3.1. Linear models assessed
In this study, the variations of X- and Y-axis coordinates of sixteen selected peaks were investigated and plotted as a function of moisture content. This means, not the changes in the spectrum heigh at the specified and fixed positions were measured, but the cautious evaluation on the real maxima coordinates of sixteen selected peaks has been conducted. Therefore, the approach presented relevantly differs from previously performed studies (Célino et al. 2014; Hou et al. 2014; Yuan et al. 2019; Consumi et al. 2021). It accounts the movements of the absorption bands’ maxima along both the axes of frequency and absorbance, as well as considers sixteen different peaks attributed to chemical groups embodied in cellulose structure. Consequently, revealing what happens with a peak’s highest point while cellulose moisture content elevates. Tracking the changes at a fixed wavelength might be misleading as it does not reflect a full spectral information on the chemical groups embodied in cellulose and their interactions with water molecules, e.g., possible shifts of some absorption bands along the frequency axis. Moreover, a not assigned random wavelength is a parameter without any significant scientific information – it is just a mathematical coordinate.
Absorption bands taken into consideration in this study are presented in Fig. 1. Sixteen selected maxima have been marked with red dots. Each red dot represents a point regarded for further mathematical analysis and the dashed lines reveal how the coordinates (X – wavenumber, Y – absorbance) have been read. The total amount of investigated cellulose samples is n = 250.
Moreover, Table 1 gathers information about a chemical meaning behind the selected absorption bands and reveals the assignment of sixteen maxima to the chemical moieties present in the structure of investigated biopolymer. Further information on the interpretation of cellulose IR spectrum is available elsewhere (Oh et al. 2005a; Cichosz and Masek 2020). Noteworthily, the wavenumber range given in Table 1 shows the variations between the lowest and the highest wavenumber values attributed to the position of the absorption band’s maximum in the recorded IR spectra.
Table 1 Assignment of the analysed absorption bands to chemical moieties embodied in cellulose structure.
Peak no.
|
Wavenumber range [cm-1]
|
Chemical group
|
Ref.
|
1.
|
444-432
|
vibration of C-O bonds
|
(Kaur et al. 2012)
|
2.
|
560-556
|
γCH, characteristic of cellulose I
|
(Chen et al. 2020)
|
3.
|
668-660
|
δCOH out of plane
|
(Lee et al. 2015)
|
4.
|
900-895
|
γCOC at β-glycosidic linkage; amorphous region
|
(Oh et al. 2005b)
|
5.
|
1032-1027
|
γCO
|
(Lee et al. 2015)
|
6.
|
1056-1051
|
γCO
|
(Lee et al. 2015)
|
7.
|
1107-1103
|
γ ring in plane
|
(Gulmine et al. 2002)
|
8.
|
1162-1159
|
γCOC at β-glycosidic linkage
|
(Oh et al. 2005b)
|
9.
|
1203-1199
|
δCOH in plane
|
(Oh et al. 2005b)
|
10.
|
1316-1314
|
δCH2 (wagging)
|
(Poletto et al. 2014)
|
11.
|
1335-1332
|
δCOH in plane
|
(Olsson and Salmén 2004)
|
12.
|
1370-1360
|
δCH
|
(Ciolacu et al. 2011)
|
13.
|
1429-1425
|
δCH2 (symmetric); crystalline region
|
(Oh et al. 2005b)
|
14.
|
1646-1636
|
absorbed water
|
(Oh et al. 2005b)
|
15.
|
2898-2888
|
γCH
|
(Morán et al. 2008)
|
16.
|
3339-3327
|
γOH covalent bond, hydrogen bonding
|
(Oh et al. 2005a)
|
Furtherly, as every cellulose IR analysis performed was accompanied by a short and precise Karl-Fischer titration experiment, the moisture content of each biopolymer sample has been successfully established. Then, the coordinates of absorption bands’ maxima, namely, both absorbance and wavenumber, were plotted as a function of water content in cellulose specimens. Consequently, to reduce the studied dependencies to a linear form, the regression adjustment has been performed separately for each peak. The results of the carried out investigation for three different models (simple linear, semilogarithmic, power) have been gathered in Table 2. The adjusted coefficient of determination (R2) was given for each model analysed. Furthermore, in case of absorbance-based models that exhibited R2 > 90% the standard error of calibration (SEC) was presented as well. Importantly, an external validation has been performed with ten additional IR-Fischer titration data packages.
