25
l  Se
Substituting m
 K d y 2 in the above equation
K s Se
26
Then, h1c y 2 .
Eq. (24) (nonlinear form of Monod equation) is used to
nd out Y, lm, and Ks and Eq. (26) is used for determination of Kd. In Eqs. (24) and (26), the rst y1 (or y2) denotes
the theoretical value and second y1 (or y2) denotes the
experimental value.
The values of kinetic coecient obtained from Fig. 1 for
Monod model are presented in the linear transform
(LTFM) columns and the kinetic coecients determined
by nonlinear regression using ve dierent error functions
are presented in Tables 2 and 3, respectively. From the
SNE values of Table 2, it is seen that signicant errors
are associated with the determination of kinetic coecients
Y, lm, and Ks by linearized model. The SNE values indicate
that the nonlinear regression with SAE parameter set produces the best overall t for the calculation of Y, lm, and
Ks. The optimum values of Y, lm, and Ks thus obtained

2137

from error analysis are 0.07 mg VSS mg COD1,

0.048 d1, and 226.1 mg l1, respectively.
The values of SNE in Table 3 indicate that estimation of
Kd from linearized equation involves maximum errors. The
nonlinear regression with ARE, and HYBRID parameter
sets is observed to produce the optimum t for evaluating
Kd. The modied value of Kd obtained by performing nonlinear regression using error analysis is 0.004. Hence, the
associated errors in kinetic coecients by linearized model
necessitate the use of nonlinear regression using dierent
error functions for the estimates of kinetic coecients for
Monod model.
4.2. Grau second-order multi-component substrate removal
model
In order to nd out the kinetic coecients (a, b and Ks2)
Eq. (15) is plotted in Fig. 2. The values of a, and b are calculated from the intercept and slope of the straight line on
the graph. The values of a, and b are found to be 0.558 and
1.043 with high correlation coecients of (R2) 0.977. In
order to analyze the error involved in the determination
of kinetic coecients due to linearization of the model,
optimization techniques have been applied with previously
mentioned ve dierent error functions. In this case,
LTFM considers Eq. (15) and the nonlinear regression considers Eq. (27), which is the modied form of Eq. (16).
y3

1
1 hH =a

27

where y 3 SSoe .
The values of kinetic coecient associated with dierent
error functions are presented in Table 4. It is seen from

Table 2
Linear and nonlinear kinetic coecients (Ks, Y, lm) of Monod model
Ks
Y
lm
SSE
SAE
ARE
HYBRID
MPSD
SNE

LTFM

SSE

SAE

ARE

HYBRID

MPSD

226.1
0.083
0.058
1.15 10+8
5.09 10+4
2.66 10+1
5.90 10+4
3.56 10+1
4.71

225.77
0.08
0.057
1.14 10+8
5.13 10+4
2.72 10+1
6.06 10+4
3.65 10+1
4.78

226.1
0.07
0.048
1.17 10+8
5.03 10+4
2.62 10+1
5.83 10+4
3.49 10+1
3.84

226.1
0.097
0.062
1.41 10+8
5.24 10+4
2.52 10+1
6.01 10+4
3.34 10+1
4.83

252.69
0.085
0.061
1.2 10+8
5.01 10+4
2.56 10+1
5.77 10+4
3.39 10+1
4.63

275.69
0.059
0.042
1.39 10+8
5.05 10+4
2.47 10+1
6.09 10+4
3.31 10+1
4.77

Table 3
Linear and nonlinear kinetic coecient (Kd) of Monod model
Kd
SSE
SAE
ARE
HYBRID
MPSD
SNE

