Methodological study on single grain OSL dating of mortars: Comparison of five reference archaeological sites

The aim of the preparation procedure for single SG-OSL measurements in this experiment was to extract the quartz grains of the size fraction 200–250 μm that were not exposed to light during sampling, following the hypotheses that the coarser fraction is supposed to be better bleached in the case of mortar (i.e. Goedicke, 2003; Jain et al., 2004). If the samples are compact, the external part that might have been briefly exposed to light during sampling is scraped off. In order to do so, the use of manual tools is preferable to using a saw since the latter causes quite important loss of material and there is a risk of sample disintegration during cutting. After removing a surface layer, the internal part of the sample is gently crushed into powder and sieved. If mortar has been sampled in dark light conditions in the form of powder, it can be directly sieved.

The chemical treatment includes three stages: 10% HCl, 15% H₂O₂ and a mixture of H₂SiF₆ with HNO₃ in the ratio of 9 to 1 (Urbanová et al., 2015). The last step takes about 1–2 weeks depending on the quantity of feldspars to be dissolved. A total dissolution of feldspar in the etched fractions from mortars is verified before starting the dating procedure by IRSL test (Urbanová et al., 2015). The external alpha contribution to the annual dose (“k” value) taken into account in this work is equal to 0.05 ± 0.02 (Blain et al., 2007; Guibert et al., 2009a; further details in Urbanová et al., 2015).

The quantity of the bulk sample needed for the preparation is variable, depending on the abundance and the grain size of quartz in mortar. Nevertheless, between 50–100 g of raw material was generally sufficient to obtain between 200 and 500 mg of quartz size fraction 200–250 μm.

The amount of quartz grains necessary for SG-OSL analyses including measurements of equivalent doses, dose recovery tests and IRSL tests depends on the proportion of luminescent grains in the sample and its degree of bleaching. To give a general idea, for the well-bleached sample with 5% of luminescent grains we needed about 53 mg of material (3800 quartz grains of size fraction 200–250 μm) to obtain the distribution of 190 individual equivalent doses.

All OSL measurements were performed using a TL/OSL DA20 Ris⊘ reader. The light detection system of the reader consists of an EMI Q9235 photomultiplier tube and 7.5 mm of Hoya U-340 filter for detection in the UV-blue wavelength range (about 280–370 nm). A ⁹⁰Sr/⁹⁰Y beta source was used as an irradiation source (dose rate 0.150 ± 0.005 mGy/s on the 1^st January 2014). Two different luminescence stimulation systems depending on the measurement protocol were used. The OSL measurements by a standard multigrain technique use Blue LEDs (NISHlA type NSPB-500s) with a peak emission at 470 nm for the stimulation, while a single grain analysis was performed by a 10 mV Nd: YVO₄ solid state diode-pumped laser emitting at 532 nm.

The measurements of recovery dose and equivalent doses were performed applying the SAR protocol (Murray and Wintle, 2000) using the parameters described in Table 2. OSL signals from multi-grain aliquots are based on the summation of the first 0.8 s of stimulation corrected for background derived from the last 8 s (time of stimulation: 40 s). Single-grain OSL signals are derived from the summation of the first 0.05 s of stimulation less the sum of the last 0.2 s (time of stimulation: 1 s). For each grain or disc, the sensitivity-corrected regenerated OSL signals (L_x/T_x) were fitted with an exponential function and each individual equivalent dose (D_e) was estimated by projecting the sensitivity-corrected natural signal (L_N/T_N) onto the fitted curve. The standard error on the individual equivalent dose was obtained by Analyst version 4.11 from counting statistics, curve fitting and the instrumental reproducibility error of 2.7% (the value calculated from the series of measurements performed in the IRAMAT-CRPAA laboratory in Bordeaux).

In general, single grain analyses reveal significant heterogeneity of luminescent properties between individual grains. Such heterogeneity cannot be noticed by any classical multigrain technique because of averaging effects that occur when tens or hundreds of grains are measured together. Selection criteria used to evaluate single grain measurements should therefore be logically less strict than for multigrain analyses. In this work, the grains were retained following criteria mentioned below also used by other authors dealing with weakly luminescent samples (for example Thomsen et al., 2004; Jacobs et al., 2006, 2013; Medialdea et al., 2014):

relative error on recycling ratio error did not exceed 25% (related to the reference value 1)

The influence of the selection criteria on the global archaeological dose (paleodose) was evaluated by comparing the archaeological dose calculated for all the grains giving a signal with the archaeological dose obtained only when the selected grains were taken into account. Even if no significant differences were observed in our case, we still consider the application of these criteria justified for obtaining reliable estimates of equivalent doses.

