How to trace back an unknown production temperature of biochar from chemical characterization methods in a feedstock independent way

Abstract Besides the feedstock composition, the highest treatment temperature (HTT) in pyrolysis is one of the key production parameters. The latter determines the feedstock’s carbonization extent, which influences physicochemical properties of the resulting biochar, and in consequence its performance in industrial and agricultural applications. The actual HTT of biomass is difficult to measure in a reliable manner in many large-scale pyrolysis units (e.g., rotary kilns). Therefore, producers and end-users often rely on unreliable or biased information regarding this key production parameter that affects biochar quality. Data from indirect chemical assessment methods of biochar’s carbonization extent correlate well with the highest treatment temperature. Therefore, this study demonstrates that the HTT can be accurately assessed posteriori and feedstock-independently via a simple-to-use model based on biochar characteristics related to the carbonization extent. For that purpose, 24 contrasting biochars from 12 different feedstocks produced in the most common production temperature range of 350−700 °C were analysed using 5 different established biochar chemical characterization methods. Then, experimental data was used to establish a multilinear regression model capable of correlating the HTT, which was successfully validated for external datasets. The correlation accuracy for biochars of various origin (lignocellulosic, manure) was satisfactorily high (R2adj. = 0.853, RSME =47 °C). The obtained correlation proved that the HTT can be predicted feedstock independently with the use of basic input data. It also provides a quick, simple, and reliable tool to verify the HTT of a given biochar.

biochar from mineral-rich feedstocks (i.e. crop residues and processed waste materials like manures and sewage sludge) obtained under the same processing conditions [5][6][7][8][9][10].The impact of production-dependent parameters, especially the HTT in pyrolysis on the aromaticity and extent of charring is more comprehensible. It is well known that upon increasing the HTT, a progressive elimination of heteroatoms (through dehydration, decarbonylation and decarboxylation reactions) occurs [11], along with rearrangements (i.e. poly-condensation reactions) in the carbonaceous structure that promote the formation of (poly)aromatic clusters [8,12,13]. Moreover, an increase in temperature increases the degree of aromatic condensation (i.e. the cluster size and the purity of the aromatic structure) as observed through 13 C NMR spectroscopy [8,14,15]. As a result, biochar obtained at higher HTT features particular levels in the aromaticity and degree of aromatic condensation which are not observed in biochar produced at a lower temperature [8]. Unfortunately, the 13 C NMR spectroscopy analysis method, despite its accuracy and reliability, requires expensive instruments, which additionally are not straightforward to use. Therefore, relatively simple and low-cost biochar chemical characterization methods were pursued and introduced, whose role is to indirectly assess the carbonization level of biochar in a less accurate, yet less time-cost expensive manner.
The simplest and most frequently used ones are based on the elemental and proximate analysis, such as H/C molar ratio or fixed carbon content (FC) on a dry basis [16]. Considering that the most stable carbonaceous material is anthracite/graphite with a very well-developed structural organisation and whose H/C is very low and with a FC content close to 100%, other carbonaceous materials can be ranked according to their carbonization level in relation to these reference materials. The R50 stability proxy is based on a very similar basis [17]. Another, relatively new method is the Edinburg stability tool (AE), which assess the resistance to chemical oxidation of biochar C [18]. It assumes that the better-developed structure, i.e. a more aromatic char, is more resistant to mineralisation, hence more stable. More complex chemical indicators J o u r n a l P r e -p r o o f are the ones obtained via analytical pyrolysis (Py-GC/MS), such as the benzene to toluene ratio (B/T ratio). Analytical pyrolysis methods are based on the assumption that more recalcitrant carbonaceous structures release less oxygenated or branched aliphatic compounds, as these compounds should already have been released upon the actual char production process. As it can be noticed, all the mentioned biochar characterization methods are indirectly related with the carbonaceous material structural organisation (e.g. aromatization and the extent thereof).
Since changes in the degree of aromatic condensation can occur partially feedstockindependently, the HTT could be considered as a basic indicator of the extent of the biochar's aromatization. Therefore, considering a large-scale production, it could be useful to biochar end-users, producers, and certifiers to know the actual temperature in which biomass was converted. The aim of this study is to create a simple-to-use correlation based on easy-tomeasure properties of given biochar, which would allow for quick assessment of its HTT after production. For this purpose, this study assesses the feedstock-independent nature of various established biochar characterization methods described in literature via statistical tools like principal component analysis (PCA). Then, the characterization methods are checked in terms of their predictive power and reliability. This study provides a multilinear correlation between selected predictors and HTT. The obtained MLR model is then validated against various external datasets to assess its accuracy and usefulness.