LTFM

SSE

SAE

ARE

HYBRID

MPSD

0.006
1.36 103
1.62 101
4.28 10+1
3.34 101
5.64 10+1
4.77

0.002
8.63 104
1.37 101
4.64 10+1
2.92 101
6.65 10+1
4.35

0.003
8.85 104
1.34 101
4.20 10+1
2.63 101
5.90 10+1
4.06

0.004
9.75 104
1.37 101
3.92 10+1
2.61 101
5.44 10+1
4.00

0.004
9.75 104
1.37 101
3.93 10+1
2.59 101
5.44 10+1
4.00

0.005
1.13 104
1.47 101
4.02 10+1
2.85 101
5.34 10+1
4.27

P. Bhunia, M.M. Ghangrekar / Bioresource Technology 99 (2008) 21322140

HRT/E

2138
20
18
16
14
12
10
8
6
4
2
0

Monod model. Hence, in this range of inuent COD,

Monod model should be preferred over Haldane model.
In order to determine errors involved in the kinetic coecients obtained from Eq. (18), error analysis with minimization of dierent error functions have been carried
out (Table 5). Negligible dierence in the kinetic coecients is observed between Eq. (18) and those obtained
from error functions. This can be explained, as the basic
form of Haldane model is nonlinear by nature; hence,
determination of kinetic coecients by nonlinear regression analysis using Eq. (18) does not involve noticeable
errors.

y = 1.0429x + 0.5582
2
R = 0.9765

10
HRT

15

20

Fig. 2. Determination of kinetic coecients of Grau second-order model.

Table 4 that determination of kinetic constants using nonlinear regression with ARE, and HYBRID produces lowest
sum of normalized errors. It is also observed that determination of kinetic constants using Eq. (15) does not involve
much error as the SNE value of LTFM is very close to the
minimum values obtained from nonlinear regression.
4.3. Haldane model
In order to acquire the kinetic coecients of Haldane
model, Eq. (18) is plotted (Fig. 1(b)). The values of kinetic
coecients obtained from Fig. 1(b) are k = 0.55 d1,
Ks = 167.211 mg l1, and Ki = 3636.36 mg l1. On account
of the large value of Ki, indicating no inhibition eect of in
reactor COD values on the growth rate of microorganisms,
it is observed that Haldane model basically directs to

4.4. Evaluation of kinetic models

From engineering point of view, development of kinetic
model is a useful tool in designing and optimization of the
process by reducing extensive and complex experimental
data to simple and convenient mathematical expressions.
The precise determination of kinetic coecients and selection of appropriate mathematical relationship between process variables is obligatory to increase the applicability of
such a model. Analysis and estimation of kinetic coecients for predicting performance of UASB reactors have
been carried out with Monod model, Grau second-order
model, and Haldane model as discussed. Among these
three models, Grau second-order model ts well with the
experimental data set. The best-t of Grau second-order
model has been reported by Sandhya and Swaminathan
(2006) even for the treatment of textile wastewater in
hybrid column upow anaerobic xed bed reactor. No sig-

Table 4
Linear and nonlinear kinetic coecient (a) of Grau second-order model

a
SSE
SAE
ARE
HYBRID
MPSD
SNE

LTFM

SSE

SAE

ARE

HYBRID

MPSD

0.558
0.160
1.537
41.463
3.022
54.809
4.24

0.74
0.142
1.479
46.645
3.658
71.518
4.64

0.705
0.143
1.463
44.621
3.437
67.582
4.47

0.573
0.157
1.520
41.434
3.0246
55.723
4.24

0.562
0.159
1.533
41.456
3.022
55.033
4.24

0.459
0.186
1.691
43.177
3.258
51.783
4.54

Table 5
Kinetic coecients of Haldane model

Ks
k
Ki
SSE
SAE
ARE
HYBRID
MPSD
SNE

Eq. (17)

SSE

SAE

ARE

HYBRID

MPSD

167.212
0.55
3636.36
1500261
4960.408
33.508
7661.201
49.235
5

167.987
0.596
3636.359
1499391
4563.842
29.934
6166.84
43.218
4.5

167.228
0.594
3636.359
1180940
4930.915
29.948
6184.903
43.21302
4.36

167.212
0.594
3636.36
1180938
4572.722
33.270
6184.697
43.211
4.39

167.219
0.594
3636.359
1180941
4572.77
29.948
7653.526
43.213
4.48

167.212
0.594
3636.36
1180938
4572.722
29.947
6184.697
49.213
4.41

P. Bhunia, M.M. Ghangrekar / Bioresource Technology 99 (2008) 21322140

2139

Table 6
Details of experimental data sets used for prediction and validation by dierent kinetic models
Sl. no.

