Prediction of carbon, hydrogen, and oxygen compositions of raw and torrefied biomass using proximate analysis

Fuel 180 (2016) 348–356

Contents lists available at ScienceDirect

Fuel
journal homepage: www.elsevier.com/locate/fuel

Full Length Article

Prediction of carbon, hydrogen, and oxygen compositions of raw
and torrefied biomass using proximate analysis
Daya Ram Nhuchhen
Mechanical Engineering, Dalhousie University, Halifax, NS B3H 4R2, Canada

h i g h l i g h t s
 Examine if the existing correlation can be used to predict the elemental compositions of torrefied biomass.
 Analyze different forms of new correlations using data from raw and torrefied biomass.
 Validate the selected correlations with another set of published data.
 Compare the existing and selected new correlations.
 Only new correlations are applicable for torrefied biomass.

a r t i c l e

i n f o

Article history:
Received 20 January 2016
Received in revised form 5 April 2016
Accepted 11 April 2016
Available online 18 April 2016
Keywords:

Biomass
Torrefaction
Proximate analysis
Ultimate analyses
Correlations

a b s t r a c t
Elemental compositions of biomass are essential for designing energy conversion systems. Only a few
correlations to estimate the elemental compositions using the proximate analysis of raw biomass have
been published so far. Recently researches on biomass torrefaction have been increasing significantly,
which require performing an elemental analysis of the torrefied biomass. Torrefaction affects both the
proximate and elemental analyses of biomass. Therefore, this study examines if the existing correlations
can be deployed or not for estimating carbon, hydrogen, and oxygen compositions of the torrefied
biomass. For this, estimation errors were calculated for the existing correlations using data from the torrefied biomass. Results suggest that existing correlations were not suitable for predicting the elemental
compositions of the torrefied biomass. New correlations were proposed using a wide range biomass,
including both raw and torrefied biomass (447 samples). New correlations C ¼ À35:9972 þ 0:7698VM
þ1:3269FC þ 0:3250ASH; H ¼ 55:3678 À 0:4830VM À 0:5319FC À 0:5600ASH,
and
O ¼ 223:6805À
1:7226VM À 2:2296FC À 2:2463ASH were selected for future use. These correlations have the MAE of
2.58%, 0.41%, 2.60%, the AAE of 5.23%, 9.94%, 8.79%, the ABE of 0.45%, 2.82%, 2.01%, and the R2 of 0.83,
0.70, 0.84 corresponding to the measured values of C, H, and O, respectively. The selected correlations
were also validated and compared with the existing correlations using another set of data that includes
raw, washed, torrefied, and carbonized biomass. Selected new correlations could be used for predicting
carbon, hydrogen, and oxygen compositions in the raw and torrefied biomass, especially those biomasses
which have negligible nitrogen and sulfur contents.
Ó 2016 Published by Elsevier Ltd.

1. Introduction
Biomass is widely available renewable energy resources with

balanced CO2 emissions and absorption. Physical, chemical and
thermodynamic properties of biomass are essential parameters
for designing any energy systems [1]. For instance, the higher
heating value (HHV), which gives the energy content of biomass,
is considered to be an important fuel parameter for a design of
the combustion system [2]. The elemental compositions of biomass

E-mail addresses: ,
/>0016-2361/Ó 2016 Published by Elsevier Ltd.

are also necessary to analyze the overall process of any thermochemical conversion methods. The elemental compositions help
to predict the flue gas flow rate, air requirement, and flue gas compositions in the combustion process. The experimental method of
finding the elemental compositions is, however, costly and
requires a sophisticated equipment [3]. It also needs highly skilled
engineers or analysts [4–6]. Therefore, having proper correlations
for predicting the major elemental compositions would always
be an asset for a design engineer. However, only a few such studies
[7–9] are available in the literature so far. While the correlation by
Vakkilainen [9] is limited only to the black liquor, correlations by


D.R. Nhuchhen / Fuel 180 (2016) 348–356

Parikh et al. [7] and Shen et al. [8] are applicable for a wide range of
biomass. The range of data points considered by Parikh et al. [7] for
volatile matter, fixed carbon, ash, carbon, hydrogen and oxygen
contents of biomass are 57.2–90.6%, 4.7–38.4%, 0.12–77.7%, 36.2–
53.1%, 4.4–8.3%, and 31.4–49.5%, respectively, Shen et al. [8], on
the other hand, have used the data points in the range of 57.2–
90.6%, 9.2–32.8%, 0.1–24.6%, 36.2–53.1%, 4.7–6.6%, and 31.4–

48.0% for VM, FC, ASH, C, H, and O, respectively. While Parikh and
his co-workers neglected the effect of ash compositions on the elemental analysis, Shen and his colleagues discussed the importance
of ash measurement and proposed new correlations including the
ash compositions. Shen et al. [8] found correlations with a better
prediction compared to those developed by Parikh et al. [7]. However, these correlations are purely based on raw biomass and some
chars.
Recently, interest on biomass torrefaction because of its ability
to improve heating values, hydrophobicity, grinadlbility, flowability and combustion characteristics of biomass have increased significantly. Studies have found that the torrefaction of biomass
has changed components of both proximate and ultimate analyses.
More on the effect of torrefaction on biomass and the technologies
of torrefaction are reviewed in different papers [10–14]. Reported
results suggest that torrefaction reduces volatile matter (VM) of
biomass and increases fixed carbon and ash (ASH) contents per unit
weight of the torrefied biomass. Reduction in volatile matter of the
torrefied biomass is due to the solid mass loss contributed by the
mild pyrolysis of hemicellulose, cellulose, and lignin during torrefaction process [15]. Decomposition of hemicellulose and
depolymerization of cellulose and lignin compositions of biomass
are two major reactions of the torrefaction process that release different light volatile gases including non-condensable gases (carbon
dioxide, and carbon monoxide) and condensable gases (water,
methanol, and carboxylic acids) [16]. As a result, it increases the
fuel ratio, making biomass more comparable with the fuel ratio
of coal (lignite) [17]. This then improves the flame stability [12]
and reduces the burnout rate of biomass [18]. Due to the decomposition and depolymerization reactions that favor decarboxylation
and deoxygenation of biomass [19], the torrefaction process
increases the carbon content of biomass and reduces the hydrogen
and oxygen contents. This is due to the removal of different light
volatile gases that have high hydrogen and oxygen atoms. This
then moves the biomass from the right-hand side of the VanKrevelen diagram to the left side (towards the coal side) [20].
Considering the fact that there will be major changes in the
proximate and ultimate analyses of the torrefied biomass. As these

