Prediction of methane yield and pretreatment efficiency of lignocellulosic biomass based on composition

Lignocellulosic biomass is considered a key resource for the future expansion of biogas production through anaerobic digestion (AD), and research on the development of pretreatment technologies for improving biomass conversion is an intensive and fast-growing field. Consequently, there is a need for creating tools able to predict the efficiency of a certain pretreatment on different biomass types, fast and accurately, and to assist in selecting a pretreatment technology for a specific biomass. In this study, seven different types of raw lignocellulosic biomass of industrial relevance were systematically analyzed regarding their composition (carbohydrates, lignin, lipids, ash, extractives, etc.) and subjected to a common pretreatment. The aim of the study was to identify the most important characteristics that make a biomass good receptor of the specific pretreatment prior to AD. A simple ammonia pretreatment was chosen as a case study and partial least squares regression (PLS-R) was used for modeling initially the ultimate methane yield of raw and pretreated biomass. In the sequel, PLS-R was used for modeling the efficiency of the pretreatment on increasing the ultimate methane yield and hydrolysis rate as a function of the biomass composition. The fit of the models was satisfactory, ranging from R 2 = 0.89 to R 2 = 0.97. The results showed that the most decisive characteristics for predicting the efficiency of the pretreatment were the lipid (r = (cid:0) 0.88), ash (r = + 0.79), protein (r = (cid:0) 0.61), and hemicellulose/lignin (r = (cid:0) 0.53) content of raw biomass. Finally, the approach followed in this study facilitated an improved understanding of the mechanism of the pretreatment and presented a methodology to be followed for developing tools for the prediction of pre-treatment efficiency in the field of lignocellulosic biomass valorization.


Introduction
Lignocellulosic biomass has been in the spotlight of research during the last few decades due to its recognized potential for contributing to the production of sustainable and renewable energy, fuels and chemicals. In Europe (EU-28), agricultural and residual lignocellulosic biomass is considered to be a key resource for the significant expansion of the biogas and biomethane production sector in the near future, which is expected to double at least until 2030 as compared to year 2015 (Meyer et al., 2018). The use of residual lignocellulosic biomass for biogas production through anaerobic digestion (AD) presents certain additional advantages. Besides avoiding the environmental pollution due to its otherwise uncontrolled degradation in the fields and storage containers, residual biomass generation does not require arable land in contrast to energy crops. While a significant effort is needed in determining the most appropriate ways for biomass collection and handling, an important limitation for its integration in AD processes is its low conversion rate and extent, which results in significantly lower methane production compared to the theoretical potential.
Lignocellulosic biomass is mainly composed of cellulose, hemicellulose and lignin, which are interconnected to form a complex and recalcitrant to biodegradation matrix. Extensive research has been carried out on identifying the most limiting factors to biodegradation, ultimately pointing out the crystallinity of the cellulose fibers and more importantly the lignin barrier for conversion to methane (Carrère et al., 2016). In an effort to improve the ultimate methane yield and production rate of lignocellulosic biomass, numerous pretreatment methods have been developed in the past few decades, i.e. mechanical, chemical, thermal, biological and combinatorial methods (Olatunji et al., 2021).
The theoretical methane potential, which can be calculated for a biomass of known elemental composition (Symons and Buswell, 1933), is a poor indicator of the ultimate methane yield under real conditions. This is due to the variable degree of matrix complexity among different types of biomass, or even among biomass of the same type depending on its source and state. In order to obtain a more reliable estimation of the ultimate methane yield, both raw and pretreated biomass is traditionally subjected to biochemical methane potential (BMP) assays. These assays consist of batch AD trials of biomass seeded with anaerobic inoculum and monitoring of the methane production under laboratory conditions. However, BMP tests are time-consuming and can be costly when it comes to fast decision-making e.g., as to whether a certain biomass should be included in industrial AD processes. Thus, the prediction of the ultimate methane yield of different types of biomass and eventually the choice of pretreatment by faster methods would undoubtedly be a valuable tool for the AD industry.
Earlier studies have focused on predicting the ultimate methane yield of AD substrates based on biomass compositional characteristics by using linear and multiple regression models (Raposo et al., 2020). Since the biodegradability of lignocellulosic biomass is a multifactor problem, models constructed with single regressors (e.g. lignin, carbohydrates, chemical oxygen demand (COD), etc.) were insufficient for modeling the ultimate methane yield (usual R 2 less than 0.65), (Gunaseelan, 2007). On the other hand, the use of multiple regressors has resulted in models with an improved prediction potential, for example for food and vegetable waste (Gunaseelan, 2007), grassland biomass (Dandikas et al., 2015) and a variety of lignocellulosic biomass types (Monlau et al., 2012;Triolo et al., 2011). Nonetheless, models developed for predicting the ultimate methane yield of raw biomass may fail to predict satisfactorily the yield of biomass that has undergone a pretreatment step. This is due to the alterations caused in the matrix that were not considered during model calibration. The rise of pretreatment technology development, along with a significant effort for developing technologies that can be implemented at industrial scale, calls for the development of models able to identify which types of biomass are good receptors of a certain pretreatment in a fast and accurate way. This, besides shedding more light on the mechanism of a specific pretreatment, would ease the pretreatment technology selection and application for industrial plants.
In this study, seven different types of lignocellulosic biomass considered of industrial relevance (two types of animal manure, four types of agricultural straw and grass) have been collected and systematically subjected to compositional analysis according to the two-step sulfuric acid hydrolysis (SAH) procedure of NREL (Sluiter et al., 2011). In the sequel, all biomass types were subjected to BMP tests. This permitted the determination of the ultimate methane yield and hydrolysis rate of each biomass, without any pretreatment (raw biomass) and after pretreatment. The aim of the study was to identify different regressors for predicting the ultimate methane yield before and after pretreatment, as well as to identify which compositional characteristics make different matrices good candidates for a specific pretreatment. For this purpose, a simple ammonia pretreatment at room temperature was selected as a case study and partial least square (PLS) regression was used for model generation. This pretreatment has so far been tested on some types of biomass with a variable efficiency (Antonopoulou et al., 2015;Jurado et al., 2013bJurado et al., , 2013aMirtsou-Xanthopoulou et al., 2014), and has recently been reported to be promising for industrial application given the mild conditions and possibility for ammonia recovery (Lymperatou et al., 2021). The developed models are valuable tools for estimating the ultimate methane yield of lignocellulosic biomass and for evaluating the suitability of the ammonia pretreatment based on the initial composition of the biomass. The present study develops a methodology that can be established as an evaluation tool of the application potential of newly designed pretreatment technologies and for enhancing the comparability among studies.

