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

Elemental compositions of biomass are essential for designing energy conversion systems. Only a fewcorrelations to estimate the elemental compositions using the proximate analysis of raw biomass havebeen 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 correlationscan be deployed or not for estimating carbon, hydrogen, and oxygen compositions of the torrefiedbiomass. 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 elementalcompositions of the torrefied biomass.

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 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 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: dy998952@dal.ca, dayashyad@hotmail.com http://dx.doi.org/10.1016/j.fuel.2016.04.058 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 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 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 À N X ðPi À M i Þ2 i¼1 Fcraw 80 60 40 20 N !, , N N X  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 Results and discussion (b) 100 80 20 40 60 80 100 VMtor VMraw 3.1 Scatter distribution of data 60 40 20 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 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 10 20 30 40 50 Ash content (% Wt, dry basis) Fig 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) FCtor FCtor Oxygen content (wt %, dry basis) Hydrogen content (wt %, dry basis) Fcraw 48 Fcraw 40 32 24 16 0 20 40 60 80 20 100 40 60 80 100 Fixed carbon content (% Wt, dry basis) Fixed carbon content (% Wt, dry basis) (b) 56 (b) VMtor Oxygen content (% wt., dry basis) Hydrogen content (% wt., dry basis) VMtor VMraw 48 VMraw 40 32 24 16 0 20 40 60 80 100 20 40 60 80 100 Volatile matter content (% Wt, dry basis) Volatile matter content (% Wt, dry basis) (c) 56 (c) ASHtor Oxygen content (wt %, dry basis) Hydrogen content (wt %, dry basis) ASHtor ASHraw 48 40 32 24 16 0 10 20 30 40 50 Ash content (% Wt, dry basis) Fig 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 10 20 30 40 50 Ash content (% Wt, dry basis) Fig 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 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 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 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 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 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 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 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 Comparison of the estimation errors of the developed and published correlations (PS-Present Study) SN 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 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 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 (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 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 shows Predicted carbon content (Wt %, dry basis) (a) 100 Eq (Shen et al., 2010) Eq (Parikh et al., 2007) 80 Eq (PS3) 60 40 20 0 20 40 60 80 100 (b) Predicted hydrogen content (Wt %, dry basis) Measured carbon content (Wt %, dry basis) 10 Eq (Shen et al., 2010) Eq ( Parikh et al., 2007) Eq 18 (PS12) 4 Conclusions 10 Predicted oxygen content (Wt %, dry basis) Measured hydrogen content (Wt %, dry basis) (c) 60 Eq (Shen et al., 2010) Eq (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 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% (