changes in the properties of biomass vary with the operating conditions of the torrefaction process, the existing correlations for raw
biomass will not be able to incorporate the changes in the fuel
properties of biomass after torrefaction. So, it would be worth noting that existing correlations will have large estimation error in
predicting the elemental compositions of the torrefied biomass
and new correlations have to be devised to address those changes
in properties of biomass after torrefaction. New correlations that
included a wider range of data than the existing correlations will
have a better accuracy of prediction. Since such correlations are
targeted mainly for those users of biomass in boilers who do not
have all experimental facilities to test, a good correlation with less
error only could help them to determine accurate controlling information for the operation of the boiler. It then helps to run a boiler
smoothly, improving the reliability of the power plant operation. In
addition to this, these new correlations could also be used to validate the experimental results from the elemental analyzer.
Though one may argue that the current expressions are valid for
a wide range of biomass materials, hence such correlations can be
useful to determine the elemental compositions of the torrefied
biomass. This study, therefore, answers if the existing correlations

349

can predict the elemental compositions of torrefied biomass or not.
This study used data presented by the author in his previous work,
which verifies the HHV correlations. This work is divided into: (i)
reviewing the published correlations to predict the elemental compositions of biomass using proximate analysis, (ii) examining if the
currently available correlations can be used to predict the elemental compositions of the torrefied biomass or not, (iii) developing
new forms of correlations using a large number of published data
of the proximate of the raw and torrefied biomass, and (iv) validating and comparing the selected new correlations with the existing
correlations using another set of data.

2. Methodology

The data points considered in this study are reviewed from different kinds of the published literature and are reported in supplementary file (Table S1 [21–50] and Table S2 [2,21–47,51–64]).
Table S1 shows how proximate and ultimate analyses were changed at different operating conditions after the torrefaction process
compared those data with the raw biomass presented in Table S2.
One can note that ranges of VM, FC, ASH, C, H, and O of raw biomass
materials are 47.70–93.60%, 0.67–36.10%, 0.01–48.70%, 19.12–
56.30%, 2.00–7.36%, and 25.18–49.50%, respectively. Corresponding components of the proximate and ultimate analyses of torrefied biomass materials are 13.30–88.57%, 11.25–82.74%, 0.08–
47.62%, 35.08–86.28%, 0.53–7.46%, and 4.31–44.70%, respectively.
This information clearly tells that the ranges of the proximate
and ultimate analyses were varied significantly after torrefaction
process. For example, the minimum value of volatile matter is
reduced to 13.30% whereas the carbon content is increased to
88.57%. This confirms that new scheme has to be developed with
the new range of data including torrefied biomass. The collected
data are in wt.% dry basis. Some data, which are, other than in
dry basis originally in the published literature, were also converted
into the dry basis. Though the collected data points have nitrogen
and sulfur contents, they are very small compared to the carbon,
hydrogen, and oxygen contents. Therefore, the author focuses
mainly on the finding the correlations only for carbon, hydrogen,
and oxygen. One may, however, argue that the oxygen content
can be obtained by difference method (by subtracting C, H, N, S,
and ASH compositions from 100%) and does not require correlation.
But, having correlation for oxygen content will avoid the necessity
of nitrogen and sulfur contents and assist in validating the oxygen
content if the difference method was to be adopted to find the oxygen content.
Before validating the existing correlations, three major compositions of raw and torrefied biomass were plotted with different
components of the proximate analysis. These plots analyze the distribution of the elemental compositions (mainly carbon, hydrogen,
and oxygen contents) with the proximate analysis.
In order to validate if the existing correlations can be used or
not, existing correlations were used to predict the elemental compositions of the torrefied biomass (Presented in Table S1). The

deviation between the predicted and measured values was examined by calculating the estimation errors.
Tables S1 and S2 were then merged to determine new correlations using the principle of the least sum square error in the Microsoft Excel. Different forms of the new correlations were selected
from the correlations presented in Parikh et al. [7] and Shen
et al. [8] for analysis purpose. Some more additional forms of possible correlations were also analyzed. Estimation errors for the
existing correlations were also calculated using both raw and torrefied biomass to determine their suitability in a wider range of
biomass types.


350

D.R. Nhuchhen / Fuel 180 (2016) 348–356

2.1. Estimation errors

(a) 100

N
X
MAE ¼
jPi À M i j=N

!,

AAE ¼

N
X
jP i À M i j=Mi

ABE ¼

N
X
ðPi À M i Þ=Mi

i¼1

i¼1

R2 ¼ 1 À

N
X
ðPi À M i Þ2
i¼1

Fcraw

80

60

40

20

N
!,

,

0

N
N
X
 2
ðP i À MÞ
i¼1

 represent the predicted, measured, and an averwhere P; M, and M
age of measured elemental compositions of the biomass sample,
respectively. N is the number of sample data points used for the
regression analysis. While the AAE measures the degree of closeness
between the predicted and measured elemental compositions, the
ABE calculates the degree of overestimation and underestimation.
On the other hand, the MAE provides the actual amount of error
in the same unit that the physical quantity has. Therefore, this study
has selected the correlation with the lowest MAE value and with a
higher degree of fitness (high R2 value) as the best correlation.
3. Results and discussion

(b)

100

80

20

40

60

80

100

VMtor
VMraw

3.1. Scatter distribution of data

60

40

20

0

0

20

40

60

80

100

Volatile matter content (% Wt, dry basis)

(c)

100
ASHtor

Carbon content (wt. %, dry basis)

