After pre-assembling the books, 297 items were analysed by two assessors in about 49 h, at ~10 min per book, bibliographic information which include title, author and date of publication being also recorded (see Additional file 2). The survey primarily reports data on historical rag papers, as the majority of the measured books pre-date the 19th century. Two large-scale studies have previously investigated the chemical and physical properties of historic rag papers [22,24], and the present study offers additional data which allows an in-depth analysis of rag paper degradation. A summary of the methodology and results of the visual assessment can be found in Additional file 1.
As the collections were separated into climate-controlled and non-controlled collections by the library only six years ago, they have been grouped together for parts of this study, as the last six years of degradation likely represent an insignificant period compared to the age of the objects (1-5% of the object’s lifetime). The collections will be separately dealt with in Part II [31] as the effects of the different environments on the future condition of books will be modelled.
3.1 Quantitative Survey Data Evaluation
3.1.1 Paper typology
The SurveNIR system discriminates between four European paper types, i.e., rag paper, groundwood pulp paper, bleached wood pulp and coated paper, using the chemometrical models obtained from a reference sample set of 1400 European paper [28,29]. Three paper types were present in this survey: rag paper (82%), groundwood pulp paper (13%), and bleached wood pulp (5%). The change in fibres follows the general evolution of European papermaking [47]; the oldest book made of bleached pulp paper found in this study was dated 1896, while the oldest book made of groundwood paper may be dated to 1834.
3.1.2 Acidity, protein and rosin content
It is well known that acidity plays a very important role in paper degradation, the more acidic the paper is, the faster the degradation proceeds [5,48]. Figure 1 shows the pH for the analysed books as a function of publication date and paper type. The estimated pH values cover a wide range, in which 75% of all the analysed books are close to neutral (6-8), most of which are rag paper. Groundwood pulp papers are characterised by low pH, while the books made of bleached pulp paper have a broad spread of pH values.
Fig. 1. pH of rag (squares), groundwood (circles), and bleached pulp (triangles) papers as a function of the approximated publication date. The error bars indicate the instrumental errors for each paper type.
Compared to Barrow’s large-scale study of rag paper [22], which measured the acidity of 1,470 book papers made between 1507 and 1949, the results on average are similar. Historic rag paper has pH averaging 6.1 in Barrow’s study and 6.6 in this study. Both studies observe a sharp increase in acidity in the mid-19th century onwards. Gelatine (derived from collagen, the connective tissue in skin, cartilage, sinews and ossein of animals) was the most common sizing agent used in Italian papermaking from 1337 until acidic alum-rosin sizing was introduced in 1807. During the 17th century aluminium potassium sulfate was added to gelatine size to stabilise the viscosity of the size at various concentrations and temperatures, to inhibit biological growth, and increase the resistance of the size to ink penetration [49,50].
It has been shown that gelatine is beneficial to paper, decreasing its degradation rate, as the protein can act as a physical barrier and chemical buffer [51,52]. The presence of gelatine sizing was measured by protein content (%) in this study and has been described as a function of publication date and paper type in Figure 2. The measurements show a continual decrease of protein content towards the early 19th century when alum-rosin size became widely used. Similar trends for protein use in Italian papermaking were recently observed by Barrett et al. [24], who found higher concentrations of protein in pre-1500 paper, as papermakers were attempting to imitate parchment.
Fig. 2. Protein content (%) of rag (squares), groundwood (circles), and bleached pulp (triangles) papers as a function of the approximated publication date. The error bar indicates the instrumental errors for all paper types.
The effects of the addition of alum to gelatine, discussed extensively in Barrow’s research [22] as a factor that increased acidity in papers produced from the mid-17th century could not be clearly observed in this study as the presence of aluminium was not tested during this survey. However, the use of alum was not recorded in the technical data included in early Italian papermaking descriptions [23] and high contents of protein (>4%) were measured until the late 18th century in this study. It is possible that most of the measured rag papers did not contain high concentrations of alum as high acidity was not observed. High contents of protein could have also acted as a buffer, as observed by Barrett et al. [24].
From the early-19th century on, rosin was added to the pulp and precipitated with aluminium sulfate to size paper internally. Figure 3 describes rosin content (mg/g) as a function of publication date and paper type from 1800 onwards. The results indicate a steady increase of rosin content in the books printed after 1850, as expected. It has previously been shown that high quantities of alum are detrimental to the long-term stability of paper [53]. The average pH of rosin containing paper (>2 mg/g) measured in this survey was 4.4, resulting in chemically unstable post-1850 books.
Fig. 3. Rosin content (mg/g) of rag (squares), groundwood (circles), and bleached pulp (triangles) papers as a function of the approximated date of publication post-1800. The error bar indicates the instrumental errors for all paper types.