From the information presented in Table 2, it is clearly visible that ATR FT-IR technique has a great potential as a tool in quantitative determination of water content in cellulose. It might be concluded that the absorbance-based simple linear model favourably described the gathered data, because R2 assigned to the simple linear fits for all peaks investigated exhibited a value higher than 90%. Therefore, the remaining semilogarithmic and power models have not been taken into consideration. Yet, regarding the wavenumber-based models, all three fits were investigated and none of them exhibited a value of R2 higher than 90%, hence, they did not describe the data with a sufficient reliability. Among the analysed models, the highest coefficients of determination values were at the level of approximately 70-75%.
Further description of the models analysed in this study, as well as the observed changes of the peaks’ coordinates with raising moisture content of cellulose have been presented in the subsequent sections of the article.
Table 2 Quality parameters for models’ calibration and external set validation for the models selected (p-value for each regression on the level <0.5); SEC – standard error of calibration, SEP – standard error of prediction, R2 – adjusted coefficient of determination.
Peak no.
|
Absorbance-based simple linear model
|
Wavenumber-based calibration
|
Calibration
|
External validation
|
SEC [%]
|
R2 [%]
|
SEP [%]
|
R2 [%]
|
Simple linear model R2 [%]
|
Semilogarithmic model R2 [%]
|
Power model
R2 [%]
|
1.
|
0.9
|
96.9
|
11.9
|
99.2
|
10.6
|
10.2
|
10.2
|
2.
|
0.8
|
96.3
|
12.7
|
99.1
|
70.0
|
75.6
|
75.6
|
3.
|
7.5
|
96.8
|
12.7
|
99.1
|
65.7
|
64.9
|
64.9
|
4.
|
8.3
|
95.9
|
9.7
|
99.5
|
13.4
|
9.8
|
9.8
|
5.
|
8.8
|
92.0
|
14.6
|
98.8
|
41.4
|
46.4
|
46.4
|
6.
|
7.9
|
92.8
|
14.4
|
98.9
|
64.0
|
69.0
|
69.0
|
7.
|
7.0
|
94.0
|
13.0
|
99.1
|
9.6
|
16.4
|
16.4
|
8.
|
6.4
|
95.6
|
11.2
|
99.3
|
30.1
|
50.3
|
50.3
|
9.
|
9.7
|
93.1
|
11.2
|
99.3
|
66.4
|
66.7
|
66.7
|
10.
|
6.6
|
97.0
|
9.4
|
99.5
|
4.8
|
3.7
|
3.7
|
11.
|
6.8
|
96.5
|
9.2
|
99.5
|
45.4
|
51.2
|
51.2
|
12.
|
7.0
|
96.4
|
9.5
|
99.5
|
41.8
|
48.6
|
48.5
|
13.
|
7.5
|
96.3
|
10.0
|
99.5
|
46.4
|
50.5
|
50.5
|
14.
|
22.5
|
92.6
|
18.3
|
98.3
|
2.3
|
1.8
|
1.8
|
15.
|
8.9
|
94.9
|
12.2
|
99.2
|
58.3
|
49.0
|
49.0
|
16.
|
5.7
|
98.7
|
6.0
|
99.8
|
6.1
|
8.0
|
8.0
|
3.2. Absorbance-moisture content dependency
Fig. 2 reveals the experimental data regarding the changes in the absorbance values (height of the peak) assigned to sixteen maxima visible in IR spectrum of cellulose fibres. Additionally, the black lines represent the regression functions that coefficients of determination (R2) and standard errors of calibration (SEC) have been shown in Table 2.
Giving a closer look at Fig. 2, it might be easily perceived that absorbance of each peak analysed is changing significantly while the moisture content in cellulose specimens elevates. The similar phenomenon has been observed by some other scientists (Célino et al. 2014; Hou et al. 2014; Yuan et al. 2019; Consumi et al. 2021). However, in this study the raise in the height of each peak recorded, for the first time, have been successfully described with linear functions exhibiting R2 > 90%. High value of adjusted determination coefficient means the models applied reliably reflected the experimental data.