Figs. 1–3 show how the elemental compositions of biomass vary
with different components of the proximate analysis. Fig. 1(a) indicates that the variations of carbon contents with the fixed carbon
content (FC) of the raw biomass and of the torrefied biomass are
in similar trend. The fixed carbon content and the elemental carbon content of raw biomass are observed to be less than 40% and
60%, respectively. However, the fixed carbon and elemental carbon
contents of torrefied biomass are reported up to 85%. This scattered
plot, therefore, tells that any correlations developed with a single
fixed carbon content term will have an error in predicting the carbon content of the torrefied biomass. Fig. 1(a) also shows that the
elemental carbon content increases with the rise in the fixed carbon content irrespective of raw or torrefied biomass. This is not
the case with the volatile matter (VM) (Fig. 1(b)). While the elemental carbon content increases with the volatile matter of raw
biomass, it reduces with the volatile matter in the torrefied biomass. However, the elemental carbon decreases with the increase
in ash content of the biomass, which is true in both the raw and
torrefied biomass (Fig. 1(c)). But, the decreasing trend is not collinear. Therefore, the correlation of the carbon content with the ash
content of the raw biomass has to be modified using the data
points of the torrefied biomass.
Fig. 2(a–c) shows the variation of hydrogen with FC, VM, and
ASH contents of raw and torrefied biomass. Fig. 2(a) indicates that

0

Fixed carbon content (% Wt, dry basis)

Carbon content (% wt., dry basis)

i¼1

FCtor

Carbon content (wt.% , dry basis)

The correlation is said to be the best-fitted regression line if the
error of the estimation tends to zero [1]. However, it would be not
possible to have such correlations. So, three forms of estimation
errors, including the mean absolute error (MAE), average absolute
error (AAE), and average biased error (ABE) were calculated to
select the best suitable correlations. On the other hand, the coefficient of determination (R2) value was calculated to determine the
degree of goodness of the proposed correlations. All the estimation
errors and the coefficient of determination are estimated as:

ASHraw

80

60

40

20

0

0

10

20

30

40

50

Ash content (% Wt, dry basis)
Fig. 1. Variation of carbon content of raw and torrefied biomass: (a) Fixed carbon
content; (b) Volatile matter content; and (c) Ash content.


351

D.R. Nhuchhen / Fuel 180 (2016) 348–356

(a) 56

(a) 8
FCtor

FCtor

Oxygen content (wt. %, dry basis)

Hydrogen content (wt. %, dry basis)

Fcraw
6

4

2

48

Fcraw

40
32
24
16
8
0

0

0

20

40

60

80

0

20

100

40

60

80

100

Fixed carbon content (% Wt, dry basis)

Fixed carbon content (% Wt, dry basis)

(b) 56

(b) 8

VMtor

Oxygen content (% wt., dry basis)

Hydrogen content (% wt., dry basis)

VMtor
VMraw
6

4

2

48

VMraw

40
32
24
16
8

0

0

0

20

40

60

80

100

0

20

40

60

80

100

Volatile matter content (% Wt, dry basis)

Volatile matter content (% Wt, dry basis)

(c) 56

(c) 8

ASHtor

Oxygen content (wt. %, dry basis)

Hydrogen content (wt. %, dry basis)

ASHtor

ASHraw
6

4

2

48
40
32
24
16
8
0

0

0

10

20

30

40

50

Ash content (% Wt, dry basis)
Fig. 2. Variation of hydrogen content of raw and torrefied biomass: (a) Fixed carbon
content; (b) Volatile matter content; and (c) Ash content.

the hydrogen content is more scattered with the fixed carbon
content of raw biomass. But, the hydrogen content decreases with
the fixed carbon content of the torrefied biomass. However, the
hydrogen content increases with the volatile matter of both raw

ASHraw

0

10

20

30

40

50

Ash content (% Wt, dry basis)
Fig. 3. Variation of oxygen content of raw and torrefied biomass: (a) Fixed carbon
content; (b) Volatile matter content; and (c) Ash content.

and torrefied biomass (Fig. 2(b)). A single correlation with the volatile matter term may able to estimate the hydrogen contents of
both raw and torrefied biomass. Hydrogen content, however,
shows more scattered points with the ash contents of both raw
and torrefied biomass.


352

D.R. Nhuchhen / Fuel 180 (2016) 348–356

Fig. 3(a–c) shows the distribution of oxygen contents with FC,
VM, and ASH contents. While oxygen content decreases with the
fixed carbon content, it increases with the volatile matter contents
of raw and torrefied biomass. However, the distribution of the oxygen content is more scattered with the fixed carbon content compared that with the volatile matter. The distribution of the oxygen
content is also more scattered with the ash contents of both raw
and torrefied biomass.
From above discussions, one can make out that the existing correlations developed with the proximate analysis of raw biomass
might not be suitable for predicting the elemental compositions
of the torrefied biomass.
3.2. Validation of existing correlations using data from torrefied
biomass
For the further verification and to determine the applicability of
the existing correlations for predicting the elemental compositions
of torrefied biomass, estimation errors were calculated using the
set of data (torrefied biomass) presented in Table S1. Table 1 presents the calculated estimation errors for the existing correlations.
Correlations presented by both Shen et al. [8] and Parikh et al. [7]
have large estimation errors. Negative ABE values for carbon content clearly indicate that the existing correlations underestimate
the carbon content of the torrefied biomass. This confirms that
the existing correlations, which were developed using the proximate analysis of raw biomass, fail to incorporate the effect of torrefaction. This is also supported by the results of large positive ABE

values calculated for hydrogen and oxygen contents. While comparing the correlations presented by Shen et al. [8] and Parikh
et al. [7], the estimation errors of the Shen correlations were found
to be higher than the Parikh correlations. This must be due to the
additional errors associated with the ash contents.
However, it is also important to note that the existing correlations are still suitable for predicting non-torrefied (raw) biomass
or not. This can be tested by finding the estimation errors for the
existing correlation using the collected information of raw biomass
(Table S2). Results of estimation errors are presented in Table 1.
One can note that the estimation errors of the existing correlations
are significantly smaller for the raw biomass compared that for the
torrefied biomass. This confirms that the existing correlations are
still good to predict the elemental analysis of the raw biomass.
But, torrefaction of biomass that releases different light volatiles
from the parental biomass leads to change in the fuel type. The
Table 1
Estimation errors of the existing correlations using the proximate analysis of the
torrefied biomass.
Biomass

Correlations

Ref.