3.1.3 Lignin content
Lignin is a polymeric structural constituent of wood and other plant tissues and is undesirable as it may cause discolouration [54,55]. The lignin content in paper varies considerably depending on the original fibres and its pulping processes, i.e., the amount of lignin in groundwood is much higher than rag paper or bleached pulp paper. The fibre sources of historic rag paper within Italy, including linen and later cotton rags, naturally contained low levels or no lignin as observed in Figure 4, which shows lignin contents (mg/g) measured in the analysed books as a function of publication date and paper type. As expected, there is a notable difference between rag, groundwood and bleached pulp paper. The lignin content of the books made of rag paper is typically below 25 mg/g, while the books made of groundwood paper have lignin contents between 73 and 303 mg/g. The books made of bleached pulp paper have less lignin than those made of groundwood paper, the lignin content of the bleached pulp paper ranging from 31 to 98 mg/g.
Fig. 4. Lignin content (mg/g) of rag (squares), groundwood (circles), and bleached pulp (triangles) papers as a function of the approximated publication date. The error bar indicates the instrumental errors for all paper types.
3.1.4 Mechanical Properties
The mechanical strength of paper reflects the chemistry, morphology and the fibre network structure of paper. Mechanical strength depends on the paper type and the environment, primarily relative humidity. Tensile strength (TS) and tensile strength after folding (TSF) were measured, as they reflect the strength of a sheet of paper not subjected to mechanical stresses (TS), while TSF represents the strength after folding using the Bansa-Hofer method [42]. Both are reported in terms of nominal force (N). Figure 5 displays the TS and the TFS as a function of publication date and paper type.
Fig. 5. TS (left) and TSF (right) of rag (squares), groundwood (circles), and bleached pulp (triangles) papers as a function of the approximated publication date. The error bars indicate the instrumental errors for each paper type.
The structure of the paper and properties of individual fibres are reflected in the TS values. The mean TS values of groundwood and bleached pulp paper books are similar to each other, 35 and 38 N, respectively, and the mean TS value for rag paper is 51 N, which is statistically significantly different even given the measurement uncertainties (Table 2). It has been reported [23] that calcium carbonate was produced from the lime used during sheet formation, creating alkaline deposits in traditional Italian papers without affecting the fibre strength, and in addition, gelatine sizing may also improve the strength of the paper sheet [51,52]. As expected, changes in papermaking processes result in a wider range of TS values for the 19th-century papers, where 49% of papers measure <45 N, which is similar to Barrow’s study [22]. A lack of folding endurance can be the result of short fibre length or lack of inter-fibre bonding [56].
3.2 Rag Paper Data Analysis
Experimental studies of paper degradation have mostly focused on the chemical and physical analysis to understand the complex relationship between a limited set of experimentally controlled parameters. Collection surveys can offer significant and complementary datasets; however, statistical analysis is required to draw conclusions on the basis of often higher data scatter and uncertainties. The specific advantage of the data collected during the survey of the Classense Library collection is that it offers an unprecedented historic rag paper dataset covering a 600-year period from 1300-1900, and further data analysis focuses specifically on this dataset.
3.2.1 Degree of Polymerisation
Since the most common analytical techniques to measure the DP, namely viscometry and size exclusion chromatography, require destructive sampling [57], condition surveys rarely report DP values. Figure 6 displays the inverse of the average degree of polymerisation (1/DP) [58–60] as a function of age for the measured rag papers only.
Fig. 6. 1/DP of rag papers as a function of age, with the regression line fitted using York regression [61], taking into account the standard errors for age and 1/DP.
The regression line in Figure 6 shows that degradation of rag paper broadly follows the Ekenstam equation [58], where the rate constant of degradation in year-1 can be calculated from the regression line slope. There is significant data scatter, however, both the slope and the intercept are statistically significant. The error bars represent the uncertainties of DP estimation (Table 2) and of paper age, and we used York regression [61] to account for the reported uncertainties for both x and y values (N = 243).
Based on the value of the intercept, average DP0 (i.e., immediately after production, at t = 0) in this study was 2360±130. Although the error interval for this value is asymmetrical due to the inverse function, the difference is small (<5%) compared to both the instrumental error (Table 2) and to the usually observed inhomogeneity of rag paper and the associated uncertainty of DP, which is typically ~10% [62]. Therefore, the negative and the positive error were averaged to give the value of 130. Although DP 2360 appears small compared to the DP of native cellulose, it is comparable to the DP of other processed cellulosic fibres, and it is useful to remember that rag fibres were substantially pre-processed and pre-degraded before being used for paper production [1].
The rate constant for chain scission is (4.2±0.6)·10-7 year-1, as deducted from the slope of the regression line in Figure 6. It represents the first experimentally observed rate constant for historical rag paper, and it is valid for all rag paper, as the methods of production were similar outside Italy. In spite of the significant data scatter, the value found for the rate constant is perfectly consistent with that reported [63] for historical Korean Hanji paper without beeswax coatings (2.17·10-7 year-1), considering that the rate of cellulose chain scission of Hanji paper was found to be about two times slower than that of rag papers [63]. The Collections Demography dose-response function [32] may help us validate this result, and in Figure 7 we estimate the range of T and RH values where such a rate would be expected for paper with average pH of 6.6, as was calculated for the 243 samples in this study. The standard error (SE) was calculated on the basis of the SE of the regression line slope in Figure 6, and reflects data scatter and uncertainties of estimations of DP and age, but it does not reflect the uncertainties of the dose response function itself, or the uncertainties of pH estimation, so it is likely that the SE is substantially bigger, although numerical error estimation is outside the scope of this article. The estimated SE of 0.6·10-7 year-1, i.e., 15%, is comparable to the estimated errors for experimentally determined degradation rates [57], which is remarkable, given that the declared instrumental uncertainties and general data scatter are high.