Biochar materials
A set of 24 biochar samples with contrasting properties which are produced using lab-scale biochar production reactors was used. They were produced using 12 different feedstocks at 10 different production temperatures with varying heating rates and residence times. The dataset also contained 8 thermo-sequences (groups of biochars from the same feedstock but produced J o u r n a l P r e -p r o o f 8 at different pyrolysis temperature). An overview of the biochars applied in this study is shown in Table 1. All samples used in this study were supplied by the UK Biochar Research Centre.

Elemental analysis
The mass fractions of carbon, nitrogen, hydrogen on dry basis (wt.%, db) were determined in triplicate, using a Flash 2000 elemental analyser (Thermoscientific, USA). The samples were pre-dried overnight at 105 °C prior to the elemental analysis. The oxygen mass fraction was calculated by difference.

Proximate analysis
Proximate analysis of biochars was determined in triplicate using TGA [19]. In brief, the moisture content of biochar was obtained from the mass loss upon heating from 30 °C to 110 °C at a heating rate of 25 °C/min and holding at 110 °C for 10 minutes. The volatile matter content on dry basis was determined from the weight loss upon heating from 110 °C at 25 °C/min to 900 °C and holding at 900 °C for 10 minutes. Moisture and volatile matter content determination were carried out in an inert N2 atmosphere, with 50 ml/min flow rate. The ash content on dry basis was determined from the weight curve after switching the carrier gas from N2 to air (same flow rate) and after being kept at 900 °C for 20 minutes. Fixed carbon content on dry basis was obtained by difference.

Thermal recalcitrance index (R50)
Determination of the R50 index from TGA was done according to the procedure described in Harvey et al. [17]. Measurement was done in duplicate. A 70 µl aluminium crucible was fully filled with ca. 10-15 mg biochar (or ca. 5 mg for low-density biochars). Each sample was then heated from 30 °C to 1000 °C with a heating rate of 10 °C/min under Nitrogen flow rate of 10 ml/min. Resulting TG profiles were corrected for moisture and ash contents and thermal recalcitrance index (R50) was obtained using the following equation: where 50, is the temperature at which 50% of the sample mass was oxidized (lost), while 50, ℎ is an external standardization factor and corresponds to the temperature at which 50% of a graphite sample is oxidized ( 50, ℎ = 885 °C) [17].

Edinburgh stability tool
The Edinburgh stability tool, i.e. accelerated aging of biochar, was performed as described by Cross and Sohi [18].
Where denotes the residual mass of biochar after oxidation, denotes the mass fraction of carbon (wt. %, db) in the residual biochar after oxidation, denotes the initial mass of biochar and denotes the corresponding carbon mass fraction (wt. %, db).

Pyrolysis-GC-MS analysis
Micro-pyrolysis experiments of biochar were performed using a micro-pyrolysis unit (Multishot pyrolyser EGA/PY-3030D, Frontier Laboratories Ltd.) coupled to a gas chromatograph (Thermo Fisher Scientific Trace GC) -mass spectrometer (Thermo ISQ MS). Samples were analysed according to the procedure described in Suarez-Abelenda et al. [20]. In brief, ca. 0.5 mg of finely ground and well homogenized biochar sample was loaded into a sample cup, which Ratios between the specific compounds evolved in the Py-GC/MS analysis applied in this study are calculated as the ratio of the relative peak areas of each compound.

