**2. Materials and Methods**

In this case study, the "SEPIA" Hyperspectral Imager (Art Innovation BV, now DEM-CON based in Enschede, The Netherlands) was used. The instrument features two identical wavelength-tunable light sources (70 spectral bands in the range 365 nm–1100 nm) and a monochrome CCD camera (4 megapixel) mounted at a 45◦/0◦ geometry with respect to the surface of the recorded document. This is a fully enclosed system, meaning that no external light can enter the recording area. Figure 1 shows a schematic drawing of the setup and an image of the instrument installed at the Nationaal Archief.

**Figure 1.** The case study Imaging spectroscopy (IS) instrument installed in the conservation laboratory of the Nationaal Archief [25].

During a measurement, the document is consecutively illuminated for pre-programmed exposure time periods by monochrome light from the light sources at each band. The instrument and the recording parameters are described in more detail in a previous publication [25].

For measuring the monitoring impact of repeated IS recordings, a set of Blue Wool Standard (BWS) acquired from Preservation Equipment Ltd. (based in Norfolk, UK) was used. BWS are dyed textile references with well-defined fading characteristics and they are available in a range of light-fastness grades [26].

These references are very well-known in the field of conservation as they are generally used to assess the impact of illumination during exhibitions in a qualitative way [27] or to calculate the cumulative light dose through calibration [28]. In this study, samples of grades 1 to 8 were mounted on a flat black-painted sample holder. To provide a reference, one half of the area of each sample was shielded with a paper board for most of the measurements, while the other half was exposed to the lights of the instrument at every IS measurement. The samples were exposed to 45 IS measurements, corresponding to a monitoring period of about 20 years. The monitoring period of 20 years was chosen as the best representation for the operational lifetime of such an instrument and because two recordings per year were considered for the exhibited documents at the Nationaal Archief as they follow a rotation schedule every six months. The effect of the IS measurements on the BWS samples was determined by repeated measurements of their reflectance spectra with a spectrophotometer (model Xrite i1pro-1) that provides reflectance values at every 10 nm over the spectral range of 380 to 730 nm, using a measurement spot diameter of 4.5 mm in a 45◦/0◦ illumination/detection geometry. After calibration with the white standard provided with the device (calibration software "i1 Diagnostic v 4.0.0.127"), 10 different spots were measured on each BWS sample. The reproducibility of these spot measurements is shown in Figure 2, with the spectrum of BWS Grade 1 before any exposure to IS as an example. The average spectral reflectance values over the 10 spot measurements are connected by straight lines. The error bars around each average value indicate ±1 standard deviation (SD) of the spectral reflectance values of the 10 individual measurements. For further analysis, the average spectra of the 10 spots were used.

To enable the assessment of the impact of IS, the reproducibility of the photospectrometer for measuring any spectral changes must be high enough so that the corresponding error on the measured colour change is below the visual detection threshold. This was verified by comparing any two photometer spectra (each averaged over 10 spots) that were measured on the same BWS reference area when it had been exposed to the same number of IS measurements. For each BWS sample, 51 unique pairs of such measurements are available. For each pair, the difference spectrum of the later minus the earlier one of both measurements and their colour difference ∆E2000 [29] were calculated. The visual thresh-

old was set at a value of ∆E2000 = 0.7, which based on the work of Pretzel [30] corresponds to the 30% probability that a human observer is able to detect a colour difference.

**Figure 2.** Spot measurement reproducibility of the Xrite spectrophotometer on the Blue Wool Standard (BWS) grade 1.

As seen in Figure 3, the mean value of the differences does not exceed the SD, and for most wavelengths it even remains considerably below the SD. This means that there is no significant bias (drift) of the photospectrometer measurements and no significant change of the spectra of the reference area in the course of the experiment. The SD of the photospectrometer measurements on the reference areas can therefore be used to estimate the measurement error: measured spectral changes on the exposed BWS areas can be assumed to be significant (i.e., real changes) if they exceed ±2 SD.

**Figure 3.** Average and standard deviation of the spectral difference curves based on 51 unique pairs of photospectrometer measurements of the BWS Grade 1 performed without an IS measurement cycle in between.

For each of the 51 pairs of reference area measurements, the colour difference (∆E2000) was calculated. The average colour difference is 0.14 (SD = 0.08), and for 95% (i.e., for 49 pairs) of the measured colour difference values were less than 0.27. The latter value can be used as a conservative estimate of the measurement error for the colour difference: if a colour difference measured on an exposed area exceeds this value, it very likely indicates a true colour change rather than a measurement error.

Spectrophotometer measurements themselves required irradiation of the BWS samples, and the overall results therefore represent an overestimation of both the measurement error and of the impact of long-term monitoring.

For testing the monitoring sensitivity, a second set of samples containing the BWS as well as common archival materials were exposed to photodegradation using a range of light doses. Table 1 reports the six test materials that were selected.

