Spectral sensitivity of the discoloration of Historical rag paper

This paper discusses the spectral sensitivity of the discoloration of historical rag paper simultaneously affected by Relative Humidity (RH) and Oxygen concentration [O2] in the ambient environment. Sacrificial samples were degraded using narrowband radiation sources centred at 450 nm, 525 nm, and 625 nm in combinations of RH and [O2] at two levels: 0% [O2] and 70% RH, 21% [O2] and 70% RH, 0% [O2] and 20% RH, and 21% [O2] and 20% RH. Diffuse reflectance was measured before and during the degradation experimental runs. Consistent qualitative results were obtained for the change in reflectance and the change in tristimulus total color change in CIELAB color space. In both cases, the increase of discoloration was modelled logarithmically over time. Among the three factors investigated in this research, wavelength of the radiation (Λ) was found to have the strongest effect. The radiation at 450 nm induced the most and fastest discoloration whereas the radiation at 625 nm induced the least and slowest discoloration. This spectral dependence was likely to be related to the photo energies at different wavelengths, but other factors were found to have played a role. Further analyses revealed that the main effects and the effects of the interactions between [O2] and Λ and between RH and Λ on the discoloration of historical rag paper were statistically significant. It suggests that managing the spectral power distribution of the radiation source can be crucial in the collection management. List of abbreviations ANOVA Analysis of variance ΔE00 Total color difference defined by CIEDE2000 formula ΔE00N Normalized total color difference defined by CIEDE2000 formula Ev Illuminance kΔE00N Normalized rate constant of change in total color difference kΔR Rate constant of change in diffuse reflectance kΔRN Normalized rate constant of change in diffuse reflectance KΔRN Normalized rate of change in diffuse reflectance LED Light-emitting diode N Number of photons in Mol [O2] Oxygen concentration ΔR Change in diffuse reflectance ΔRN Normalized change in diffuse reflectance R Diffuse reflectance R2 Coefficient of determination RH Relative humidity SPD Spectral power distribution [Λ] Narrowband radiation centered at Λ Λ Wavelength of radiation in nm λ Wavelength of reflectance in nm


List of abbreviations
Narrowband radiation centered at Λ Λ Wavelength of radiation in nm λ Wavelength of reflectance in nm

Introduction
Managing the discoloration of collection materials is among the top priorities for museums and galleries, especially when it comes to decision making for display strategies. But this strong need has not led to sufficient scientific research on the issue. Most of the limited studies often focus on the colorants of artworks and manuscripts [21,28] with little consideration given to the supporting materials, despite the fact that the properties of substrates can affect the photochemical behavior of colorants [7][8][9][10]16]. Moreover, the discoloration of substrates has been observed to penetrate through the surface layers and significantly affect the overall measurements of discoloration of collection materials [27].
As one of the most commonly used supporting materials for art works and manuscripts, the photochemistry of paper received interest mainly in the 20 th century. Experiments were designed using laboratory grade cellulose and monochrome and heterogeneous ultra violet (UV) radiation. The wavelength of radiation was found to play a decisive role in the degradation reaction kinetics [11,17,19,25]. Far and middle UV (λ < 300 nm) was found to initiate direct photolysis, whereas near UV (λ = 388 nm) mainly caused photo-oxidation. This observation can be explained by the fact that the energy of the photons in far and middle UV (95 kcal•mol − 1 at 300 nm) is higher than the energy required for the scission of C-C and C-O bonds (80-90 cal•mol − 1 ) [22].
In addition to the wavelength of radiation, the effect of the moisture content of samples and the effect of the oxygen concentration of ambient environment were discussed in relation to the photochemical reaction kinetics [12][13][14]17]. Using UV radiation at 254 nm, where direct photolysis was the main reaction mechanism, oxygen played a small role whereas moisture strongly inhibited the degradation of cellulose, as indicated by reflectance, degree of polymerisation, and alpha cellulose content [18]. In contrast, when UV radiation at 388 nm and heterogeneous UV radiation were used, the rate of degradation was found to increase with increasing oxygen content, as indicated by decreasing degree of polymerisation and increasing gaseous degradation products [25].
These early studies are solid and inspirational, but are not practically helpful for collection management. In the museums and archives in the Western world, concern over the discoloration of paper-based collections mostly involves the presence of iron gall ink, which is considered susceptible to fading [1]. Due to the popular use of iron gall ink before the substantial change of papermaking in the late 19 th century [15], most collection items in concern are made of historical rag paper. Historical rag paper mainly contains hemp, linen and cotton fibres, gelatine sizing, and an accumulation of natural degradation products [26], which can behave differently from laboratory grade model paper. In addition, UV has been excluded in museums and archives and the Light-Emitting Diode (LED) has been commonly used for lighting [6]. The results obtained using UV radiation may not be directly applicable to today's lighting environment. Therefore, research on the discoloration of paper needs to be updated with historical rag paper and visible range of the electromagnetic spectrum in order to maximise its support for collection management.
Only a small number of research projects on the discoloration of historical paper under visible range of the electromagnetic spectrum has been carried out recently [3,20]. Liu [20] investigated the effects and interactions between Relative Humidity (RH), Oxygen concentration [O 2 ] and Illuminance (E v ) on Western rag paper based on factorial experimentation. Dang et al. [3] focused on the effect of different wavelengths on the discoloration of Chinese paper in a static ambient environment. It should be noted that the discoloration in the research by Dang et al. [3] was only assessed in tristimulus color space and customized colorimetric terms. This seems to be common practice in assessing discoloration of heritage collections [4,5]. Given that different formula for tristimulus color change were used in different research, transformations need to be carried out before interpreting and comparing the results.
To bridge the gaps in research on the discoloration of paper-based collections, in this research, an experiment was designed to investigate the synergistic effect of RH, oxygen and the wavelength of radiation, with an emphasis on the spectral sensitivity of the discoloration behavior. Sacrificial historical rag paper samples were used to represent the uncontrollable natural variations in the materials. Given that the tristimulus parameters are formulated based on the physiological perception of color by human vision [2,24], both spectroscopic and tristimulus responses of the samples were analyzed to examine whether consistent results could be achieved for the colorimetric study of historical rag paper. The results not only provided insights into the spectrally dependent discoloration of historical rag paper, but also have implications for museum lighting and collection management.