MAE

AAE

ABE

Torrefied

C = 0.635FC
+ 0.460VM À 0.095ASH
H = 0.059FC + 0.060VM
+ 0.010ASH
O = 0.340FC
+ 0.469VM À 0.023ASH
C = 0.637FC + 0.455VM
H = 0.052FC + 0.062VM
O = 0.304FC + 0.476VM

[8]

8.02

13.67

À13.56

0.66

23.79

20.93

7.90

40.09

39.90

[7]

7.78
0.55
7.43

13.20
19.66
37.64

À13.01
16.34
37.46

C = 0.635FC
+ 0.460VM À 0.095ASH
H = 0.059FC + 0.060VM
+ 0.010ASH
O = 0.340FC
+ 0.469VM À 0.023ASH
C = 0.637FC + 0.455VM
H = 0.052FC + 0.062VM
O = 0.304FC + 0.476VM

[8]

2.04

4.47

À0.88

0.38

7.05

À0.60

2.26

5.69

2.00

2.09
0.38
2.27

4.63
6.87
5.71

À0.36
À1.03
2.18

Raw

[7]

elemental analysis of such new type of upgraded biomass fuel cannot be predicted by the existing correlations.
3.3. Proposed new correlations
There is no standard method to select and develop new forms of
correlation. Different forms of correlations and their purposes are
discussed in Parikh et al. [7]. However, those correlations may
not have a reasonable prediction accuracy. Therefore, in this study,
different forms of correlations proposed by both Shen et al. [8] and
Parikh et al. [7] were analyzed. Using the observation from the
scatter distribution plots, two more forms of correlation of carbon
(Eqs. 7 and 8), hydrogen (Eqs. 16 and 17), and oxygen (Eqs. 25 and
26) were also proposed and examined.
Total data used was 447, including both the raw and torrefied
biomass. The ranges of reported data for VM, FC, ASH, C, H, and O
were 13.30–93.60%, 0.67–82.74%, 0.01–48.70%, 19.12–86.28%,
0.53–7.46%, and 4.31–49.50%, respectively. Table 2 presents the
summaries of the estimation errors calculated for the published
and new proposed correlations using the data points from both
raw and torrefied biomass (Tables S1 and S2). Correlations with
the lowest MAE values and the highest R2 values are recommended
for future use in predicting the elemental compositions of torrefied
and raw biomass. The updated versions of the existing correlations,
which contents similar terms as they had in their original forms,
were also recommended for the future use (bold rows).
The ABE values for existing correlations for predicting carbon
content were observed to be negative. This supports the previous
discussion that the existing correlations cannot address the
changes occurred in the biomass after torrefaction. Since the estimation errors were calculated using both raw and torrefied biomass, they were smaller compared that to those presented in
Table 1. Observed errors were higher in Shen correlations than in
Parikh correlations. This can also be confirmed from the low R2 values for Shen Correlations with that of Parikh correlations (Table 2).

Estimation errors for 27 new forms of correlations are also presented in Table 2. The calculated values of the constant term a, b, c,
and d of all proposed correlations are presented in Table S3. Among
all the proposed new forms of correlations for the carbon content,
Eq. 12 (PS6 – bold row) has the lowest MAE value and has the highest
R2 value. The selected correlation for predicting carbon content is:

C ¼ 1:0396FC þ 0:0757VM 1:3773
Among all the proposed new forms of correlations for the
hydrogen content, the lowest MAE value was found to be 0.41 with
the maximum R2 value of 0.7. There are few other correlations that
are having similar MAE and R2 values. Therefore, the author has
selected Eq. 18 (PS12) with the lowest AAE value of 9.94. However,
one should note that this correlation having positive ABE of 2.82
will slightly overestimate the hydrogen content. One should also
note that Eqs. 19 (PS13), 20 (PS14), and 24 (PS18) can also predict
the hydrogen content with a reasonable estimation error. The
selected correlation (PS12) for predicting hydrogen content is:

H ¼ 55:3678 À 0:4830VM À 0:5319FC À 0:5600ASH
Similarly, Eq. 30 (PS24) for predicting the oxygen content was
found to have the lowest MAE and AAE, and the highest R2 value.
However, this correlation slightly overestimates the oxygen content compared that by the Eq. 26 (PS20). The selected correlation
(PS24 – bold row) for predicting oxygen content is:

O ¼ À0:0198FC þ 0:7244VM 0:9239
Knowing the fact that only two new papers [7,8] were published to predict the elemental compositions of the biomass, the
author also modified their correlations with the different constant


353

D.R. Nhuchhen / Fuel 180 (2016) 348–356
Table 2
Comparison of the estimation errors of the developed and published correlations (PS-Present Study).
SN
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

26
27
28
29
30
31
32
33

Correlation

Ref.

C ¼ 0:635FC þ 0:460VM À 0:095ASH
H ¼ 0:059FC þ 0:060VM þ 0:010ASH
O ¼ 0:340FC þ 0:469VM À 0:023ASH
C ¼ 0:637FC þ 0:455VM
H ¼ 0:052FC þ 0:062VM
O ¼ 0:304FC þ 0:476VM
C ¼ a þ bFC

[8]

C ¼ a þ bFC þ cFC 2
C ¼ a þ bVM þ cFC þ dASH
C ¼ aVM þ bFC þ cASH
C ¼ aFC þ bVM
c
C ¼ aFC þ bVM
C ¼ aFCb þ cVM