Biomass and inoculum source
Seven different types of biomass were selected for evaluating the effect of ammonia pretreatment on the increase of the CH 4 yield and hydrolysis rate, namely, wheat straw, barley straw, rye straw, rape straw, grass, chicken manure and horse manure. The wheat and barley straw were collected from the experimental field of the University of Copenhagen (Taastrup, Denmark), and the rye and rape straw from Vestforsyning (Holstebro, Denmark) during the month of September. The grass was harvested from the gardens of the Technical University of Denmark (Kgs. Lyngby, Denmark) during the month of September. The manure samples were provided from Madsen bioenergi (Spøttrup, Denmark), which receives livestock manure from neighboring farms on a weekly basis, and were stored at − 20 • C until used. Prior to BMP tests, all biomass samples were cut to a 4 mm length with a cutting mill (Retsch SM 2000, Germany); The grass, horse manure and chicken manure were previously dried in a conventional oven at 45 • C to reach a total solids (TS) content higher than 90%. The inoculum used for the BMP tests was a mixture of 50% inoculum collected from a manurebased lab-scale digester and 50% inoculum collected from the Solrød biogas plant (Solrød, Denmark). TS, volatile solids (VS), soluble COD, NH 4 + -N concentration and pH of the inoculum were 2.3% wet mass, 1.4% wet mass, 2.8 g O 2 /L, 3.7 g NH 4 + -N/L and 8.2 respectively.

Ammonia pretreatment
Each biomass was subjected to ammonia pretreatment in a 2-L bluecap bottle sealed with Parafilm tape (Fig.S1) at the following conditions: 50 g TS biomass/L reagent, 4 days duration, 20 • C temperature, and 15% w/w NH 3 in water. During the pretreatment, all bottles were kept in a fume hood. The conditions applied were chosen based on previous knowledge from the statistical optimization of the pretreatment of swine manure fibers and wheat straw for maximizing their CH 4 yield (Lymperatou et al., 2020(Lymperatou et al., , 2017. The pretreatment was considered finalized after the removal of NH 3 by means of vacuum evaporation as described in Lymperatou et al. (2020). The final NH 3 concentration of all pretreated biomass was less than 1.0 g/L.