Methodology
The experiment was carried out in a reaction chamber as illustrated in Fig. 1 (a). A LED light bulb (240 V, 13 W, 1521 lm, Philips, Netherlands) was installed in the middle of the reaction chamber. The temperature (T) was constantly kept at 21 • C to limit as much as possible the interference of thermal degradation during the experiment. Air and N 2 supplies were used to achieve [O 2 ] (v/v %) at either high (21%) or low (0%) levels respectively. Gas flows from these supplies were regulated by Aalborg GFC aluminium body mass flow controller (Caché Instrumentation, UK). The flows then passed through a V-GenTM Dew Point/RH Generator (InstruQuest Inc., US) to achieve RH at high (70%) and low (20%) levels before entering the reaction chamber. The controlled atmospheres were continuously circulated at 200 ml•min − 1 through the system 24 h before and during the experiment.
For each experimental run, five distinct sacrificial rag paper sheets were used. These paper sheets did not have associated dates and were identified as rag paper by visual examination of the presence of sieve marks and random fibre orientations. 1.0 cm * 1.5 cm subsamples were cut from each sheet ( Fig. 1 (b)) and were attached to Whatman filter  the runs. For each run, it took ~1 h for the environment inside the reaction chamber to stabilise after interruption and the samples were degraded for another ~2.5 h (~0.14 day in total) to reach a measurable change in R before the first measurement was taken. Subsequent measurements were taken at intervals of 0.2-5 days until the measurements reached a plateau. The duration of the runs ranged from 11 to 30 days with at least eight data points obtained. The measurements were taken using a hyper-spectral imaging system (Camlin, Lisburn, UK). The system was equipped with a VNIR camera (spectral range: 400-1000 nm, spectral resolution: 2 nm) with an XENOPLAN 2.8/50-0902 lens (Schneider-KREVZNACH, Germany) and a halogen light source. Both dark and white calibrations were carried out before each scan at 0.8 mm•s − 1 with 50 ms exposure time and f/5.6 aperture. The spatial resolution of the images was 625 ppi.
R of each sample was represented by the average R of all the pixels of uncontaminated paper area on that sample. For the analyses of reflectance, change in reflectance (ΔR, day − 1 ) was calculated by subtracting the reflectance spectrum before degradation from the spectrum collected at each time point. For each band-pass filter, ΔR was averaged across all the samples for each experimental condition to construct a data cube, where the first dimension was the experimental condition, the second dimension was ΔR over the reflectance spectral range (λ) of 424-853 nm, and the third dimension was the time of degradation (t). The spectra of ΔR were smoothed using Savitzky-Golay filtering (2 nd order polynomial, window width 19) [23].
For the analyses of tristimulus colorimetry, R of each sample was truncated to 424-780 nm and analyzed in the CIELAB color space (Commission Internationale de l'Éclairage, abbreviated as CIE). X, Y, Z values in CIE 1931 XYZ color space were first calculated for each sample using the following equations: where x, y, and z are CIE 1964 standard 10 • colorimetric observer functions which are plotted in Fig. 2 (a), S is the reflectance of the sample, I is the relative SPD of the CIE standard illumination D65 as shown in Fig. 2 (b), and n is a normalising constant. X, Y, and Z were then used to obtain L*, a*, and b* based on: , X 0 , Y 0 and Z 0 are the tristimulus values of the reference white for the CIE standard Illuminant D50. Based on these calculations, the total color change of each historical rag paper sample (ΔE 00 ) was calculated using the formula given by CIE [2,24] at each time a measurement was taken. ΔE 00 was then average across all the samples degraded under the same condition for further analyses. All the data analyses were carried out in MATLAB® R2017a.