Inuent COD
(mg l1)

Euent COD
(mg l1)

Sludge bed
volume (l)

X
(g l1)

Inuent ow
rate (l d1)

SRT (d)

HRT (h)

Reference

1
2
3
4
5
6

3534
3007
3007
3278
3327
4005

259
213
224
302
342
590

6.08
10.21
12.56
10.95
17.73
13.48

47.15
35.09
30.75
40.68
36.1
28.95

17.71
23.88
31.65
46.83
69.57
71.82

107.98
98.75
76.25
58.37
58.61
35.51

16
12
9
6
4
4

Ghangrekar
et al. (2002)

nicant dierences are noticed with the linearized equation

of Grau second-order model.
In case of Monod model, linear and nonlinear regression
with error functions reveals dierences in the values of
kinetic constants. Errors are minimal, when the kinetic
constants have been determined by nonlinear regression
methods. This shows the suitability of nonlinear regression
analysis for the determination of kinetic constants by
Monod model. In addition, this study enable us to recognize the value of Eq. (9), linearized forms of Monod model,
over other linearized forms (Eqs. (10) and (11)) for determination of kinetic coecients of Monod model. Haldane
model indicates that because of the large value of inhibition
coecient, basically it leads to the Monod model. Apart
from the above three models, Contois model has also been
tested. However, because of the very poor correlation
(R2 < 0.2) observed with the same 34 data sets (Table 1),
the details have not been shown here.
In order to observe usefulness of those three models,
apart from linear and nonlinear regression analysis with
dierent error functions, euent COD concentration has
been predicted with values (considering only properly
start-up reactor values) reported by Ghangrekar et al.
(2002) (Table 6). In this case, kinetic coecients corresponding to the minimum SNE values of linear regression
as well as nonlinear regression are used for the analysis.
Mean of absolute percent error (MAPE) is calculated to
visualize the percentage variations between predicted euent COD and observed euent COD. The MAPE values
for euent COD were (i) 37% by linear regression, and
31% by nonlinear regression for Monod model; (ii) 14%
by linear regression, and 15% by nonlinear regression for
Grau second-order substrate removal model; and (iii)
52% by nonlinear regression of Haldane model. The
observed response of prediction analysis is quite similar
to that of the error analysis discussed earlier. In this case
also, it is observed that Grau second-order model gives better prediction among the kinetic models considered in this
study. Negligible dierence in MAPE between linear and
nonlinear forms of Grau second-order model further signies the usefulness of linearized form of this model for prediction of euent COD.
Association of considerable errors have been noticed
for predicting euent COD by Monod model. Also, prediction of euent COD based upon kinetic coecients of
linearized Monod model involves error compared to its