Principal component analysis (PCA)
PCA on different datasets was performed in R Studio  [21]. Next to HTT, biochar properties are also influenced by the retention time, albeit to a lesser extent. However, in small-scale reactors with few heat transfer limitations, Ronsse et al. [22] found no significant differences in elemental and proximate composition in biochars produced with varying retention time (>10 min) once the HTT was 450 °C and above and using lignocellulosic feedstocks. With the exception of the SP-350.1 biochar, all biochars in the dataset being produced at short RT's have been produced at higher temperatures. Hence, the retention time was deemed not significantly influential and as such not included in the model.
The selection of the parameters (indicators based on the characterization methods for carbonization extent) for the temperature prediction model was done by the following sequence.
First, indicators' correlations to the production temperature were identified through the determination coefficient (R 2 ). The indicators showing a R² value higher than 0.3 were retained as MLR candidate parameters. Moreover, multicollinearity in the dataset was avoided by J o u r n a l P r e -p r o o f considering the variance inflation factor (VIF) test. Parameters with a VIF value above 5 were removed, resulting in the final set of parameters from which MLR+ANOVA analysis started [23,24].
J o u r n a l P r e -p r o o f

Results and discussion
Results from the elemental and proximate analysis, thermal recalcitrance index (R50) and Edinburgh stability tool (AE) measurements are presented in Table 2.

3.1.Elemental and proximate analysis
Results of elemental and proximate analysis showed a significant difference between the biochar samples tested. The same typically observed trends with increasing pyrolysis temperature, such as relative C enrichment, increase in FC content and reduction of VM content, were observed in the studied thermo-sequences (Table 2), especially for biochar produced form lignocellulosic feedstock. Figure 1 shows a van Krevelen diagram of the investigated samples, with indication of proposed International Biochar Initiative (IBI) and European Biochar Certificate (EBC) limits ( ≤ 0.7 H/Corg and ≤ 0.4 O/C) for stable biochar [16,25]. According to the IBI and EBC guidelines, it is recommended to do an acid treatment prior to organic C determination in order to avoid the impact from inorganic carbon species [16,25], but this acid treatment was not applied in this study. The data in Figure 1 is presented with the assumption that all C from elemental analysis can be considered as organic C.

3.3.Edinburgh stability tool (AE)
The Edinburgh stability tool (AE) depicts the oxidative degradation of biochar in soil . Moreover, it can be used as a proxy for the environmental aging of approximately 100 years under temperate conditions [18]. According to Crombie et al. [26] the stable carbon fraction in biochar increases with the biochar production temperature due to the elimination of the volatile fraction.
Results of the Edinburgh stability tool in this study (

Py-GC/MS analysis
Analytical pyrolysis allows thermal degradation of the compounds under inert atmosphere [27].
Hence, it provides information regarding the biomolecular composition of chars [28]. Pyrolysis product ratios obtained through Py-GC/MS analysis is shown in Table 3. Typically, benzene, toluene, ethylbenzene, PAHs, and phenols are predominantly presented in pyrograms of biochar [29,30]. Therefore, these compounds and their homologues with alkyl side chains can be transformed into ratios. Next, they can be used as an indicator of the degree of thermal alteration and dealkylation in the pyrolysis products [20,27]. Due to the significant thermal stability of the char produced at high HTT, their pyrograms are characterized with fewer pyrolysis products out of which benzene is the predominant one [28,31,32].
J o u r n a l P r e -p r o o f Therefore, the B/T ratio derived from Py-GC/MS analysis was used as an indicator to assess carbonization level of biochar in several studies and showed a good correlation with the biochar HTT [20,27,29,30]. In this study as well, the B/T ratio of biochars showed a good positive correlation with the biochar HTT (R 2 = 0.78). However, it is not that much stronger as previously reported [20,27,29,30]. This may be due to the diversity of the biochar feedstock material used in this study. Suarez-Abelenda et al. [20] reported that biochars from N rich, hence protein-rich feedstocks produced at low temperatures are able to introduce bias into the measured B/T ratio via the addition of toluene derived from incompletely converted protein, especially the amino acid phenylalanine produces toluene upon pyrolysis. Moreover, in this Phenol tends to be increasingly released from chars treated between 400 °C to 800 °C due to demethoxylation of methoxyphenols (as decomposition products from lignin) and starts to decrease at 800 °C because of phenol dehydroxylation [27]. However, none of these ratios showed strong correlation with biochar HTT.