Spectral colorimetric change
Given that the monochrome light sources were generated using band-pass filters with a broadband LED radiation source with nonuniform spectral power distribution (Fig. 1 (c)), ΔR was normalized by the number of photons that reached the samples. The number of photons (N, mol) was calculated based on the Planck-Einstein relation: where Λ is the wavelength of the radiation in nm, E is the total photon energy reached the sample surface in J The progression of ΔR N over time (t/day − 1 ) followed a logarithmic pattern. This is similar to the behavior of historical rag paper irradiated by a broadband LED light source, which has been discussed in detail by Liu [20]. Therefore, a similar modelling approach was taken, where the relationship between ΔR N and t was modelled using a logarithmic relationship: where A 0 is the normalized amount of spectral change took place in one day, the first measurement was taken at t = 0.14 day, and k ΔRN is the rate constant of the normalized rate of spectral change (K ΔRN ) derived according to the following equations: Based on these four equations, k ΔRN was calculated over 424-853 nm at approximately 50 nm intervals for the samples exposed to [ [20], k ΔRN-B , k ΔRN-G , and k ΔRN-R decreased as λ increased with positive maxima reached at the short end of the λ range. k ΔRN-B , k ΔRN-G , and k ΔRN-R obtained at high RH were generally higher than those obtained at low RH, suggesting that RH was likely to play a positive role in driving the rate of discoloration of historical rag paper. It should be noted that k ΔRN is generally positive when λ < 600 nm and negative when λ > 600 nm. This indicates that R increased at short wavelengths and decreased at long wavelengths, which led to blue-shift of the reflectance spectra as the samples degraded. It should also be noted that k ΔRN-B , k ΔRN-G , and k ΔRN-R are much smaller than the rate constant obtained with broadband radiation using the same experimental setup [20]. This suggests that the signal to noise ratio was likely to be relatively lower in this research. For the even smaller k ΔRN obtained at λ > 600 nm, where the data scatter was also larger, the effective spectral change was likely to be un-measurable and the values mainly represented noise.
To elucidate the wavelength dependence of the spectral changes, the relationship between k ΔRN and Λ was explored as shown in Fig. 3. It clearly shows that under all the combinations of [O 2 ] and RH, [Λ 450 ] induced the fastest changes across 424-853 nm whereas [Λ 625 ] induced the slowest. On average, k ΔRN-B was found ~12 times as much as k ΔRN-R and ~3 times as much as k ΔRN-G across different environmental conditions. This is likely to be an effect of the higher energy carried by the photons of 450 nm than the photons of 625 nm. Given that the relationship between photon energy (E) and Λ is reciprocal according to this equation: the relationship between k ΔRN and Λ − 1 was further plotted, as shown in Fig. 4. Nonlinearity was observed between k ΔRN and Λ − 1 at the selected λ in different conditions, with the possibility of linearity at a certain λ. This indicates that photon energy alone is insufficient to explain the observed changes and there were other factors that affected the discoloration of historical rag paper in the experiment. However, it is beyond the scope of this research to investigate these factors where additional experiments need to be designed and measurements of higher resolution need to be obtained.