BMP tests
The effect of the ammonia pretreatment on the ultimate CH 4 yield and hydrolysis rate of the substrates was assessed through BMP tests. Three replicates of each ammonia-treated biomass, three replicates of each raw biomass and three control experiments (only inoculum) were set up in 320 mL vials at a substrate to inoculum ratio of 1:2 on a VS basis (Fig.S2). All vials were flushed with a mixture of 20% CO 2 and 80% N 2 and sealed with rubber stoppers and aluminum crimps for ensuring anaerobic conditions. The vials were incubated under mesophilic conditions (37 • C) and headspace sampling was carried out three times per week for determination of CH 4 production through Gas Chromatography (GC). All gas volumes reported correspond to STP conditions (1 atm, 0 • C). All vials were manually stirred after gas sampling and placed back in the incubator. The ultimate CH 4 yield of each biomass was calculated as the average cumulative CH 4 yield of triplicates during the last measurements when CH 4 production from each test ceased.

Analytical methods
Determination of TS, VS and ash content was carried out according to APHA standard methods (APHA, 2005). All raw biomass was milled down to 1 mm prior to compositional and elemental analysis and determination of NH 4 + -N concentration. Soluble (non-structural) sugar quantification and NH 4 + -N determination was performed in raw milled biomass (after suspension in Millipore-grade water for 3 h) and in pretreated biomass liquors after a centrifugation at 10000 rpm for 10 min and filtration through 0.45 μm. The NH 4 + -N concentration of the biomass was determined by HACH Lange kits 305 (HACH Lange, Germany). Soluble sugars (oligosaccharides and free sugars) were quantified in both raw biomass and in biomass after ammonia pretreatment, following a mild sulfuric acid hydrolysis procedure as described in Bjerre et al. (1996). All raw substrates were subjected to compositional analysis for the determination of extractives and volatiles (Sluiter et al., 2008), structural carbohydrates and lignin (Sluiter et al., 2011) and to elemental analysis. Elemental analysis was performed with an EA3000 elemental analyzer (EuroVector, Italy) on all raw biomass using He as carrier gas and sulphanilamide as standard. Sugar detection and quantification (glucose, xylose, arabinose) was carried out by means of high performance liquid chromatography (HPLC, Shimadzu, USA) equipped with a refractive index detector (Lymperatou et al., 2017). The CH 4 content in biogas samples was determined by means of a gas chromatograph (GC82-22, Mikrolab Aarhus, Denmark) equipped with a thermal conductivity detector (Lymperatou et al., 2022).

Calculations and assumptions
The theoretical CH 4 yield of biomass was calculated based on the elemental composition of each substrate and by using the Buswell formula (Symons and Buswell, 1933): The increase of ultimate CH 4 yield and hydrolysis rate of the biomass after the ammonia pretreatment were calculated as: where x i , is the ultimate CH 4 yield in mL/g TS or the hydrolysis rate in d − 1 of each i biomass. The hydrolysis rate was calculated by fitting a 1st order kinetic model on the BMP curves of each biomass and by following the assumption that the hydrolysis is the limiting step in the AD of lignocellulosic biomass: where: B and B 0 is the CH 4 yield after t (d) digestion time and at the end of digestion respectively, and k h , is the hydrolysis rate in d − 1 .
The protein content of biomass was calculated based on the elemental N content, subtracted by the NH 4 + -N content, and multiplied by the factor 6.25 (Galí et al., 2009). The content of lipids in biomass was assumed to correspond to the ethanol extractives quantified gravimetrically according to the protocol of NREL (Sluiter et al., 2008).

PLS regression analyses
PLS regression analyses were performed using the statistical software The Unscrambler® Version 11 (CAMO A/S, Norway). A data matrix was formed with the main compositional characteristics of the seven substrates used as x variables (regressors), and the y variables (one at a time) being: the ultimate CH 4 yield of raw biomass, the ultimate CH 4 yield of ammonia pretreated biomass, the % increase of the ultimate CH 4 yield after ammonia pretreatment, and the % increase of k h of the biomass after the pretreatment. All variables and responses were centered and weighted. The validation of the models was performed by full cross validation; based on this approach, a model is generated by keeping out of the calibration set one sample (biomass) at a time, which is used for model validation. This is done sequentially until all samples have been used for validation once. The final model is selected based on the highest coefficient of regression R 2 and the lowest root mean square error of the calibration (RMSEC).