**Table 1.** Specifications of the materials used for the accelerated light-ageing tests.


The main light sources used in the exhibition *De Verdieping van Nederland* at the Nationaal Archief at the time of the experiments are quartz-tungsten halogen reflector lamps without infrared suppression, which are mounted on the room ceiling for spot illumination of the objects in their glass cases. The dedicated setup that was built for accelerated light ageing [33,34] was designed to approximate the light spectrum of the illumination of the objects in the exhibition. To achieve this, the same type of halogen reflector lamps were used and the light was transmitted through the top glass plate of an exhibition case. As opposed to the situation in the exhibition, a homogeneous distribution of the light intensity on the samples at the intended level was required. Therefore, the light transmitted through the glass was diffused by scattering it from white cardboard in two steps before it reached the sample areas. The homogeneity with each area was verified with a lux meter.

Figure 4 shows the spectral power distribution of the homogenized halogen light with which the samples were irradiated in this research.

**Figure 4.** Spectral power distribution of the light sources used during accelerated light ageing.

The setup was used to induce accelerated light ageing at five intensity levels, by placing a set of samples at a suitable distance from the diffused light source. The light intensity to which each sample set was exposed was measured with a lux meter and monitored after each exposition time.

To prevent excessive heating of the samples, which could impact on their ageing behaviour in an uncontrolled way, forced air cooling was applied to the confined space of the sample chamber. The air temperature and relative humidity were monitored inside and just outside of the sample chamber. During the light ageing experiments the temperature ranged from 20 ◦C to 26 ◦C and the relative humidity from 35% to 62% RH.

The illumination in the *De Verdieping van Nederland* exhibition at the time of the experiments were 50 lux for 7 h/day over a period of three months.

The samples in this research were exposed to light doses ranging from ca. 1.5 to 500 klx·h, which corresponds to a range of 4–1400 days of exhibition at the above conditions. Table 2 reports the range of light doses obtained using different combinations of irradiation intensities and exposure times. In addition, 5 subsamples of each material underwent the same preparation but remained covered (i.e., 0 lx) during the accelerated light ageing process to serve as reference samples, having experienced only a small light dose during sample preparation.


**Table 2.** Light doses in klx·h achieved through different combinations of intensity and exposure times, as applied to each of the 6 tested materials. The values in parentheses indicate the number of exhibition days equivalent to the light doses.

Table 2 also indicates the number of days in the exhibition that would result in the same accumulative light dose. The row with 0 lx indicates the reference subsamples that remained covered during the accelerated aging.

The number of 6 materials and the number of light intensities were chosen to optimize the recording surface of the IS instrument. For each of the six materials, 5 × 6 = 30 subsamples were cut out, to be used in the experiments. The samples were arranged to fit into the field-of-view (FOV) of the tested IS instrument (120 mm × 120 mm) in order to optimize the duration of measurements and improve the quality of the results.

Round samples (Ø 5 mm) were cut taking care they were taken from a homogeneous area of each test material. The pieces were mounted on five buffered RagMat Museum Board 4 ply Natural, without liquid adhesive but instead using Filmoplast® by Neschen that covered only small margins of the samples as shown in Figure 5a.

**Figure 5.** (**a**) The mounted samples used for the light ageing experiments. Rectangular regions-of-interest (ROI)s defined on the IS datacube before (**b**) and after (**c**) the light ageing experiment. Each ROI is drawn in the same colour in the 680 nm spectral grayscale images in both recordings.

The resulting set of 180 sub-samples was measured with the IS instrument twice: before and after the accelerated light-ageing experiment. Each measurement results in a calibrated hyperspectral datacube (a stack of calibrated spectral images) that contains

a spectral reflectance curve for each image pixel. For each of the 180 material samples a region-of-interest (ROI) consisting of 38 × 52 = 1976 pixels was defined, which corresponds to an area = 2.3 × 3.2 mm. Any slight shift of the samples in the field-of-view of the instrument, visually detected in the second measurement with respect to the first measurement, was compensated by a corresponding shift of the ROI areas, see Figure 5b,c. *Heritage* **2021**, *4* FOR PEER REVIEW 9

*Heritage* **2021**, *4* FOR PEER REVIEW 8

For each ROI, the mean spectral curve and the standard deviations (SD) over all 1976 pixels were calculated for both measurements. Figure 6 shows the mean spectral curves of one particular ROI for each test material before artificial ageing. The standard deviation reflects the combination of measurement noise of individual pixel values and the inhomogeneity of the material within the ROI area. Note that the peak in the curves of all samples at 380 nm is most likely caused by an imperfection of the corresponding spectral filter of the instrument, i.e., it is a measurement artefact rather than a reflectance characteristic of the samples.