The use of a SAR protocol for SG-OSL measurements on young samples with low sensitivity, such as mortars, needs to follow certain adjustments, unlike a standard procedure (Murray and Roberts, 1998). First, a test dose cannot represent 10% of an expected archaeological dose as is usually employed for sediments, since such low doses (about 0.1–0.3 Gy) would result in the loss of any OSL signal and in the impossibility of signal normalization. Therefore, a test dose between 1.5 and 4.8 Gy was employed. In addition, due to a supposed high variability of equivalent doses related to poor bleaching, a growth curve has to be sufficiently extended to estimate equivalent doses in a larger interval of values. The first two regeneration doses were chosen sufficiently close to the expected archaeological dose but still high enough to get the OSL response, while two higher regeneration doses were chosen 10 to 30 times higher to get an approximate equivalent dose estimation for poorly bleached grains. Nevertheless, the highest equivalent doses corresponding to extremely poorly bleached grains did not enter into this interval (see Table 3, the last column). The equivalent dose estimation of these grains is performed by the extrapolation of the exponential growth curve. If the proportion of such grains is significant, higher regeneration doses should be employed in order to extend the dose-response curve and to avoid the imprecisions in equivalent dose determination of the poorly bleached grains.

The choice of the convenient preheat temperature is crucial for a correct equivalent dose estimation. The preheat temperature should not be lower than 160°C in order to avoid unstable light sensitive traps contributing to a detected signal. However, a highly elevated temperature can provoke thermal transfer which can also negatively affect a measured luminescence.

The experiment on mortar dating was started with Gallo-Roman samples, being older and therefore generally more luminescent than mortars from younger periods. The choice of measurement parameters was based on multigrain measurements which indicated the best results for the preheat temperatures about 240°C. Apart from single grain dose recovery measurements, preheat plateau tests and thermal transfer tests were performed by standard multigrain technique in order to verify the pertinence of this choice (e.g. Murray and Olley, 2002, Wintle and Murray, 2006; Medialdea et al., 2014). However, exploitable results could have been obtained only for well bleached samples from Antibes (Urbanová et al., 2016). Given that these tests are carried out by multigrain technique that allows the measurement of an average signal, it cannot be applied on poorly bleached samples with large scatter of equivalent doses. Performing the same tests by single grain technique would be a time-consuming procedure due to the generally low sensitivity of mortars to the OSL stimulation. Thus, for poorly bleached samples, the solution how to test the dependence of a natural dose on preheat temperatures could be to perform two series of single grain measurements at two different preheat temperatures. This was done in the case of the sample BDX 15541 from Palais-Gallien for which no significant difference between the measurements at 190°C and 240°C was observed.

The single grain measurements of mortar samples from medieval monuments studied in the second part of the project were systematically carried out at 190°C. If the dose recovery ratio entered the interval 1.00 ± 0.10, the preheat temperature of 190°C was used for equivalent dose determination. Even if we did not observe any increase in equivalent dose with rising preheat temperature, we recommend using the preheat of 190°C and the cut-heat of 160°C as initial parameters in order to minimize the potential risks of thermal transfer that would significantly change kinetics of luminescence. If the dose recovery tests are satisfactory, these temperatures can be kept for equivalent dose determination. The preheat temperature between 180–200°C for young samples is also recommended by other authors (Murray and Olley, 2002; Jain et al., 2004; Goedicke, 2011; Medialdea et al., 2014). The SG-OSL measurement parameters applied on the sample sets studied are detailed in Table 2.

The OSL signal was dominated by the fast component for the quartz samples presented in this study, which is a primary condition for dating mortar. This aspect is usually evaluated by the LM-OSL test consisting in the linear increase of stimulation power during the OSL measurement. However, generally, this test can only be performed by multigrain technique as was the case in this study since routine experimental apparatus does not allow the performance of this kind of measurement by single grain procedure.