Composition of raw biomass and effect of ammonia pretreatment
The substrates used in this study (agricultural straw, animal manure and grass) were chosen based on their high potential for contributing to a significant expansion of the AD industry (Meyer et al., 2018). Agricultural straw refers to the parts of a crop plant remaining after the harvest of the seed that is usually left to decompose in the fields or stored in bales for other uses. Straw is characterized by a high C and a low N content that can be variable, depending, among others, on the extent of degradation prior to collection. On the other hand, livestock manure contains the recalcitrant part of lignocellulose that has not been digested during animal digestion, together with bedding material, impurities from the collection and handling process and urine. As a result, manure is characterized by a significantly higher N content than straw. Grass is also characterized by a high N content due to its increased need for N for the formation of chlorophyll compared to crop plants. The results obtained from the biomass elemental analysis (Table 1) are in agreement to these trends. The C content was higher in agricultural straw ranging from 45 to 49% of TS, while the highest N content corresponded to the chicken manure and grass (2-4% TS). From a C/N ratio point of view, the only biomass that presented a value close to the optimal range for AD  (25-30:1) (Wang et al., 2012) was horse manure. As expected, the C/N ratio of the agricultural straw was excessively high (>100:1) and the ratios of the chicken manure and grass were very low (≤19:1). These characteristics make chicken manure and grass prone to induce NH 3 inhibition events in AD, while agricultural straw is considered a suitable substrate for balancing the C/N ratio in co-digestion practices (Neshat et al., 2017). All substrates were further analyzed for their composition in structural carbohydrates, total extractives and volatiles, lipids, proteins, lignin and ash content ( Table 2). The compositional analysis showed a large variation in total structural carbohydrates (cellulose and hemicellulose) in different types of straw, with the rye straw presenting the highest content (75%TS) and rape straw the lowest (51%). On the contrary, the N-rich biomass presented significantly lower structural carbohydrates (35-40%) but a higher lipid, protein and total extractives content ( Table 2). The horse manure presented middle values in all components except for lignin, which was the highest compared to the rest of substrates. Overall, the total structural carbohydrate content ranged from 35 to 75% TS, the lipids content from 2 to 17% TS, the proteins content from 0.2 to 22.0% TS and the lignin content from 12 to 23% TS. The variation of the biomass composition was considered to be satisfactory to support the regression analyses.
In the sequel, all biomass samples were subjected to ammonia pretreatment and liquid fractions were collected for quantification of the soluble sugars content. As it may be observed in Fig. 1, all biomass samples presented an increase of soluble sugars after the pretreatment, which is more evident for C-5 sugars (xylose and arabinose) that were derived from the hemicellulose fraction. On the other hand, the glucose content was increased in all biomass samples after the pretreatment except in grass and chicken manure, where a slight reduction was observed. Previous applications of ammonia pretreatment at room temperature have shown that most substrates exhibited a significantly reduced hemicellulose fraction (quantified as structural C-5 sugars) while the cellulose content remained invariable (Lymperatou et al., 2020(Lymperatou et al., , 2017, or deviated slightly from the initial content (Antonopoulou et al., 2015;Jurado et al., 2013aJurado et al., , 2013bMirtsou-Xanthopoulou et al., 2014). The slightly reduced glucose content in the soluble fraction of chicken manure and grass indicated that the pretreatment may have resulted to a partial glucose degradation in these biomass samples. This could be a result of a higher fraction of glucose being degraded than solubilized by ammonia, or due to already soluble glucose being degraded by ammonia. Previous studies on the effects of ammonia on lignocellulosic cell walls have identified a series of nitrogenous compounds formed during pretreatment (Balan et al., 2012;Chundawat et al., 2010). While most of these reactions take place at harsher ammonia pretreatment conditions, e.g. at high temperature, earlier studies have shown that ammonia reacts with free sugars even at low temperature when applied at prolonged durations (Kort, 1970).