C ¼ FC þ aVM
C ¼ a þ bFC þ cVM
H ¼ a þ bVM
H ¼ a þ bVM þ cVM2
H ¼ a þ bVM þ cFC þ dASH
H ¼ aVM þ bFC þ cASH
H ¼ aFC þ bVM
c
H ¼ aFC þ bVM
H ¼ aFC b þ cVM
H ¼ FC þ aVM
H ¼ a þ bFC þ cVM
O ¼ a þ bVM
O ¼ a þ bVM þ cVM2
O ¼ a þ bVM þ cFC þ dASH
O ¼ aVM þ bFC þ cASH
O ¼ aFC þ bVM
c
O ¼ aFC þ bVM
O ¼ aFC b þ cVM
O ¼ FC þ aVM
O ¼ a þ bFC þ cVM

MAE

AAE

ABE

R2

PS1
PS2

5.33
0.54
5.36
5.22
0.47
5.11
3.66
3.63

9.53
16.26
24.62
9.35
13.90
23.28
7.65
7.60

À7.85
11.25
22.86
À7.32
8.53
21.59
1.04
1.06

0.15
0.29
0.24
0.21
0.46
0.33
0.65
0.65

PS3
PS4
PS5
PS6
PS7
PS8
PS9
PS10
PS11

2.58
2.58
2.60
2.53
2.58
2.67
2.58
0.44
0.41

5.23
5.23
5.28
5.08
5.21
5.42
5.23
10.87
10.09

0.45
0.45
0.55
0.36
0.50
0.68
0.45
3.29
2.84

0.83
0.83
0.83
0.84
0.83
0.82
0.83
0.67
0.69

PS12
PS13
PS14
PS15
PS16

0.41
0.41
0.41
0.41
0.51

9.94
9.97
10.07
10.07
11.23

2.82
2.81
3.15
3.23
0.02

0.70
0.70
0.70
0.70
0.60

PS17
PS18
PS19
PS20

10.78
0.41
2.61
2.59

284.23
9.97
8.80
8.90

153.87
2.81
2.03
1.54

NA
0.70
0.84
0.84

PS21
PS22
PS23
PS24

2.60
2.61
2.61
2.59

8.79
8.79
8.81
8.84

2.01
2.00
2.09
2.14

0.84
0.84
0.84
0.84

PS25

34.25

8.83

2.18

0.84

PS26
PS27

11.39
2.61

54.78
8.79

35.34
2.00

NA
0.84

[7]

terms (bold rows in Table 2). Such modifications on the existing
correlations are inevitable because the torrefaction process leads
to devolatilization and depolymerization reactions, releasing several light volatile gases such as CO2, CO, acetic acids, lactic acids
and so on. This upgrades the fuel qualities of biomass and alters
the fuel type. Torrefaction process, which increases the fuel ratio
(FC/VM) and decreases O/C and H/C ratios, makes biomass to
behaving like coal. The need for new correlations or modification
of existing correlations is also confirmed by the observation from
the scatter distribution plots (Figs. 1–3). Changes in the proximate
and ultimate analyses after torrefaction cannot be seen directly
from the developed correlations, but they could be only incorporated in the correlation by updating the type of materials and the
range of data points used while devising the correlation. Author
has, thus, incorporated all changes in biomass properties after torrefaction process by including a wide range of torrefied biomass

materials while developing the new correlations. The modified correlations were found to have similar estimation errors that the
selected correlations have. Therefore, the modified correlations
could also be used in predicting the elemental compositions of torrefied biomass. The modified Shen and Parikh correlations, respectively are:

Author, however, would like to emphasize here that though the
selected (three) new correlations have low estimation errors, correlations for oxygen and hydrogen do not have ash content. Since
the ash content of biomass also increases after torrefaction, it is
good to have ash content in the selected correlation. As discussed
in the earlier section, effects of ash content on the elemental compositions are very significant. One may note that the carbon content at a given ash content is higher in the torrefied biomass
compared that in the raw biomass (Fig. 1(c)) whereas the oxygen
content is lesser in the torrefied biomass compared that in the
raw biomass (Fig. 3(c)). However, one cannot select the correlation
with the ash content if the estimation error was too high. But in
Table 2, the estimation errors of Eq. 9 (PS3) are comparable with
that of the Eq. 12 (PS6). Similarly, the estimation errors of Eq. 27
(PS21) are also comparable with Eq. 30 (PS24). So correlations for
the carbon and oxygen contents of torrefied biomass can also be
predicted using Eqs. 9 and 27, respectively as:

C ¼ 0:4097VM þ 0:9671FC À 0:0348ASH

Considering the fact that torrefaction of biomass would affect
all three compositions of the proximate analysis and correlations
giving small estimation errors could predict more accurately, Eqs.
9, 18, and 27 (bold and italic rows in Table 2) containing VM, FC,
and ASH contents of biomass were selected for predicting carbon,
hydrogen, and oxygen contents of the torrefied biomass.
The author also suggests to readers of this paper that one could
also use other forms of new correlations, which are having low
estimation errors and high R2 value, to predict the elemental

H ¼ 0:0708VM þ 0:0215FC À 0:0063ASH
O ¼ 0:5147VM þ 0:0061FC À 0:0097ASH
C ¼ 0:9612FC þ 0:4093VM
H ¼ 0:0204FC þ 0:0707VM
O ¼ 0:0045FC þ 0:5146VM

C ¼ À35:9972 þ 0:7698VM þ 1:3269FC þ 0:3250ASH
O ¼ 223:6805 À 1:7226VM À 2:2296FC À 2:2463ASH


354

D.R. Nhuchhen / Fuel 180 (2016) 348–356

compositions. For that purpose, all constants of the proposed
correlations can be extracted from Table S3.
3.4. Validation and verification of the selected correlations
Another set of data points in Table S4 [65–69] (30 samples) consisting raw, torrefied, and washed biomass materials have been
adopted to validate and verify the selected new correlations (Eqs.
9, 18, and 27). The same data points were also used to compare
the prediction accuracy between the selected new and existing
correlations. The range of percentage errors for the new selected
and existing correlations is presented in Table S5. Fig. 4 shows

Predicted carbon content (Wt. %, dry basis)

(a)

100

Eq. 1 (Shen et al., 2010)
Eq. 4 (Parikh et al., 2007)

80

Eq. 9 (PS3)

60

40

20

0

0

20

40

60

80

100

(b)