Interestingly, signals originated from both oxygen-embodying (peaks no.: 1, 3, 4, 5, 6, 8, 9, 11, 16) and oxygen-lacking (peaks no.: 2, 7, 10, 12, 13, 15) chemical groups responded to the variations in moisture content in cellulose fibres. It is not surprising that changes in the maxima assigned to oxygen-rich chemical moieties, being able to form hydrogen bonds with water molecules, altered during the water absorption process (Mihranyan et al. 2004a). Due to the differences in the electronegativity between the atoms in oxygen-embodying groups, these moieties are able to create electrostatic attraction between cellulose and water molecules. Consequently, water bonds to the cellulose surface via physical forces (Hofstetter et al. 2006; Salmén and Bergström 2009). Because of the changes in cellulose-water interactions, the investigated chemical groups differently reacted to the infrared radiation and absorbance/wavenumber shifts of the peaks could have become visible in IR spectrum (peaks no.: 1, 3, 4, 5, 6, 8, 9, 11, 16).
However, it is not certain what kind of mechanism stands behind the recorded variations in absorbance/wavenumber of the peaks assigned to non-polar CH2 and C-H chemical groups that are not able to directly interact with water molecules (peaks no.: 2, 7, 10, 12, 13, 15). Nevertheless, Celino et al. (Célino et al. 2014) proposed a theory that could bring some elements of understanding. The changes observed might be favourably described with the surrounding signals originated from oxygen-embodying chemical moieties. The shoulders of the maxima assigned to carbon- and oxygen-rich chemical groups overlap, hence, affecting the shape, as well as the height of the peaks visible in IR spectrum.
Another explanation has been proposed by Yuan et al. (Yuan et al. 2019) who concluded that IR spectrum might also react to the cellulose chain stiffening caused by moisture absorption. The scientists referred that the changes in the macromolecule’s stiffness might contribute to the shifts along both absorbance and frequency axes.
Furtherly, the attention should be drawn to some interesting results that have been shown in Fig. 2n. Peak no. 14 is generally assigned to the moisture absorbed in cellulose (water content) (Oh et al. 2005b). Importantly, this maximum is not present in IR spectrum of the biopolymer in a dried state. Therefore, its appearance could be only related to the water absorbed by cellulose fibres. Consequently, it should precisely reflect the amount of moisture bonded to cellulose. However, an opposite effect could have been observed. Peak no. 14 exhibits one of the lowest values of adjusted determination coefficient R2 = 92.6% and the highest standard error of calibration SEC = 22.5% (Table 2). This means, the mathematical function applied does not precisely reflect the experimental data.
This observation could be explained with the nature of a certain peak. Often, maxima visible in IR spectrum are the common effect of many smaller signals that overlap. Therefore, their position and height are determined by many different interactions (Bledzki et al. 2002; Łojewska et al. 2005; Kondo et al. 2016), e.g. hydrogen bonds (Peršin et al. 2011), van der Waals forces (Missoum et al. 2013). However, peak no. 14 is assigned only to the moisture content in cellulose fibres. Hence, it reflects merely the vibrations of water molecules. Probably, this absorption band might be hardly deconvoluted into separated signals and could be significantly affected by the shoulders of neighbouring maxima. Therefore, the peak’s position is not stable and highly influenced by its surrounding.
Taking into account the above reasoning, if assume that peaks consisting of multiple signals and separated from the remaining absorption bands (restricted overlapping with neighbouring maxima) describe water uptake more reliably, a mathematical model assigned to an alone absorption band that might be deconvolved into different signals (especially corresponding to water-cellulose hydrogen bonds) should more accurately describe the changes in cellulose moisture content. Interestingly, this theory has found a confirmation in practice. The highest value of the adjusted coefficient of determination (R2 = 98.7%) has been recorded for partially separated peak no. 16 that could be successfully deconvoluted into three types of hydrogen bonding: intramolecular (3OH⋯O5 and 2OH⋯O6), intermolecular (6OH⋯O3’) (Łojewska et al. 2005; Oh et al. 2005a). Most likely, the position of this peak is stabilized with three species of interactions that might be affected by the presence of water molecules, as moisture bonds to the same active centres that create inter- and intramolecular hydrogen bonds in cellulose structure. Then, the outcome is a describable and observable peak’s shift along the Y axis.