Methane potentials and hydrolysis rates of biomass
The theoretical CH 4 yields of the different substrates ranged from 373 to 461 mL/gTS, with the highest corresponding to rye straw and the lowest to grass. The ultimate CH 4 yields obtained experimentally highlighted the importance of taking the matrix complexity into account, as the substrates reached a biodegradation extent (ultimate CH 4 yield expressed as % of the theoretical) between 47 and 82% (Table 3). For instance, substrates with a similar theoretical CH 4 yield (407-409 mL/g TS) such as wheat straw, rape straw and horse manure, resulted to a much wider range of experimentally obtained yields (192-337 mL/g TS). The pretreatment resulted in an increase of the ultimate CH 4 yield in most substrates but also in a decrease in three of them (Table 3). Interestingly, two of the substrates exhibiting a negative effect on the ultimate CH 4 yield after the ammonia pretreatment, grass and chicken manure, also presented a reduced soluble glucose fraction (section 3.1). Furthermore, these substrates presented a high share of extractives, lipids and proteins in contrast to other substrates. It could be hypothesized that the ammonia pretreatment affected negatively the ultimate CH 4 yield of these substrates due to the destruction of readily available organic matter already present in the extractives fraction (section 3.1), since the presence of common inhibitory by-products has been found to be very limited even in harsher conditions of this pretreatment compared to the ones applied in this study (Antonopoulou et al., 2015). This hypothesis is supported by the relatively high biodegradability of the respective raw substrates prior to pretreatment (74% and 73% for grass and chicken manure respectively). Overall, the increase of the ultimate CH 4 yield of the substrates that responded positively to the pretreatment ranged from 4 to 17%.
Ammonia pretreatment at room temperature has been applied in the Table 3 Methane yields and hydrolysis rates of raw and ammonia-treated biomass.  past in various biomass types with a variable efficiency. Jurado et al. (2013b) reported an increase of the CH 4 yield of 30-80% after ammonia application on raw swine manure fibers and 178% on digested manure fibers. Antonopoulou et al. (2015), tested the effect of ammonia pretreatment on sunflower straw, poplar and grass and reported an increase of 38%, 149% and 26% of CH 4 yield respectively. However, these results correspond to ammonia application at a much higher concentration (32% w/w) than the one applied in this study (15% w/w). Later experiments have shown that 18% and 7% NH 3 w/w corresponded to the optimal concentrations for maximizing the CH 4 yield of wheat straw and swine manure fibers respectively (Lymperatou et al., 2020(Lymperatou et al., , 2017. It is expected that the optimal pretreatment conditions vary depending on the biomass in question. However, in an industrial set up it is more likely to aim at conditions that will ensure pretreatment efficiency on most substrates. For this reason, middle range conditions of ammonia pretreatment were chosen in this study. As discussed earlier, the limitation of using lignocellulosic biomass in AD processes may result not only due to the limited biodegradation extent, but also due to the slow hydrolysis rate. The ammonia pretreatment resulted in an increase of the hydrolysis rate of all substrates tested in this study, including the ones that did not present an increased ultimate CH 4 yield ( Table 3). As a matter of fact, the highest increase of the hydrolysis rate (77%) was observed in grass, which had a − 16% ultimate CH 4 yield after the pretreatment compared to that of the raw biomass. Thus, depending on the hydraulic retention time (HRT) of the AD process applied, the ammonia pretreatment can be efficient also on substrates that do not exhibit an improved ultimate CH 4 yield, (the CH 4 production will be higher when the AD duration is shorter).

PLS regression models
After the termination of the BMP experiments and the biomass compositional analyses, data were collected for modeling the ultimate CH 4 yield of raw and pretreated biomass, as well as the efficiency of the ammonia pretreatment (expressed as % increase of the ultimate CH 4 yield and as % increase of the hydrolysis rate). Initially, all measured compositional characteristics (including cellulose, hemicellulose, xylan, arabinan, total structural carbohydrates, lignin, ash, total extractives, water extractives, lipids, proteins, hemicellulose/lignin ratio, and xylan/arabinan ratio) were used as regressors aiming at identifying the most important factors showing a strong correlation to the response. In continuation, models were constructed by using as regressors the most strongly correlated x-variables to the respective response. Previous attempts included modeling of the biogas or CH 4 yield of biomass by using only one regressor (monocausal regression models), usually the lignin content, since its role in impeding biodegradation is well known and documented (Dandikas et al., 2014;Triolo et al., 2011). While the estimation of the response based on only one biomass compositional characteristic could be faster or easier, in this study, the focus was on identifying all compositional characteristics that play an important role for maximizing the prediction accuracy of the constructed models. This approach may also permit a better understanding of the mechanism of a pretreatment when evaluating as a response the pretreatment efficiency.

