**3. Results**

The eight different BWS samples were subjected to a number of IS measurements that correspond to 20 years of regular monitoring, in order to assess the impact. Before any IS measurements and after certain numbers of IS measurements, the samples were measured with the photospectrometer in order to quantify the impact of the multiple IS measurements on the materials with an independent instrument. Figure 7 shows for BWS1 the differences in the photospectrometer spectra recorded after different numbers of IS measurements with respect to the initial spectrum before any IS measurement. BWS grade 1 with the lowest light-fastness was chosen among the eight samples as it is deemed to be representative for the most light sensitive archival materials such as newspapers, photographs and watercolours. The grey and the red dashed curves indicate the estimated error of the photometer measurements with, respectively, ±1*σ*(*λ*) and ±2*σ*(*λ*) ranges around the mean of the spectral differences measured for the reference area where both measurements were exposed to the same number of IS measurement. After 16, 22, and 41 IS measurements, the differences in measured reflectance for the wavelength range of 530 to 650 nm are consistently outside the ±2*σ*(*λ*) range. The photospectrometer is thus capable of detecting the corresponding small spectral differences in this wavelength range, which occur on the exposed area of the BWS1 sample after multiple IS measurements.

**Figure 7.** Changes of spectral reflectance values of BWS Grade 1 measured with the photospectrometer after increasing numbers of IS measurements as indicated. Additional dashed lines show the ±1*σ* and ±2*σ* intervals around the mean value from the unexposed reference areas at the corresponding wavelengths.

While comparing measurements at individual wavelengths is useful, it is possible that small, correlated changes in the spectra give rise to statistically significant differences before they can be detected at any of the individual measurements. In the present case the main question was whether any changes caused by repeated IS measurements could become visible to the human eye.

Figure 8 shows ∆E2000 for BWS Grade 1 as a function of the number of repeated IS instrument measurements. As described in the Materials and Methods section of this article, the ∆E2000 = 0.27 is the estimated error of the spectrophotometer measurement and values higher than this indicate statistically significant colour changes. The limit value of ∆E2000 = 1 is per definition considered to be the smallest colour difference a trained human observer can detect. The ∆E2000 = 0.7 level has been shown to be the threshold value above which the statistically more than 30% of the human observers are able to detect a difference in colour. All the measured colour differences remain below this more stringent threshold value for the 41 repeated IS measurements.

**Figure 8.** ∆E2000 of BWS Grade 1 as a function of the number of repeated IS instrument measurements. A value of 0.27 is the estimated measurement error of the spectrophotometer and 0.7 is the threshold value for visual perception of colour difference.

Results show that even very light-sensitive documents can be monitored with the SEPIA over periods of more than 20 years with the frequency of measurement of twice a year, before light exposure inherent to the measurement itself starts to effect any visible changes. These results are specific for the SEPIA device but other IS measurement systems could lead to similar results depending on their design. One of the key design features of the SEPIA instrument is that the wavelength selection for each spectral image is done inside the light sources. For each spectral image the measured object is irradiated only by the relatively low-intensity monochromatic light required to record this particular image. Other IS instrument concepts that rely on broadband illumination and sequential image recording with spectral filters at the camera expose the monitored objects to unnecessary high light doses and are therefore much less suitable for long-term monitoring. The camera lens and image sensor (CCD or CMOS) of the IS instrument has to be efficient in collecting the light reflected from the object, so that for a given light intensity the camera exposure time can be as short as possible. After the camera exposure has finished, the light source has to be switched off, so that the overall light dose received by the object during the entire measurement is minimized. At the time of the experiment, the Nationaal Archief was monitoring a range of circa 30 documents.

The Xrite spectrophotometer proved to be sensitive enough for control measurements as the colour changes induced in the most light-sensitive BWS became just measurable at the 15th repeated measurement, although still well below the visibility threshold. However, any further data analysis for the less light-sensitive BWS became obsolete with this, as even Grade 1 BWS was not affected significantly.

The second set of experiments addressed the sensitivity of IS measurements for detecting light-induced changes of six different materials. Figures 9 and 10 show the measured spectral changes for the six materials as induced by exposure to the maximum intensity of 5000 lux and a range of exposure periods. Please note the different y-axis scales in the diagrams of both figures. Figure 9 also includes the spectral changes at the minimum light intensity of 313 lux for BWS Grade 1. At the maximum light dose (i.e., at the maximum exposure time at the given intensity), and typically already at much lower light doses, all materials exhibit statistically significant spectral changes at several wavelengths.

For each material, the particular wavelength was determined where the maximum absolute spectral change was measured. The capability of the IS instrument to measure spectral differences in a given material can be expressed as the minimum light dose required to induce a spectral change that exceeds the measurement error at the materialspecific wavelength with maximum change. The wavelengths and values of maximum change, the error limits at these wavelengths and the minimum detectable light doses of all materials are given in Table 5.


**Table 5.** For each material at the maximum light dose of 500 klx·h: wavelength of maximum spectral change (column 2); measured spectral change (column 3), corresponding error limit at this wavelength (column 4); light dose at the minimum detectable change (i.e., at the error limit, column 5); reference to the graph (column 6).