The presence of medium and slow components in the OSL signal can also be deduced from the form of the OSL decay curve. In this study, we systematically observed a very fast OSL decay during the first 0.1 s of stimulation for grains that showed they were well-bleached (Fig. 1a). On the contrary, for certain poorly bleached grains giving high equivalent doses the OSL decay was slower (Fig. 1b). The occurrence of medium and slow components in some cases of LM-OSL tests, as seen for some samples from Palais-Gallien and Saint Seurin (Fig. 1c), is principally linked to the presence of one or several poorly bleached grains with high signals on the same multigrain disc, but this does not characterize a behavior of all the grains on the disc (as observed also by Bulur et al., 2001). Therefore, a classical LM-OSL test carried out on multigrain discs gives only an average signal and is not a reliable and direct indicator of luminescent properties for individual grains.

Figure 1

“It is clear that there is significant over-dispersion in natural well-bleached samples, but some studies have also shown that there is significant over-dispersion even in laboratory dose recovery experiments where all extrinsic sources of variability have been eliminated.” (Thomsen et al., 2012) That is what we call “intrinsic variability”. Table 4 presents over-dispersion values from dose recovery measurements (OD DRT) of mortar samples bleached artificially in the laboratory. In this first case, over-dispersion corresponds to intrinsic variability. At the same time, over-dispersion values from equivalent dose measurements (OD ED) of natural mortar samples are shown. In this second case, only part of over-dispersion is accounted for by intrinsic variability, while another part is caused by the combination of extrinsic factors. For all the samples, natural over-dispersion is systematically higher than over-dispersion from dose recovery tests.

If we assume a transmissibility of properties in grains analyzed in DRT tests to grains used for ED measurements, the difference between OD ED values and OD DRT values will give an approximate contribution of poor bleaching and microdosimetric effects to the scatter in ED distributions. One must be conscious of the fact that DRT tests are not performed on the same set of grains as the ED measurements and that the process of artificial bleaching only simulates the process of natural bleaching, so some differences can occur. In our experiment, dose recovery ratio values on the samples artificially bleached in a solar simulator were about 0.05–0.09 more elevated than when the same samples were bleached directly in the OSL reader (Murray and Wintle, 2000). Similar discrepancies were also observed by Choi (Choi et al., 2009), who raised the hypothesis that the ultraviolet component of the solar simulator source may provoke excitation of deep OSL traps, contributing consequently to the measured OSL signal. Other experiments focused on the comparison of different kinds of artificial bleaching would be useful in future.

Since OD DRT represents the best estimation of intrinsic variability when no well-bleached variant of poorly bleached sample to be dated is available (Thomsen et al., 2012), we suggest performing artificial bleaching in the OSL reader, that generally gives satisfactory results. As we will see further, these measurements will be important for the use of minimum age models.

Microdosimetric variations can be caused by the heterogeneous distribution of radioelements in the mortar matrix. As a consequence, despite the fact that two grains could have been well-bleached when the building was constructed, they will give two different equivalent doses. Taking into account the size of the mortar samples from which quartz grains for dating purposes are extracted, we assume that these grains received the same gamma and cosmic annual dose rates. By contrast, we consider that if some microdosimetric variations at the grain scale occur, these are principally due to the heterogeneity of beta irradiations, originating from three series of radioelements: potassium-40, the series of thorium-232 and the series of uranium-238 and uranium-235. Among minerals contained in mortar, potassium is mainly present in potassium feldspars, in micas and in certain clays, whereas uranium and thorium can be found in abundance in heavy minerals such as zircons or apatites. Three analytical methods were used to assess microdosimetric variations: petrography, beta autoradiography (Fig. 4) and cartography performed by scanning electron microscopy coupled with energy dispersed X-ray spectroscopy (SEM-EDX).

Fig. 4

Beta autoradiography enables the study of the distribution of beta emissions in the mortar matrix, however, it does not make it possible to distinguish between different emitting radioelements (Ruffer and Preusser, 2009). Apart from the sporadic presence of zircons in the Antibes and Chassenon mortars, no significant quantity of zircons or apatites was noted in the rest of the studied samples. It was therefore assumed that the most important part of microdosimetric variations arises from potassium feldspars. SEM-EDX images in Fig. 5 show elemental mapping of potassium in thick sections of different mortar samples studied. The higher the number of X rays emitted by potassium and detected by the instrument, the brighter the tones of grey obtained in corresponding spots that clearly correspond to potassium feldspars. The relationship between their quantity and size and microdosimetric variations expected is discused in the following paragraphs.