Prediction of ultimate CH 4 yield of raw and pretreated biomass based on the initial composition
The regression analyses showed that only a few biomass characteristics affected significantly the ultimate CH 4 yield for both raw and pretreated biomass. Two factors were sufficient for explaining the variation of 95% and 94% of the x variables and 89% and 90% of the response for modeling the ultimate CH 4 yield of raw and pretreated biomass respectively. The strongest correlations to the ultimate CH 4 yield of raw biomass (model 1) corresponded to hemicellulose (r = +0.42), lignin (r = − 0.43) and ash (r = − 0.34), (Fig. 2a). Similar correlations were observed for model 2 constructed for the prediction of the CH 4 yield after ammonia pretreatment, with the addition of a negative correlation to the total extractives content of the raw biomass (Fig. 2b).
It is well known that the conversion of lignin to CH 4 is very limited under anaerobic conditions, presenting also a physical barrier to biodegradation as it impedes microbial access to structural carbohydrates. Thus, a negative correlation to the CH 4 yield was expected, as also reported in multiple studies (Dandikas et al., 2014;Monlau et al., 2012;Triolo et al., 2011;Tsapekos et al., 2018). Similarly, a positive correlation of CH 4 to hemicellulose was anticipated as, due to its amorphous structure, it is more easily hydrolyzed than cellulose. On the other hand, cellulose fibers are composed by glucose units in both amorphous and highly crystalline zones (Carrère et al., 2016). Cellulose has been found to be both positively and negatively correlated to the ultimate CH 4 yield of various biomass types (Raposo et al., 2020), possibly due to the variable degree of crystallinity as also reported by (Monlau et al., 2012). However, in agreement also to the results of the present study, the cellulose content is a weak regressor for predicting the ultimate CH 4 yield of lignocellulosic biomass and, as a result, it is usually excluded from regression models (Dahunsi, 2019;Dandikas et al., 2014;Tsapekos et al., 2018). Finally, the strong correlation of the ultimate CH 4 yield to the ash content is an interesting finding. Some researchers have Fig. 2. Correlation loadings of PLS models (a) 1 (eq.4) and (b) 2 (eq.5) for the prediction of the ultimate CH 4 yield of raw and pretreated biomass respectively.
A. Lymperatou et al. excluded the ash content as a regressor for modeling the CH 4 yield of biomass and chose to express the response in mL per g VS instead of g TS. However, a preliminary attempt to model the CH 4 yield per g VS in this study resulted to models of lower quality (R 2 = 0.70, R 2 = 0.85 for CH 4 yield of raw and pretreated biomass respectively), indicating the importance of including the inert material as a factor in the regression analysis and expressing the yield per g TS.
Model 2 presented some differentiation regarding the degree of correlation of the regressors to the response in comparison to model 1. Both regressors of hemicellulose and lignin became weaker for describing the CH 4 yield of the pretreated biomass (r = +0.34, r = − 0.30 for hemicellulose and lignin respectively). This indicates, that while these were still important factors for CH 4 yield prediction, their influence on the response after ammonia pretreatment was weaker in contrast to the ash content that became more important (r = − 0.45). The most evident difference between the two models though lies on the negative correlation of the CH 4 yield of pretreated biomass to the total extractives and volatiles content of the raw biomass (r = − 0.15). Extractives, depending on the type of biomass, may consist of fats, waxes, soluble proteins, soluble sugars, starch, resins, salts and other volatile or non-volatile soluble organic material along with soluble ash (Karimi and Taherzadeh, 2016). Their correlation to the ultimate CH 4 yield of raw biomass can be considered limited, due to their low share of total organics converted to CH 4 as related to the total biodegradable material of the biomass. This is also demonstrated in low correlations found between the CH 4 yield of substrates and their soluble COD, which serves as an indirect measurement of soluble organics, as reported previously (Lymperatou et al., 2017;Tsapekos et al., 2015),. The negative correlation between the extractives content of raw biomass and the ultimate CH 4 yield of ammonia-treated biomass observed in this study may be explained by the destruction of soluble organics by the chemical reagent, confirming also our previous hypothesis (section 3.2).
Overall, the fit of the data and the calibration errors were satisfactory for both model 1 (R 2 = 0.89, RMSEC = 18 mL/g TS) and model 2 (R 2 = 0.90, RMSEC = 20 mL/g TS). Fig. 3a and 3b show the fit of the models to the data, where the predicted responses are plotted versus the actual values of the substrates in this study as well as versus data retrieved from literature (Monlau et al., 2012) when this was possible (Table S1). The availability of literature data for validating model 2 was limited, as in previous ammonia applications on biomass, more often than not, the total extractives content has not been determined or reported, and the conditions of the pretreatment differed significantly. As it may be observed in Fig. 3a, the model presented a fair fit to the literature data with a few exceptions. However, it is important to point out that not all compositional characteristics retrieved from literature were within the calibration range. For instance, maize cobs were characterized by a 34.6% TS hemicellulose and Jerusalem artichoke leaves a 57.9% TS ash content, when the ranges of these regressors were 15-29 % TS and 3-18% TS respectively in the calibration data. The final models for the calculation of the ultimate CH 4 yield of raw and pretreated biomass respectively are: where, CH 4 yield i is the ultimate CH 4 yield of the raw or ammonia pretreated biomass, Lig, Ash, hemi and tot.extr. , are the lignin, ash, hemicellulose (sum of xylan and arabinan) and total extractives and volatiles contents of the raw biomass in % TS, respectively.