Predicted hydrogen content (Wt. %, dry basis)

Measured carbon content (Wt. %, dry basis)
10

Eq. 2 (Shen et al., 2010)
Eq. 5 ( Parikh et al., 2007)

8

Eq. 18 (PS12)

6

4

2

0

4. Conclusions
0

2

4

6

8

10

Predicted oxygen content (Wt. %, dry basis)

Measured hydrogen content (Wt. %, dry basis)

(c)

60

Eq. 3 (Shen et al., 2010)
Eq. 6 (Parikh et al., 2007)

50

Eq. 27 (PS21)
40
30
20

Existing proximate analysis based correlations cannot be used
to predict the elemental compositions of torrefied biomass. Correlations presented by both Shen et al. [8] and Parikh et al. [7] require
adjustment if those correlations are going to be used in predicting
the elemental compositions of torrefied and carbonized biomass.
Total 447 data points from both raw and torrefied biomass were,
therefore, deployed to adjust existing correlations and to develop
new correlations. Considering the estimation errors of the analyzed
correlations and their ability to incorporate all changes in proximate analysis after torrefaction, following correlations were
selected to predict carbon, hydrogen and oxygen contents of the

torrefied biomass as:

C ¼ À35:9972 þ 0:7698VM þ 1:3269FC þ 0:3250ASH
H ¼ 55:3678 À 0:4830VM À 0:5319FC À 0:5600ASH

10
0

the deviation of the predicted and measured carbon content values
of the new selected and the existing correlations. The predicted
values, which are close the main solid line, indicate a better accuracy of prediction. In Fig. 4(a), predicted carbon content in the
range of 40–55% are close the straight line for both existing and
new selected correlations. But, existing correlations have more
errors for carbonized and torrefied biomass. For instance, the carbon content of the hazelnut shell was increased from 46.2% to
62.8% after the torrefaction process and 70.0% after the carbonization process. While the predicted carbon contents in torrefied
hazelnut shell using Shen and Parikh correlations were found to
be 52.3% (17% error) and 52.3% (17% error) respectively, the present model predicts the carbon content of 64.2% (<3% error). This
confirms that existing correlations are not suitable for predicting
the carbon contents of the torrefied sample. The error level is further increased in predicting the carbon content of the carbonized
hazelnut shell. Deviations of the predicted hydrogen and oxygen
contents with the measured values are also shown in Fig. 4
(b) and (c).
In Fig. 4(a), predicted carbon content deviated only for carbonized and torrefied biomass when it was predicted by using
existing correlations. The existing correlations for carbon content
can be used for raw and washed biomass materials with a good
accuracy. This is also supported by results presented in Table 1.
The proposed new correlation for oxygen content is applicable to
a wider range of data points. Fig. 4(c), however, shows that existing
correlations fail to predict the oxygen content in the torrefied and
carbonized biomass. In the Fig. 4(b), plotted points are more scattered. This indicates that hydrogen correlations have high estimation errors. Since the absolute values of hydrogen content are

smaller compared to the carbon and oxygen contents, a small deviation would also result in more scattered points. However, the
selected new correlation for hydrogen is suitable for wider range
of data compared that to the existing correlations. From Fig. 4(c),
one can also see that the error level in predicting the oxygen content of microalgal residue and microalgae is significantly higher
compared that to other biomass types. This high error was mainly
from the high nitrogen content of the microalgae. So author recommends to the reader of this paper that the correlations based on the
proximate analysis are not suitable for a more nitrogenous
biomass.

O ¼ 223:6805 À 1:7226VM À 2:2296FC À 2:2463ASH
0

10

20

30

40

50

60

Measured oxygen content (Wt. %, dry basis)
Fig. 4. Validation and comparison of the proposed correlation with the existing
correlations for (a) Carbon, (b) Hydrogen, and (c) Oxygen.

Selected correlations were also validated and verified using
another set of data with a variety of material types. The new correlations, which is based on a wider range of data points than the

existing correlations, have a better prediction power compared
that to the existing correlations. Though existing correlations can


D.R. Nhuchhen / Fuel 180 (2016) 348–356

still be used in predicting the elemental compositions of raw biomass, only new selected correlations should be deployed for the
torrefied biomass. Selected new correlations would be of a great
interest in the present context where there is a growing research
on biomass torrefaction and an increasing experimental cost of
the elemental analysis. At the end, one should be very careful to
deploy such correlations to predict the elemental composition of
biomass with high nitrogen contents.

Appendix A. Supplementary material
Supplementary data associated with this article can be found, in
the online version, at />
References
[1] Nhuchhen DR, Salam PA. Estimation of higher heating value of biomass from
proximate analysis: a new approach. Fuel 2012;99:55–63.
[2] Channiwala SA, Parikh PP. A unified correlation for estimating HHV of solid,
liquid and gaseous fuels. Fuel 2002;81:1051–63.
[3] Demirbas A. Calculation of higher heating values of biomass fuels. Fuel
1997;76:431–4.
[4] Cordero T, Marquez F, Rodriquez-Mirasol J, Rodriguez JJ. Predicting heating
values of lignocellosic and carbonaceous materials from proximate analysis.
Fuel 2001;80:1567–71.
[5] Fernandez P, Diaz RM, Xiberta J. Correlations of properties of Spanish coals
with their natural radionuclides contents. Fuel 1997;76(10):951–5.
[6] Kucukbayrak S, Durus B, Mericboyu AE, Kadioglu E. Estimation of calorific