Fig. 7. Range of RH and T values that could lead to the observed rate of rag paper degradation (average pH 6.6) of 4.2·10-7 year-1, based on the Collections Demography dose response function for historic paper [32]. The green range represents ±1SE interval for the rate, the yellow one ±2SE, and the orange one ±3SE.
The coloured areas in Figure 7 thus represent the potential past average storage conditions that the surveyed samples may have endured since production in order for paper to decay at the observed rate. Since the majority of the surveyed rag papers have been stored in the Classense Library building since their acquisition, it is reasonable to assume that the current average indoor climate reflects the past indoor climate reasonably well. The 2014 averages in the non-climate controlled areas were measured to be 7-17 oC and 50-70% RH in December and 24-28 oC and 53-63% RH in July/August [64,65]. The external walls of the building are composed of fired clay bricks. The data measured during two monitoring campaigns shows that the Classense Library has a high thermal inertia [64]. Therefore, while August may be the hottest month indoors, December may not be the coldest one, as the annual minimum temperature in Ravenna is in January, and allowing for a time lag due to the thermal mass of the building, the coldest month indoors could be February. On the other hand, the average annual temperature over the last 10 years (from January 2009 to January 2019) in Ravenna has been 15 oC [66], meaning that a passive storage area in a building with a large thermal mass could have a lower annual average temperature than that. In addition, the climate has been warming up since the Little Ice Age in the 16th century [67], meaning that the past temperatures experienced by the collection may have been significantly lower than those measured in the same building today, and thus the above calculated rate of rag paper degradation may be a valid estimation.
3.2.2 PCA and MLR
Using PCA to visualise the relationships between all the data of rag paper books, the first two principal components account for 83% of the cumulative variance, which provides a useful approximation of the relationship between variables. TSF, lignin and rosin content were not included in PCA. TSF is very strongly correlated to TS (see Additional file 1: Fig. S6), the lignin content in rag paper is very small, and rosin is present only in books dated post-1800. Figure 8 shows the resulting biplot where the observations and positions of the considered variables are displayed.
Fig. 8. PCA of the observation data for rag paper.
TS, pH and protein have similar heavy loadings for PC1, as well as TS, pH, and publication date for PC2. DP adds little to the first component having the maximum loading for PC2. Besides, protein content is negatively correlated to publication date, while vectors corresponding to TS and pH are positively correlated, as reflected in pairwise correlations (see Additional file 1: Fig. S6). It is known that tensile strength is influenced by intra- and inter-fibre bonding [68], which could be represented by DP and protein content, respectively. However, their contributions have never been quantitatively examined, and MLR could help us evaluate this relationship. As all of the observed variables depend on date, this was also included in the MLR analysis.
Figure 9 and Table 3 show the model outputs where age, DP and protein are considered as independent variables to determine tensile strength. The model explains ~60% of data variability, and indicates that all three variables are statistically significant (p < 0.05), p-value of DP being the smallest, as reported in Table 3. The relationship between tensile strength and individual independent variables within the multiple-regression model can be visualised through partial leverage plots (Fig. 9), which are constructed from the residuals of tensile strength (Y) and the independent variable (X).
Table 3. Outputs of the TS multiple regression model with age, DP and protein as independent variables.
Variable
|
Coef. Value
|
Standard Error
|
p-Value
|
Intercept
|
15.77055
|
2.08439
|
8.2854E-13
|
DP
|
0.01592
|
0.00098
|
2.54946E-40
|
Protein (%)
|
2.19776
|
0.22617
|
5.03676E-19
|
Age
|
0.00685
|
0.00285
|
0.01714
|
Fig. 9. Partial leverage plots of the TS multiple regression model with age, DP and protein classified as independent variables.
Age shows the largest spread of data within its partial leverage plot and it can be concluded that within the measured sample set, age of the paper has the least effect on tensile strength. Both DP and protein content can be viewed as significant predictors of tensile strength. The partial leverage plots (Fig. 9) show that while both DP and protein content have a strong linear relationship to tensile strength, the X and Y residuals of the DP partial leverage plot are more closely correlated, indicating that DP has a greater effect on tensile strength than protein content. In agreement, Zou et al. [69] found an almost linear correlation between DP and TS in artificially aged Whatman paper. The present results suggest that intra-fibre bonding is the principal driver for increased tensile strength of rag papers, and inter-fibre bonding is of secondary importance, which is a significant conclusion related to the use of DP as a general proxy for rag paper mechanical strength.