For each sample, the difference spectrum was calculated by subtracting the corresponding ROI spectral curve before light ageing from the spectral curve of the same ROI after ageing. These difference spectra correspond to the spectral changes over the wavelength range of 365 to 1100 nm induced by ageing.  (**a**) (**b**) (**c**) **Figure 5.** (**a**) The mounted samples used for the light ageing experiments. Rectangular regions-of-interest (ROI)s defined on the IS datacube before (**b**) and after (**c**) the light ageing experiment. Each ROI is drawn in the same colour in the 680

An estimate of the measurement error at each spectral band was obtained from the difference spectra as follows. For each spectral band, the mean value and standard deviation for all 30 difference values (5 reference subsamples of all 6 materials) were calculated. The mean value can be interpreted as a systematic error, whereas the standard deviation reflects a random contribution to the error. Measured difference values outside the range of the mean value ± 2 SD are considered to be statistically significant, i.e., the object has changed. However, since the wavelength-dependent measurement errors vary for the different materials, more conservative (i.e., larger) error limits can be derived by extending the ±2 SD limits by the minimum and the maximum difference values for the particular material. nm spectral grayscale images in both recordings. The resulting set of 180 sub-samples was measured with the IS instrument twice: before and after the accelerated light-ageing experiment. Each measurement results in a calibrated hyperspectral datacube (a stack of calibrated spectral images) that contains a spectral reflectance curve for each image pixel. For each of the 180 material samples a regionof-interest (ROI) consisting of 38 × 52 = 1976 pixels was defined, which corresponds to an area = 2.3 × 3.2 mm. Any slight shift of the samples in the field-of-view of the instrument, visually detected in the second measurement with respect to the first measurement, was compensated by a corresponding shift of the ROI areas, see Figure 5b,c.

**Figure 6.** Mean reflectance spectra of the ROI of each of the 6 materials measured with the IS instrument before ageing **Figure 6.** *Cont*.

paper, contemporary printer paper, rag paper The error bars indicate ±1 SD at the corresponding wavelength.

*Heritage* **2021**, *4* FOR PEER REVIEW 9

**Figure 6.** Mean reflectance spectra of the ROI of each of the 6 materials measured with the IS instrument before ageing (from **top** to **bottom)**: Blue Wool Standard grade 1, Blue Wool Standard grade 2, iron = gall ink on paper, lignin-containing paper, contemporary printer paper, rag paper The error bars indicate ±1 SD at the corresponding wavelength. **Figure 6.** Mean reflectance spectra of the ROI of each of the 6 materials measured with the IS instrument before ageing (from **top** to **bottom)**: Blue Wool Standard grade 1, Blue Wool Standard grade 2, iron = gall ink on paper, lignin-containing paper, contemporary printer paper, rag paper The error bars indicate ±1 SD at the corresponding wavelength.

When considering degradation of archival materials, an important criterion is whether such degradation can be detected visually (e.g., by comparison with colour charts). Therefore, in addition to the spectral differences at all wavelengths, the colour difference values ∆E2000 were calculated for samples before and after ageing. These results are reported in Table 3.

**Table 3.** Estimated errors for the calculated ∆E2000 colour difference values and standardized Euclidian vector distance values, ∆Euclid, for the IS measurements of the aged test materials.


As an estimate for the measurement error ε(∆E2000), for each material the maximum colour difference measured between the exposed and unexposed samples was used. It therefore takes into account the repeatability of the IS measurement itself, including any residual ROI positioning error and also any change in the reference samples not exposed to ageing. Any detected colour change >ε(∆E2000) can thus be assumed to be a true change rather than measurement error. This means that ε(∆E2000) defines the detection limit of the method, for the respective material. Note that for all materials, the measurement error ε(∆E2000) is well below the threshold value of ∆E2000 = 0.7, above which there is a 30% chance that a colour change is visually noticeable for a human observer.

The ∆E2000 value indicates whether a colour change induced by the ageing could be visibly detectable. Taking into account the particulars of human colour perception in combination with standardized illumination condition, reflectance changes in different wavelength regions are weighted differently and by definition, reflectance changes outside the visible range are not taken into account for colour measurements.

An alternative measure of spectral change that takes into account all wavelengths in the measured spectral range is the so-called standardized Euclidian distance ∆Euclid. It is not based on (and limited by) human colour perception, and it uses all spectral values over the entire spectral range provided by the used instrument. The contribution from each wavelength to ∆Euclid is weighted according to the estimated error. It is defined as:

$$\Delta\_{\text{Euclid}} = \sqrt{\sum\_{i} \mathcal{W}\_{i} \cdot \left(\mathcal{R}\_{1}(\lambda\_{i}) - \mathcal{R}\_{0}(\lambda\_{i})\right)^{2}} \tag{1}$$

where

$$\mathcal{W}\_{\text{i}} = \frac{\sigma^{-2}(\lambda\_{\text{i}})}{\sum\_{j} \sigma^{-2}(\lambda\_{j})} \tag{2}$$

while the other symbols used are defined in Table 4. Note that all spectral measurements were carried out at the same spectral bands, which means that the sums in Equations (1) and (2) have the same number of terms for all samples. The standardized Euclidian distances are therefore comparable for all samples without the need to normalize for the number of spectral bands.

As an estimate for the measurement error ε(∆Euclid) the maximum ∆Euclid value determined for the 5 unexposed samples was used for each material reported in Table 1. Since the standardized Euclidian distance includes measurement data from additional spectral bands outside the visible range, there is the chance that allows the detection of changes of the monitored material before they can be detected by measurements in the visible range only.


**Table 4.** Symbols used in Equations (1) and (2).