values of Turkish lignites. Fuel 1991;70:979–81.
[7] Parikh J, Channiwala SA, Ghosal GK. A correlation for calculating elemental
composition from proximate analysis of biomass materials. Fuel
2007;86:1710–9.
[8] Shen J, Zhu S, Liu X, Zhang H, Tan J. The prediction of elemental composition of
biomass on proximate analysis. Energy Convers Manage 2010;51:983–7.
[9] Vakkilainen EK. Estimation of elemental composition from proximate analysis
of black liquor. Paper JA PUU – Paper Timber 2000;82(7):450–4.
[10] Acharya B, Sule I, Dutta A. A review on advances of torrefaction technologies
for biomass processing. Biomass Convers Bioref 2012;2(4):349–69.
[11] Chen WH, Peng J, Bi XT. A state-of-the-art of biomass torrefaction,
densification and applications. Renew Sustain Energy Rev 2015;44:847–66.
[12] Nhuchhen DR, Basu P, Acharya B. A comprehensive review on biomass
torrefaction. Int J Renew Energy Biofuels 2014;2014:1–56.
[13] Tumuluru JS, Wright CT, Hess JR, Kenney KL. A review of biomass densification
systems to develop uniform feedstock commodities for bioenergy application.
Biofuels Bioprod Bioref 2011;5:683–707.
[14] van der Stelt MJC, Gerhauser H, Kiel JHA, Ptasinski KJ. Biomass upgrading by
torrefaction for the production of biofuels: a review. Biomass Bioenergy
2011;35(9):3748–62.
[15] Anca-Couce A, Obernberger I. Application of a detailed biomass pyrolysis
kinetic scheme to hardwood and softwood torrefaction. Fuel
2016;167:158–67.
[16] Prins MJ. Thermodynamic analysis of biomass gasification and torrefaction.
PhD Thesis, Eindhoven University, Netherland; 2005.
[17] Nhuchhen DR, Basu P. Experimental investigation of mildly pressurized
torrefaction in Air and Nitrogen. Energy Fuels 2014;28(5):3110–21.
[18] McNamee P, Darvell LI, Jones JM, Williams A. The combustion characteristics of
high-heating-rate chars from untreated and torrefied biomass fuels. Biomass
Bioenergy 2015;82:63–72.

[19] Shoulaifar TK, DeMartini N, Ivaska A, Fardim P, Hupa M. Measuring the
concentration of carboxylic acid groups in torrefied spruce wood. Bioresour
Technol 2012;123:338–43.
[20] Nhuchhen DR, Basu P, Acharya B. Torrefaction of Poplar in a continuous twostage indirectly heated rotary torrefier. Energy Fuels 2016;30(2):1027–38.
[21] Soponpongpipat N, Sittikul D, Sae-Ueng U. Higher heating value prediction of
torrefaction char produced from non-woody biomass. Front Energy 2015;9
(4):461–71.
[22] Strandberg M, Olofsson I, Pommer L, Wiklund-Lindstrom S, Aberg K, Nordin A.
Effects of temperature and residence time on continuous torrefaction of spruce
wood. Fuel Process Technol 2015;134:387–98.
[23] Chen D, Zheng Z, Fu K, Zeng Z, Wang J, Lu M. Torrefaction of biomass stalk and
its effect on the yield and quality of pyrolysis products. Fuel 2015;159:27–32.
[24] Nam H, Capareda S. Experimental investigation of torrefaction of two
agricultural wastes of different composition using RSM (response surface
methodology). Energy 2015;91:507–16.
[25] Chen D, Zhou J, Zhang Q, Zhu X, Lu Q. Upgrading of rice husk by torrefaction
and its influence on the fuel properties. Bioresources 2014;9(4):5893–905.
[26] Thanh KL, Commandre JM, Valette J, Volle G, Meyer M. Detailed indenfication
and quantification of the condensable species released during torrefaction of
lignocellulosic biomasses. Fuel Process Technol 2015;139:226–35.

355

[27] Kambo HS, Dutta A. Comparative evaluation of torrefaction and hydrothermal
carbonization of lignocellulosic biomass for the production of solid biofuel.
Energy Convers Manage 2015;105:746–55.
[28] Zhang S, Dong Q, Zhang Li, Xiong Y, Liu X, Zhu S. Effects of water washing and
torrefaction pretreatments on rice husk pyrolysis by microwave heating.
Bioresour Technol 2015;193:442–8.
[29] Bach QV, Tran KQ. Dry and wet torrefaction of woody biomass – a comparative

study on combustion kinetics. Energy Proced 2015;75:150–5.
[30] Bach QV, Tran KQ, Skreiberg O, Trinh TT. Effects of wet torrefaction on
pyrolysis of woody biomass fuels. Energy 2015;88:443–56.
[31] Park J, Meng J, Lim KH, Rojas OJ, Park S. Transformation of lignocellulosic
biomass during torrefaction. J Anal Appl Pyrol 2013;100:199–206.
[32] Yang H, Liu B, Chen Y, Chen W, Yang Q, Chen H. Application of biomass
pyrolytic polygeneration technology using retort reactors. Bioresour Technol
2016;200:64–71.
[33] Bach QV, Tran KQ, Skreiberg O. Accelerating wet torrefaction rate and ash
removal
by
carbon
dioxide
addition.
Fuel
Process
Technol
2015;140:297–303.
[34] Ibrahim RHH, Darvell LI, Jones JM, Williams A. Physicochemical
characterization of torrefied biomass. J Anal Appl Pyrol 2013;103:21–30.
[35] Barontini F, Biagini E, Bonini F, Tognotti L. An experimental investigation on
the devolatization behavior of raw and torrefied lignocellulosic biofuels. Chem
Eng Trans 2015;43. xxx–xxx. ISSN 2283-9216.
[36] Cao L, Yuan X, Li H, Li C, Xiao Z, Jiang L, et al. Complementary effects of
torrefaction and co-pelletization: energy consumption and characteristics of
pellets. Bioresour Technol 2015;185:254–62.
[37] Costa FF, Costa M. Evaluation of particle fragmentation of raw and torrefied
biomass in a drop tube furnace. Energy Proced 2015;66:277–80.
[38] Pohlmann JG, Osorio E, Vilela ACF, Diez MA, Borrego AG. Integrating
physicochemical information to follow the transformation of biomass upon