Prediction of efficiency of pretreatment based on raw biomass composition
Two responses were considered for modeling the efficiency of the pretreatment based on the composition of the raw biomass; the increase of the ultimate CH 4 yield (model 3), and the increase of the hydrolysis rate, k h , after the pretreatment (model 4), in order to express the effect of the pretreatment on the biodegradation extent and rate respectively. The models generated by including the biomass characteristics with the strongest correlations as regressors explained 98% and 94% of the variation of the x variables and 97% and 94% of the variation of the response when modeling the increase of the ultimate CH 4 yield and hydrolysis rate, respectively.
The increase of the ultimate CH 4 yield due to the ammonia pretreatment was found to be strongly and negatively correlated with the lipids (r = − 0.88) and the proteins content (r = − 0.61), (Fig. 4a). Following the observations discussed in section 3.3.1, this result was of no surprise as lipids and proteins are associated to the total extractives content of biomass, which were found to affect negatively the ultimate CH 4 yield of pretreated biomass. Lipids are composed by long chain fatty acids (LCFAs) and glycerol, and are generally characterized by a high CH 4 potential. Nevertheless, the accessibility of LCFAs by microbes during AD may be limited due to their low solubility and tendency to float on the surface (Carrère et al., 2016). Application of alkaline reagents results to saponification, improving thus the solubility of fatty substances. As a matter of fact, saponification has been considered as a Fig. 3. Experimental versus predicted values of ultimate CH 4 yield of lignocellulosic biomass (a) without pretreatment (model 1, eq.4), and (b) after ammonia pretreatment (by model 2, eq.5). Grey dots correspond to the data used for calibration and dark red dots to data extracted from literature (Table  S1). Vertical bars correspond to standard error of predicted values. Dashed line represents a perfect fit of the models to the experimental data (x = y).
promising pretreatment for improving the anaerobic biodegradability of high-fat residues (e.g. slaughterhouse waste) in previous studies (Harris et al., 2017;Mouneimne et al., 2003). Moreover, an improved reduction of lipids during AD of ammonia-treated manure fibers has been reported recently in continuous AD (Lymperatou et al., 2021). However, the strong negative correlation found between the biodegradation extent after the ammonia pretreatment and the lipid content indicates that the pretreatment may be harsh on biomass with a high lipids share, resulting to a loss of fatty material easily convertible to CH 4 . Both grass and chicken manure shared this characteristic (Table 2) and presented a reduction of their ultimate CH 4 yield after pretreatment (Table 3). Interestingly, the biodegradation extent was also negatively correlated to the hemicellulose content (r = − 0.47) and to the hemicellulose/lignin ratio of the raw biomass (r = − 0.53). Ammonia is known to attack the ester and ether bonds of lignin-carbohydrate complexes, which is assumed to facilitate microbial and enzymatic access to structural sugars (Kim et al., 2016). Thus, a low ratio of hemicellulose to lignin limits the efficiency and importance of the pretreatment, as the biomass is probably more easily hydrolyzed. The same assumption may explain the negative correlation to the hemicellulose fraction, i.e. a high hemicellulose content may not require an ammonia pretreatment as this may result to destruction of easily hydrolyzed sugars.
The increase of the hydrolysis rate presented correlations with some similar compositional characteristics as the increase of the CH 4 yield, however, in an opposite way. The model suggests that raw biomass with a high lipid (r = +0.0865) and protein (r = +0.1032) content result to a higher increase of its hydrolysis rate after pretreatment. More importantly though, strong positive correlations to the response were found with the ash (r = +0.79) and cellulose content (r = +0.32) of raw biomass. Regarding the influence of the pretreatment on the cellulose content of the raw biomass, it has been reported that ammonia can affect significantly the crystallinity of cellulose due to the formation of cellulose-ammonia complexes, improving this way microbial access for enzymatic attack, as explained by Bellesia et al. (2011). However, this effect has been confirmed in ammonia applications that involve high temperatures. Thus, the effect of ammonia application at low temperatures on the cellulose crystallinity of biomass needs to be further investigated. Similarly, the positive correlation of the increase of the hydrolysis rate to the ash content is an interesting finding that has not been reported previously and deserves further investigation.
The fit of the data and the calibration error of models 3 and 4 were R 2 = 0.97, RMSEC = 2% and R 2 = 0.92, RMSEC = 7% respectively. The models were capable of predicting the respective responses satisfactorily for most substrates as it may be observed also in Fig. 5. However, the chicken manure was detected as an outlier in both models for the prediction of the efficiency of the pretreatment. The AD of chicken manure is a topic that has attracted a lot of research due to its high N content and tendency to cause NH 3 inhibition (Bayrakdar et al., 2017). Our findings (negative ultimate CH 4 yield increase and limited hydrolysis rate increase) imply that this biomass would not benefit from an ammonia pretreatment process, however the AD may be improved if the biomass undergoes only the NH 3 recovery step that would be associated to ammonia pretreatment application in an industrial setup (Lymperatou et al., 2021).
The final expressions for modeling the increase of the CH 4 yield and the increase of the hydrolysis rate after the ammonia pretreatment, are respectively: CH 4 yield increase = 69.6236 − 3.2328*Lip − 1.6263*Pro − 1.0995*hemi − 16.5006*hemi/lig (6) k h increase = − 36.2244 + 1.0376*Cell + 3.2870*Ash + 0.6725*Lip + 2.0579*Pro (7) where CH 4 yield increase, is the % increase of the ultimate CH 4 yield of the biomass after the ammonia pretreatment as compared to the ultimate CH 4 yield of the raw biomass, k h increase is the % increase of the hydrolysis rate of the biomass after the ammonia pretreatment as compared to the hydrolysis rate of the raw biomass, Lip, Pro, hemi, Cell, Ash and hemi/lig is the lipid, protein, hemicellulose, cellulose, ash content and hemicellulose/lignin ratio of the raw biomass before pretreatment.