> From the spectral reflectance curves measured with the IS instrument before and after accelerated ageing the function of induced colour change ∆E2000 vs. total light dose was calculated. Figure 11 shows the data for BWS Grade 1.

> In line with the reciprocity principle [35], it would be expected that for a given light dose the same colour change is obtained regardless of the actual combination of light intensity and exposure duration. However, the data points corresponding to 5000 lx clearly show a slower increase than the rest. This could indicate deviation from reciprocity at high intensities, however, it might also reflect an uncertainty of the intensity levels in the light ageing setup. In any case, the dependence of colour change on light intensity is sufficiently low for the purpose of this experiment, i.e., to quantify the lowest light dose that induces colour change detectable by the IS instrument.

**Figure 9.** Differences in spectral reflectance values as a function of wavelength for BWS Grade 1 exposed to 313 lux and 5000 lux, and for BWS Grade 2 exposed to 5000 lux, for different durations as indicated. The dashed black curves indicate the upper and lower limits of the measurement uncertainty at each wavelength.

**Figure 10.** Differences in spectral reflectance values as a function of wavelength for four different materials (paper with iron gall ink, lignin-containing paper, contemporary printer paper, rag paper) exposed to 5000 lux for different durations as indicated. The dashed black curves indicate the upper and lower limit of the measurement uncertainty at each wavelength. For contemporary printer paper at wavelengths <400 nm, the curves are influenced by fluorescence. The changes induced by light ageing at these wavelengths can probably be attributed largely to break-down of these fluorescent components.

**Figure 11.** ∆E2000 as a function of light dose for BWS Grade 1, obtained at different combinations of light intensity and exposure duration during the light ageing experiment. The actual light intensities are as indicated. The maximum colour difference measured for unexposed samples is ∆E2000 = 0.19, which is used to estimate the measurement error in Table 3.

For the BWS1 sample, the ∆E2000 = 0.7 visibility threshold is obtained at light doses ~20–30 klx·h, corresponding to ~60–90 exhibition days at 50 lx, 7 h/day, from the halogen light sources as used in the exhibition *De Verdieping van Nederland*. Furthermore, the conservative estimation of ∆E2000 = 0.5 as induced by IS measurements over 20 years can be estimated as equivalent to ~40–60 days of exhibition under the same conditions.

In addition to the ∆E2000, the standardized Euclidian distance ∆Euclid was calculated for the spectral reflectance curves measured after and before accelerated light ageing, as shown in Figure 12 for the BWS Grade 1.

**Figure 12.** Standardized Euclidian distances, ∆Euclid, for BWS Grade 1, as a function of light dose as obtained with different intensities, as indicated. The maximum distance measured for the unexposed samples is <sup>∆</sup>Euclid = 0.16 <sup>×</sup> <sup>10</sup>−<sup>3</sup> , which is used as an estimation of the measurement error in Table 3.

Tables 6 and 7 list ∆E2000 and ∆Euclid, respectively, at the maximum light dose of 500 klx·h, for all the sample materials. As expected on the basis of Figures 9 and 10, the biggest change is measured for BWS Grade 1 and the smallest change is measured for contemporary printer paper.

Based on the detection limits ε(∆E2000) and ε(∆Euclid) an estimate of the light dose required to induce the minimum detectable change was made for each material. The corresponding values are listed in the last column of Tables 6 and 7, respectively. The required light dose of 3.5 klx·h to achieve a detectable colour change in BWS Grade 1 samples corresponds to ~10 days of exhibition under the stated conditions.

**Table 6.** Colour change for each material at the maximum applied light dose of 500 klx·h, except in cases denoted with \*, where the maximum was measured at the stated lower light dose. The light dose corresponding to the minimum detectable change is defined as the minimum light dose at which the measured colour change exceeds the measurement error ε(∆E2000) as listed in Table 3. This minimum light dose is determined by linear interpolation of the measurement point just below and the point just above ε(∆E2000).


**Table 7.** Standardized Euclidian distance for each material at the maximum applied light dose of 500 klx·h, except in cases denoted with \*, where the maximum was measured at the stated lower light dose. The light dose corresponding to the minimum detectable change is defined as the minimum light dose at which the measured spectral change exceeds the measurement error ε(∆Euclid) as listed in Table 3. This minimum light dose is determined by linear interpolation of the measurement point just below and the point just above ε(∆Euclid).


For contemporary printer paper with its much higher light-fastness, the colour and the spectral changes at the maximum applied light dose of 500 klx·h is at the same level as the corresponding detection limits given by the estimated measurement errors. Note that the fact that the maximum colour and spectral difference were measured at lower light doses is also an effect of the measurement uncertainty. For the iron-gall ink on paper sample, the ∆E2000 colour change measured at 500 klx·h is also just at the level of the measurement error so that this light dose can only be used as a lower limit. The ∆Euclid value at 500 klx·h is about twice the estimated measurement error, allowing an estimate of 135 klx·h for the minimum detectable light dose.

Both ∆E2000 and ∆Euclid are generic measures for spectral change. It can be expected that by using material-specific measures that give higher weights to those spectral regions where the strongest changes are induced, the sensitivity of the IS measurements for detecting light-induced changes can be further improved.