Tristimulus total color change
For the management of discoloration of historical rag paper in practice, it is essential to understand the relationship between the spectral response and the physiological perception of color by humans. To investigate if tristimulus colorimetric assessment was in agreement with the measurements of spectral responses of historical rag paper, tristimulus total color change ΔE 00 in the CIELAB color-space was calculated as described in detail in Methodology. Similar to the normalisation of ΔR, normalized ΔE 00 (ΔE 00N ) was achieved by dividing ΔE 00 by the number of photons N (Equation 1) as: The progression of ΔE 00N over time is presented in Fig. 5. Across all the experimental conditions, ~50% of the total color change of the samples was completed in the first five days. Similar to the observations made for reflectance, the effect of [Λ] on ΔE 00N decreased as Λ increased. [Λ 450 ] induced remarkably faster and larger change in ΔE 00N , which was likely to be associated with the higher energy of photons of The dependence of k ΔE00N on Λ was investigated and illustrated in Fig. 6 (a). The relationship seemed to follow part of an exponential or a hyperbolic curve. Given the relationship between photon energy and Λ as described in Equation 5, the relationship between k ΔE00N and Λ − 1 was investigated as shown in Fig. 6 (b). Gentle curvature in this relationship was observed across the environmental conditions, which suggests that the relationship between k ΔE00N and photon energy of the light sources was not linear. This is consistent with what was observed for the spectral response. It indicates that for historical rag paper, the tristimulus compression and transformation of the spectral change did not seem to cause loss of qualitative information. This provides evidence of plausibility to use tristimulus colorimetry to qualitatively monitor the discoloration of historical rag paper in practice.
However, the tristimulus transformation was found to have quantitative impact on the results obtained by reflectance spectrometry. To reveal the impact, the relationship between k ΔE00N-B , k ΔE00N-G and k ΔE00N-R was examined and is shown in Fig. 6 (c). Linearity was observed across the four environmental conditions. k ΔE00N-B was about 2.5 times and 8.5 times as much as k ΔE00N-G and k ΔE00N-R respectively. These   Fig. 6 (a) reveals that faster discoloration took place when RH was high. At the same RH level, faster total color change was observed at higher [O 2 ]. This suggests that both RH and [O 2 ] promoted the total color change of historical rag paper. This observation is consistent with the discoloration behavior of historical rag paper induced by broadband radiation [20] and the discoloration of cellulose irradiated by near UV [25]. Based on this observation, the effects of [O 2 ], RH, Λ and their interactions on logarithmic k ΔE00N were further analyzed by analysis of variance (ANOVA) and a summary of the results is presented in Table 2. The results of ANOVA provided supporting evidence that all of [O 2 ], RH and Λ affected the discoloration of historical rag paper. Within the range of the factors investigated in the experiment, the main effects were found significantly strong with Λ showed the strongest effect. The interaction between [O 2 ] and Λ and between RH and Λ were also found significant. It suggests that [O 2 ], RH, Λ, [O 2 ]*Λ and RH* Λ need to be given attention in the management of discoloration of historical rag paper. In particular, controlling the SPD of light sources can be crucial in controlling the rate and the amount of discoloration.

Conclusions
This paper discussed the wavelength dependent discoloration of historical rag paper simultaneously affected by [O 2 ] and RH. Narrowband monochrome light sources centred at 450 nm, 525 nm, and 625 nm were used. The results obtained in normalized diffuse reflectance and tristimulus total color change were discussed and compared. The compression and transformation of tristimulus analyses of the spectral data was not found to cause loss of qualitative information regarding the progression of discoloration over time and the relative strength of the effect of [O 2 ], RH and Λ. This provides scientific evidence to support the use of tristimulus colorimetry as a way to qualitatively monitor the discoloration of rag paper-based collections in practice.
The progression of change in both diffuse reflectance and total color change over time was modelled using a logarithmic relationship and the normalized rate constant, k ΔRN and k ΔE00N , were obtained. In both cases, Λ was observed to have the strongest effect on the rate constant, where k ΔRN and k ΔE00N decreased as Λ increased. [O 2 ] and RH were observed to have promoted the discoloration. This suggests that lowering [O 2 ], RH, and the proportion of short wavelengths in the light source can be a practical management strategy to control the discoloration of historical rag paper-based collections. ANOVA confirmed these results and additionally revealed that the interactions between [O 2 ] and Λ and between RH and Λ were also significant. This reinforces that the spectral power distribution of light sources plays an important role in affecting the discoloration of historical rag paper. However, all of [O 2 ], RH, Λ, [O 2 ] *Λ and RH*Λ need to be monitored and assessed in the collection management to achieve the best outcome.
For k ΔRN , [Λ 450 ] induced about twice faster discoloration than that induced by [Λ 525 ] and more than 10 times faster response than that induced by [Λ 625 ]. For k ΔE00N , [Λ 450 ] induced ~1.5 times faster discoloration than that induced by [Λ 525 ] and ~7.5 times faster response than that induced by [Λ 625 ]. The discrepancy in the coefficients of proportionality between k ΔRN and k ΔE00N suggests that the tristimulus transformation caused quantitative distortion of the spectral assessments and may not be used for quantitative assessment of the spectrally dependent discoloration of historical rag paper in collection management. The coefficients of proportionality are likely to be associated with the higher energy carried by photons of 450 nm than photons of 525 nm and photons of 625 nm. Since the coefficients of proportionality were not equal to that of the photon energies, other factors related to λ and Λ were likely to have played significant roles in the discoloration processes of historical rag paper.