torrefaction and low temperature carbonization. Fuel 2014;131:17–27.
[39] Xue G, Kwapinska M, Kwapinski W, Czajka KM, Kennedy J, Leahy JJ. Impact of
torrefaction on properties of Miscanthus x giganteus relevant to gasification.
Fuel 2014;121:189–97.
[40] Li J, Zhang X, Pawlak-Kruczek H, Yang W, Kruczek P, Blasiak W. Process
simulation of co-firing torrefied biomass in a 220 MWe coal-fired power plant.
Energy Convers Manage 2014;84:503–11.
[41] Berrueco C, Recari J, Guell BM, Alamo GD. Pressurized gasification of torrefied
woody biomass in a lab scale fluidized bed. Energy 2014;70:68–78.
[42] Yang Z, Sarkar M, Kumar A, Tumuluru JS, Huhnke RL. Effects of torrefaction and
densification on switchgrass pyrolysis products. Bioresour Technol
2014;174:266–73.
[43] Pala M, Kantarli IC, Buyukisik HB, Yanik J. Hydrothermal carbonization and
torrefaction of grape pomace: a comparative evaluation. Bioresour Technol
2014;161:255–62.
[44] Eseltine D, Thanapal SS, Annamalai K, Ranjan D. Torrefaction of woody
biomass (Juniper and Mesquite) using inert and non-inert gases. Fuel
2013;113:379–88.
[45] Wannapeera J, Worasuwannarak N. Upgrading of woody biomass by
torrefaction under pressure. J Anal Appl Pyrol 2012;96:173–80.
[46] Park SW, Jang CH, Baek KR, Yang JK. Torrefaction and low-temperature
carbonization of woody biomass: evaluation of fuel characteristics of the
products. Energy 2012;45:676–85.
[47] Lu KM, Lee WJ, Chen WH, Liu SH, Lin TC. Torrefaction and low temperature
carbonization of oil palm fiber and eucalyptus in nitrogen and air
atmospheres. Bioresour Technol 2012;123:98–105.
[48] Bridgeman TG, Jones JM, Williams A, Waldron DJ. An investigation of the
grindability of two torrefied energy crops. Fuel 2010;89:3911–8.
[49] Arias B, Pevida C, Fermoso J, Plaza MG, Rubiera F, Pis JJ. Influence of
torrefaction on the grindability and reactivity of woody biomass. Fuel

Process Technol 2008;89:169–75.
[50] Wannapeera J, Fungtammasan B, Worasuwannarak N. Effects of temperature
and holding time during torrefaction on the pyrolysis behaviors of woody
biomass. J Anal Appl Pyrol 2011;92:99–105.
[51] Parikh J, Channiwala SA, Ghosal GK. A correlation for calculating HHV from
proximate analysis of solid fuels. Fuel 2005;84:487–94.
[52] Yin CY. Prediction of higher heating values of biomass from proximate and
ultimate analyses. Fuel 2011;90:1128–32.
[53] Phichai K, Pragrobpondee P, Khumpart T, Hirunpraditkoon S. Prediction
heating values of lignocellulosics from biomass characteristics. Int J Chem
Mole Nucl Mater Metall Eng 2013;7:214–7.
[54] PhanPhanich M, Mani S. Impact of torrefaction on the grindability and fuel
characteristics of forest biomass. Bioresour Technol 2011;102:1246–53.
[55] Xiao R, Chen X, Wang F, Yu G. Pyrolysis pretreatment of biomass for entrainedflow gasification. Appl Energy 2010;87:149–55.
[56] Mafu LD, Neomagus HWJP, Everson RC, Carrier M, Strydom CA, Bunt JR.
Structural and chemical modifications of typical South African biomasses
during torrefaction. Bioresour Technol 2016;202:192–7.
[57] Chen WH, Cheng WY, Lu KM, Huang YP. An evaluation on improvement of
pulverized biomass property for solid fuel through torrefaction. Appl Energy
2011;88:3636–44.
[58] Chen WH, Hsu HC, Lu KM, Lee WJ, Lin TC. Thermal pretreatment of wood
(Luan) block by torrefaction and its influence on the properties of the biomass.
Energy 2011;36:3012–21.
[59] Rousset P, Aguiar C, Labbe N, Commandre JM. Enhancing the combustible
properties of bamboo by torrefaction. Bioresour Technol 2011;102:8225–31.


356

D.R. Nhuchhen / Fuel 180 (2016) 348–356

[60] Huang YF, Chen WR, Chiueh PT, Kuan WH, Lo SL. Microwave torrefaction of
rice straw and pennistum. Bioresour Technol 2012;123:1–7.
[61] Wang MJ, Huang YF, Chiueh PT, Kuan WH, Lo SL. Microwave-induced
torrefaction of rice husk and sugarcane residues. Energy 2012;37:177–84.
[62] Bach QV, Tran KQ, Khalil RA, Skreiberg O. Comparative assessment of wet
torrefaction. Energy Fuels 2013;27:6743–53.
[63] Ren S, Lei H, Wang L, Bu Q, Wei Y, Liang J, et al. Microwave torrefaction of
douglass fir sawdust pellets. Energy Fuels 2012;26:5936–43.
[64] Deng J, Wang GJ, Kuang JH, Zhang YL, Luo YH. Pretreatment of agricultural
residues for co-gasification via torrefaction. J Anal Appl Pyrol 2009;86:331–7.
[65] Bilgic E, Yaman S, Haykiri-Acma H, Kucukbayrak S. Is torrefaction of
polysaccharides-rich biomass equivalent to carbonization of lignin-rich
biomass? Bioresour Technol 2016;200:201–7.

[66] Bach QV, Tran KQ, Skreiberg O. Hydrothermal pretreatment of fresh forest
residues: effects of feedstock pre-drying. Biomass Bioenergy 2016;85:76–83.
[67] Tapasvi D, Kempegowda RS, Tran KQ, Skreiberg O, Gronli M. A simulation
study on the torrefied biomass gasification. Energy Convers Manage
2015;90:446–57.
[68] Wu KT, Tsai CJ, Chen CS, Chen HW. The characteristics of torrefied microalgae.
Appl Energy 2012;100:52–7.
[69] Deng L, Zhang T, Che D. Effect of water washing on fuel properties, pyrolysis
and combustion characteristics, and ash fusibility of biomass. Fuel Process
Technol 2013;106:712–20.