PCA on combined indicators derived through biochar characterization
PCA was conducted to see the relationship between production temperature and different    With the aim to build a multilinear model to correlate HTT to biochar carbonization extent indicators, only those predictors that showed a feedstock-independent correlation were retained.
Hence, a threshold value of 0.3 for the determination coefficient (R 2 ) between predictor for the whole dataset and production temperature was set. The threshold translates to an absolute Pearson correlation coefficient of >0.5 (existence of a correlation). As a result, ash content, Py-GC/MS ratios of Ph/B, EtB/Ph and B/EtB were no longer retained as HTT predictors. These predictors also correspond to those which explained low variance for PC1 and high variance for PC2 in PCA.
J o u r n a l P r e -p r o o f

Analysis of repeatability and reliability of R50, AE and B/T ratio
Since the results of the elemental and proximate analysis had been proven through numerous publications to be consistent and reliable [26,33], these predictors do not require additional analysis and can be retained in the construction of a multilinear regression model further on.
The more complex, and less common indicators such as R50, AE and B/T ratio require additional checking to confirm that they are consistent among different datasets. The mentioned indicators were mutually correlated with other feedstock-independent predictors, using external data. By doing so, (i) it was assessed which predictors were not biased by the applied methodology, hence, which were reliable and repeatable and (ii) correlations were obtained to replace these complex predictors.
In the comprehensive review of Klasson [33], a correlation between R50 and Cdaf had been introduced as shown in eq. (4). The correlation was built on experimental data of lignocellulosic biochar from Harvey et al. [17], which summarise the data from other authors [10,34,35]. Figure 3 shows experimental data from this study, along with data from Windeatt et al. and Harvey et al. [17,36] with the correlation proposed by Klasson [33]. Almost all experimental data points from this study are consistent with the literature sources ( Figure 3). It shows that biochars from this study having a certain Cdaf showed the same R50 comparable with literature data. It proves that R50 can be used as a reliable and repeatable predictor. Additionally, it can be stated that the correlation provided by Klasson [33] is stable (R 2 for 3 different datasets = 0.72) and can be applied for biochar originating from lignocellulosic, manure and algae biomass.
In the work of Klasson, (2017) [33] is also presented a correlation between the AE and molar O/C ratio, shown in eq. (5). This correlation had been established using the data of lignocellulosic biochars from Crombie et al. [26]. Figure 4 shows experimental data from this study and from Crombie et al. [26] with the correlation proposed by Klasson [33].  Use of the AE as predictor is therefore only reliable and repeatable for the L and M derived biochars. When merging the L+M datasets, the accuracy of eq. (5) is getting lower (R 2 = 0.542) and there is a tendency to underpredict the AE value. Nevertheless, the correlation is still satisfactory, and that parameter showed acceptable accuracy and reproducibility.
The last complex predictor investigated in this assessment is the B/T ratio, originating from Py-GC/MS data. In literature reports [28,32,37], the B/T value of biochar can be found, but only few have been obtained with the same analytical procedure. Since the Py-GC/MS method is very sensitive to measurement conditions, only data from similar procedures can be compared. Figure 5 compares this study's B/T ratio and literature data obtained using the same procedure.
It is worth mentioning that Kaal et al. and Pereira et al. [28,29] only used lignocellulosic derived biochars, but Suarez-Abelenda et al. [20] included manure and algae derived biochars in their dataset. [ 26] This study [L+M] This study [A] This study [W] eq. (5) Figure 5. Comparison between B/T ratio data from this study and literature sources.
As Figure 5 shows, the B/T ratios in this study are for every HTT, on average, several times higher than those from the literature sources. This is most likely due to the different analytical instruments used. The B/T ratios from the works of other studies consider here [20,28,29,31] were obtained on the Pyroprobe series 5000 (CDS analytics) pyrolyzer connected to an HP- [20] [29] [28] [31] This dataset J o u r n a l P r e -p r o o f Closer data analysis indicates that the results from this study and literature show similar trends with the treatment temperature, albeit with different magnitude. The best fit between B/T ratio and HTT is obtained through an exponential function. Hence, it can be concluded that the B/T ratio suffers from two major issues. One is being the poor reproducibility in terms of using different analytical setups; the second is being the non-linearity. Therefore, its incorporation into a multilinear model would be in contradiction to the principles of linear model construction.
For this reason, it was decided not to retain the B/T ratio in the selected set of the temperature predictors for the MLR.