Conclusions
This study showed that the compositional characteristics of raw lignocellulosic biomass that are decisive to an improved biodegradation efficiency after ammonia pretreatment, differ from those that result to an increased hydrolysis rate. Substrates that are more recalcitrant to biodegradation with a relatively low content in lipids (r = − 0.88), proteins (r = − 0.61) and hemicellulose/lignin (r = − 0.53), are the best candidates for ammonia pretreatment when aiming at improving their ultimate CH 4 yield. On the contrary, an increased hydrolysis rate after Fig. 4. Correlation loadings of PLS regression models for the prediction of (a) the increase of the ultimate CH 4 yield after ammonia pretreatment (eq.6), and (b) the increase of the hydrolysis rate after ammonia pretreatment (eq.7).
A. Lymperatou et al. ammonia pretreatment can be expected from substrates with a high ash (r = +0.79) and cellulose (r = +0.32), content. Alkaline reagents are considered to be appropriate for lignocellulosic biomass pretreatment as they are known to be selective on lignin-carbohydrate complex cleavage and often a significant lignin removal is reported. Interestingly, in this study, no strong direct correlation was found between the efficiency of the ammonia pretreatment and the lignin content of lignocellulosic biomass. This is more likely due to the lack of heat application, resulting to milder effects of the pretreatment. Finally, the generated models (R 2 = 0.92-0.97) may be used for identifying trends of efficiency of the ammonia pretreatment on different biomass types based on their main compositional characteristics, thus eliminating the need of actually applying the pretreatment and the subsequent BMP test in order to estimate the efficiency of this pretreatment. The approach followed for the development of models for the prediction of ammonia pretreatment efficiency on improving the AD of lignocellulosic biomass may be used as a methodology for evaluating other promising pretreatments for lignocellulosic biomass.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability
Data will be made available on request.

Fig. 5.
Experimental versus predicted values of increase of (a) the ultimate CH 4 yield (model 3, eq. (6)) and (b) hydrolysis rate (model 4, eq.7) of raw biomass after ammonia pretreatment. Vertical bars correspond to standard error of predicted values. Dashed line represents a perfect fit of the models to the experimental data (x = y).