nonlinear form. Signicant errors represented by MAPE

values associated with Haldane model, once again invalidate the applicability of these models to predict euent
COD within the range of COD concentration studied.
Thus, the error analysis, prediction observation, and optimization (based upon the statistical tools to decide on
appropriate mathematical steady-state model) render minimum errors in the case of Grau second-order model.
Thus, it is rational to conclude that among the models
tested, Grau second-order model is the best-t for the performance evaluation, prediction and designing of UASB
reactors.
5. Conclusion
In this study the kinetics of UASB reactor treating
wastewater with COD in the range of 3004000 mg l1
has been explored using Monod model, Grau second-order
model, and Haldane model. The results of this analysis
indicate that Grau second-order multi-component substrate removal model ts well for estimates of kinetic coefcients in UASB reactors. Among the three kinetic models,
Grau second-order model is observed to be the preeminent
model for predicting the performance of UASB reactors. It
is also observed that linearized form of Grau second-order
model does not engross considerable errors compared to
nonlinear ones.
The observed values of kinetic coecients of linearized
Monod model are in good agreement with reported values
in literature. The present investigation of nonlinear regression using dierent error functions as well as prediction
analysis reveals association of errors with the linearized
forms of Monod model used for evaluating the kinetic
coecients. Hence, it is advisable to undertake nonlinear
regression technique using dierent error functions in the
determination of kinetic coecients of Monod model.
References
Allen, J.S., Mckay, G., Porter, F.J., 2004. Adsorption isotherm models for
basic dye adsorption by peat in single and binary component systems.
J. Colloid Interface Sci. 280, 322333.
APHA, AWWA, WPCF, 1998. Standard Methods for Examination of
Water and Wastewater, 19th ed. American Public Health Association,
Washington, DC, USA.
Buyukkamaci, N., Filibeli, A., 2002. Determination of kinetic constants of
an anaerobic hybrid reactor. Process Biochem. 38, 7379.

2140

P. Bhunia, M.M. Ghangrekar / Bioresource Technology 99 (2008) 21322140

Castillo, A., Llabres, P., Alvarez, M.J., 1999. A kinetic study of a

combined anaerobicaerobic system for treatment of domestic sewage.
Water Res. 33, 17421747.
Fukuzaki, N.N., Nagai, S., 1990. Kinetics of the methanogenic fermentation of acetate. Appl. Environ. Microbiol. 56, 31583163.
Ghangrekar, M.M., 1998. Studies on granulation, start-up and performance of UASB reactor under dierent operating conditions, Ph.D.
Thesis, CESE, Indian Institute of Technology, Powai, Bombay.
Ghangrekar, M.M., Kahalekar, U.J., Sakle, J.J., 2002. Applicable loading
rates for design of upow anaerobic sludge blanket (UASB) reactor. J.
Inst. Eng. 82, 4853.
Ghangrekar, M.M., Asolekar, S.R., Joshi, S.G., 2005. Characteristics of
sludge developed under dierent loading conditions during UASB
reactor start-up and granulation. Water Res. 39, 11231133.
Grau, P., Dohanyos, M., Chudoba, J., 1975. Kinetics of multicomponent
substrate removal by activated sludge. Water Res. 9, 637642.
Isik, M., Sponza, T.D., 2005. Substrate removal kinetics in an upow
anaerobic sludge blanket reactor decolorizing simulated textile wastewater. Process Biochem. 40, 11891198.

Lokshina, L.Y., Vavilin, V.A., Kettunen, R.H., Rintala, J.A., Hollinger,

C., Nozhevnikova, A.N., 2001. Evaluation of kinetic coecients using
integrated Monod and Haldane models for low temperature acetoclastic methanogenesis. Water Res. 35, 29132922.
Ong, L.S., 1990. A comparison of estimates of kinetic constants for a
suspended growth treatment system from various linear transformations. Res. J. Water Pollut. Cont. Fed. 62, 894900.
Pavlostathis, S.G., Giraldo-Gomez, E., 1991. Kinetics of anaerobic
treatment. Water Sci. Technol. 24, 3559.
Sandhya, S., Swaminathan, K., 2006. Kinetic analysis of treatment of
textile wastewater in hybrid column upow anaerobic xed bed
reactor. Chem. Eng. J. 122, 8792.
Schellinkhout, A., Osorio, E., 1994. Long-term experience with the UASB
technology for sewage treatment on large scale, The Netherlands. In:
7th Int. Symp. on Anaerobic Digestion, 2327 January 1994, pp. 251
252.
Versyck, J.K., Claes, E.J., van Impe, F.J., 1997. Practical identication of
unstructured growth kinetics by application of optimal experimental
design. Biotechnol. Prog. 13, 524531.