Model calibration
The initial predictors that were accurate, reliable, and repeatable were retained, being: Cdaf, Despite the inhomogeneous input dataset, the model showed a R 2 adj. higher than 0.85 and a root mean squared error (RSME) lower than 50 °C. Among the predictors, the AE had the strongest relative influence (>50%) on the predicted outcome. An accurate measurement of the AE value is therefore likely to result in a higher accuracy of prediction of production temperature (supplementary information, section D).
J o u r n a l P r e -p r o o f

Model validation
To prove the model's reliability and usefulness, it was validated against literature data.
However, no literature datasets were found that contained simultaneously both R50 and AE values. Therefore, datasets with the missing parameters were completed using the appropriate auxiliary equations (section 3.6.2). It needs to be emphasized that all the production temperatures, specified in literature, are regarded as the HTT, despite the lack of complete certainty of it and possible introduction of a random error to the model's prediction.
For a first validation of the MLR model, data from lignocellulosic biochars from Crombie et al. [26] was used, containing experimental values of the AE. The value of the R50 (which was not present in the original dataset) used for validation was calculated from eq. (4). The validation results are presented in supplementary information (section E). The obtained value of the R 2 is 0.843 and of the RSME is 63°C. The model very accurately predicted the HTT for pine wood derived biochar, and a moderate accuracy for rice husk and wheat straw derived biochar was obtained, presumably to the higher ash content found in those biochars.
The validation dataset contained data of biochars derived from lignocellulosic (L), manure and manure mixed with lignocellulosic biomass (M). This dataset lacked values of R50 and AE, which for validation purposes were calculated using eq. (4) and eq. (5). The validation results and residuals are presented in supplementary information (section E and F). Considering that the model's predictions were solely based on data from elemental and proximate analysis, the overall model performance is more than satisfactory. The accuracy of HTT prediction for lignocellulosic derived biochars was slightly higher than for manure and the mixture dataset.
This was likely due to the greater share of lignocellulosic derived biochars in the training dataset. The model predicts the HTT in the range between 350 °C and 700 °C with the highest accuracy, but still a small over-estimation is noticed in the middle of the mentioned range.

J o u r n a l P r e -p r o o f
Results also show rapid accuracy loss beyond both ends of the range. It is strongly related with the training dataset's temperature range, which did not contain samples produced below 350 °C and only one sample produced above 700 °C ( Figure 6). Figure 6. Comparison between measured HTT and predicted HTT.

Model summary
The summarised outcome of both model validations is presented in Table 4. J o u r n a l P r e -p r o o f correlation is reliable and to some point applicable to various biochars obtained from lignocellulosic biomass and manure. Eq. (6) presents the temperature correlation obtained in this study. The presented model predicts the HTT very well for biochars produced in the common biochar production temperature range of between 350 °C and 700 °C with typical biochar ash and fixed carbon contents.
[ ] = −437.2 O/C + 495.9 R50 + 447.3 AE Also, the application of the equations proposed by Klasson [33] allows for temperature prediction in datasets lacking R50 and AE data. The combination of eq. (6) with the correlations in eq. (4) and eq. (5), yielding eq. (7) allows the fairly accurate prediction of biochar's HTT based solely on elemental and proximate analysis data. Where, value is in the percent.
[ ] = 555 + 2 − 1440 / However, if the used dataset is completed with experimental data for R50 and AE, higher accuracy is expected. Indeed, using the correlations in eq. (4) or eq. (5) introduces additional variance, considering their R 2 with 0.719 and 0.727, respectively.

Conclusions
Strong inter-correlation between HTT used in biochar production and characterization data was