Next, all investigated models have been validated with external experimental data set. The results of the validation performed are presented in Table 2 and Fig. 3. The mathematical functions developed in this study revealed the satisfactory similarity between the newly recorded data and the values predicted with the linear functions. The coefficients of determination for validated models exhibited values from 98.3-99.8% with the best fit for peak no. 16, which is also easily visible with a naked eye while comparing the graphs presented in Fig. 3. This means, the information collected with ATR FT-IR technique could be successfully correlated with volumetric measurements of moisture content in cellulose (e.g., Karl-Fischer titration). Consequently, ATR FT-IR method might be favourably regarded as a quantitative technique for the determination of moisture content in cellulose fibres.
Yet, the data presented in this study could be investigated differently. Experimental points of each peak characterised (Fig. 2) may be favourably divided into two parts depending on a specific bend in the shape created by the plotted data: the first stage from approximately 0-2wt.% of moisture content (before the bend) and the second stage from 2-10wt.% (after the specific bend). Therefore, this alternative approach has been carefully analysed regarding the adjustment of linear models analysed in this study, namely: simple linear fit, semilogarithmic, power. Results of this additional analysis are presented in Table 3.
Table 3 Quality parameters of fit adjustment for the alternative two-stage approach of data analysis (in case of most correlations evaluated p-value on the level <0.5); SEC – standard error of calibration, SEP – standard error of prediction, R2 – adjusted coefficient of determination.
Peak no.
|
1st stage (moisture content: 0-2wt.%)
|
2nd stage
(moisture content: 2-10wt.%)
|
Simple linear model R2 [%]
|
Semilogarithmic model R2 [%]
|
Power model
R2 [%]
|
Simple linear model R2 [%]
|
1.
|
69.9
|
85.8
|
83.1
|
93.2
|
2.
|
69.1
|
86.9
|
81.5
|
92.1
|
3.
|
67.9
|
85.7
|
82.4
|
93.0
|
4.
|
58.0
|
71.0
|
69.8
|
91.6
|
5.
|
57.5
|
72.9
|
65.6
|
86.1
|
6.
|
56.2
|
72.0
|
64.7
|
88.1
|
7.
|
57.5
|
73.8
|
68.3
|
90.4
|
8.
|
57.8
|
74.3
|
70.2
|
92.4
|
9.
|
34.4
|
42.9
|
42.6
|
88.4
|
10.
|
63.9
|
80.5
|
77.5
|
94.1
|
11.
|
60.6
|
76.1
|
73.7
|
93.5
|
12.
|
60.6
|
76.4
|
73.3
|
93.1
|
13.
|
55.1
|
69.0
|
67.0
|
92.8
|
14.
|
1.1*
|
1.6*
|
1.2*
|
89.4
|
15.
|
65.7
|
83.2
|
79.2
|
89.1
|
16.
|
74.9
|
93.2**
|
87.4
|
97.8**
|
* p-value > 0.5
** Calibration: SEC = 0.001%, External validation: SEP = 6.2%, R2 = 99.8%
Nonetheless, the division did not result in a more precise description of experimental points. Taking into consideration values of the coefficients of determination (Table 3), almost none of the mathematical models was able to reliably describe the changes in experimental points at the 1st stage of moisture absorption (moisture content: 0-2wt.%). However, simple linear fit, again, successfully reflected the increase in absorbance with raising water content during the 2nd stage of moisture absorption process (moisture content: 2-10wt.%).
Similarly, as in the previous investigation, the most adequate adjustment has been observed for the peak no. 16 which underlines the reliability of this absorption band regarding the possibility of quantitative analysis of water content in cellulose-based systems. The 1st stage of moisture absorption could have been efficiently described with semilogarithmic model (R2 = 93.2%, SEC = 0.001%) and the 2nd stage – with simple linear fit (R2 = 97.8%, SEP = 6.2%).
Taking into account the considerations set out above, it would be also worth considering whether it is reasonable to use only one absorption band to determine the water content of natural cellulose fibres, or whether the results calculated from the different absorption bands should be averaged. Therefore, additional calculations were made. The determined moisture contents obtained from the 16 absorption bands were averaged for each experimental point. Then, external validation was performed. The adjusted coefficient of determination was at the level of R2 = 99.5%, and the standard SEP prediction error – SEP = 9.9%. These values indicate a slightly lower accuracy of such a method along with the determination of the moisture content in cellulose for, e.g., peak no. 16. Thus, proving once again the accuracy of the determinations made using this absorption band.
3.3. Wavenumber-moisture content relationship
Additionally, this is the first study that have deeply investigated the wavenumber-moisture content relationship and the possibility of the mathematical models adjustment (Célino et al. 2014; Hou et al. 2014; Yuan et al. 2019; Consumi et al. 2021). The gathered experimental points plotted as a function of moisture content in cellulose have been shown in Fig. 4. Similarly, as in the previous subsection, the dots reflect the recorded data plotted as a function of moisture content. However, this time, adjusting of any regression was not possible. According to the data presented in Table 2, none of the investigated models (linear, semilogarithmic, power) did not exhibit a sufficient value of adjusted determination coefficient (R2 < 90%), hence, indicating inaccurate description of experimental data.
Although the linear regression models were unable to sufficiently reflect the collected experimental points, the data presented in Fig. 4 contains some important scientific information and deserves a brief discussion.
It might be easily noticed that the behaviour of the maxima differs. Some peaks shift along X axis (peaks no.: 2, 3, 4, 6, 9, 11, 12, 13, 15), while the others remain relatively unchanged (peaks no.: 7, 10, 16) or exhibit specified positions irrespective of the moisture content (peaks no.: 1, 8, 14). Most likely, the shifts described above might be related to the water-cellulose intermolecular interactions and different responses of the specified chemical groups to these attraction forces (Gao et al. 2005; Sun 2008).
The first group distinguished consists of the absorption bands that change the position with increasing moisture content are mostly the signals assigned to polar moieties (able to form hydrogen bonds with water molecules (Luo and Zhu 2010)) and neighbouring to them maxima attributed to non-polar groups (probably affected by the overlapping of the signals originated from polar moieties (Leszczyńska et al. 2019)). Interestingly, some absorption bands shifted to the lower wavenumbers (peaks no.: 2, 3, 4, 6) while the remaining peaks moved toward higher values of frequency (peaks no.: 9, 11, 12, 13, 15). It is supposed that such a phenomenon is related to the type of interactions created in the water-cellulose system and the reaction of the above-mentioned system to infrared radiation. However, the reason for such a behaviour is not entirely clear and needs to be furtherly analysed in the future.
Moving forward, the maxima that did not significantly change the position along the axis of frequencies are attributed to ring in plane vibrations (peak no. 7) and CH2 non-polar moieties (peak no. 10) which could be barely affected by water molecules (A.N. and K.J. 2017). However, the last maximum exhibiting relatively stable position is the peak no. 16 assigned to hydroxyl groups that actually might interact with water molecules. On the other hand, based on previously done observations and literature research, it was evidenced that peak no. 16 exhibited relative stability in the position. Firstly, it is partially separated from remaining absorption bands. Therefore, the effect of the signals overlapping is restricted. Secondly, this is the broad maximum consisting of smaller maxima attributed to the inter- and intramolecular hydrogen bonds (Cao and Tan 2004), hence, providing peak’s shape stability via dependency on different kinds of interactions.
The last group consists of the absorption bands revealing shifts along the axis of frequencies that are irrespective of the amount of absorbed water. They exhibit a few specified and fixed wavenumber values (especially peaks no.: 8, 14) which does not change describable with increasing moisture content. This means regardless of the amount of absorbed water, these peaks reveal several specific values assigned to the X axis. The behaviour described could be inherent to the material, which is known to be subjected to chemical intrinsic variability (Thygesen et al. 2005; Bourmaud et al. 2018), as well as be assigned to non-specific experimental variations (Garside and Wyeth 2006; Agarwal et al. 2011). However, this phenomenon undoubtedly requires further investigation as its nature is not fully understood.
Summarising, it was not possible to observe the wavenumber-moisture content relationships that could have been described with the investigated regression models. However, data indicating the correlation between the position of the maximum on the axis of frequencies and the moisture content have been gathered. Basing on the collected information, most likely, wavenumber shifts cannot be successfully applied in quantitative assessment of moisture content in cellulose-based materials. Yet, the position of some specific peaks (peaks no.: 2, 3, 4, 6, 9, 11, 12, 13, 15) on the X axis might favourably provide some valuable information on the cellulose-water molecules interactions by applying, e.g., 2Dcos analysis and correlation spectra (Hou et al. 2014), simple calculation of wavenumber shifts (Awa et al. 2014), deconvolutions revealing shifts of hydrogen bonds (Łojewska et al. 2005).