Fig. 5

The early medieval (BDX 16492 and BDX 16493), Romanesque (BDX 16593) and Renaissance samples (BDX 16587 and BDX 16591) from the Saint Seurin basilica in Bordeaux correspond to the group of the weakest radioactivity from mortars studied (Fig. 4). Hardly observable differences of tonality imply small relative variations in beta irradiations. Nevertheless, we have to mention that here we approach the detection limits of the method and the absence of differences in tonality does not necessarily mean that no variations occur.

According to SEM-EDX cartography and petrographic observations, all these mortars contain very low quantity of K feldspars. If some K feldspars occur (e.g. in the sample BDX 16493, Fig. 5f), their dimensions do not exceed 500 pm in diameter. A minor quantity of K feldspars partially explains low radioactivity and the apparent homogeneity of beta irradiations. Therefore, we assume that the contribution of microdosimetric effects to the dispersion in ED distributions won’t be significant for this group of samples.

The last aspect to be evaluated is a bleaching degree. We considered that the mortar samples were well-bleached if the age calculation using the archaeological dose obtained with the Central age model (CAM, Galbraith et al., 1999) was in agreement with the reference age, taking into account that all grains meet the selection criteria. Afterwards, we checked what could be an indication of good bleaching if the age of the sample wasn’t known.

It was observed that for well-bleached mortars studied in this work over-dispersion values did not exceed 50%, whereas for poorly bleached mortars this value was systematically higher than 100%. These findings are in agreement with those suggested by Arnold et al. (2009). Also, well bleached distributions were symmetrical; the ratio of the lowest measured dose to the mean of the distribution (R_min, Table 6) was similar to the ratio of the mean of the distribution to the highest measured dose (R_max, Table 6). In contrast, for poorly bleached samples R_max was 2–3 orders of magnitude higher than R_min(Table 6). It is necessary to specify that the mentioned criteria seem to be valid for all the studied mortars whose age does not exceed 2000 years. For Paleolithic sediments the asymmetry of the distribution can be less evident.

The samples identified as well bleached originate from Antibes, Chassenon and Basel as demonstrated by narrow ED distributions centered around the value of 3, 11 and 2 Gy, respectively (Fig. 2). All of them come from Antiquity or the late Antiquity period. Archaeological doses of these samples were determined by the Central age model (CAM, Galbraith et al., 1999) and are presented in Table 7.

For the Antibes and Chassenon mortars, we observe agreement between the reference ages and the ages obtained by SG-OSL. For Antibes, the coherence between multigrain and single grain measurements, confirming a good bleaching degree, has already been demonstrated (Urbanová et al., 2016). For Chassenon, only one out of four samples could be dated because of the insufficient quantity of quartz grains in the other three mortars. Since the factor of poor bleaching is eliminated for these series of samples (σ_x, = 0), the over-dispersion value measured for these samples arises from the combination of microdosimetric variations and intrinsic variability (σ² = σ<!-- s -->m2+σ<!-- s -->a2 $\sigma^2_m+\sigma^2_a$ ). By subtracting σ<!-- s -->a2 $\sigma^2_a$ (the square of OD DRT) from σ² (the square of OD ED) (Table 4) and subsequent root extraction, we obtain the contribution of microdosimetric variations to the dispersion of ED distributions, equal approximately to 32% for Antibes samples (average for five samples)and 45% for Chassenon. Despite this fact, the use of the central age model for the determination of the archaeological dose leads to a correct estimation of the ages.

The Basel samples present some particularities and have to be discussed separately. As mentioned before, 5 out of 8 Basel samples did not emit any OSL signal. The age of the Basel Roman sample BDX 16304 is over-estimated despite its good bleaching. Based on petrographic observations, it is possible that an important quantity of luminescent quartz grains originates from crushed brick fragments present in this mortar. The annual dose rate of these grains is higher than the average annual dose rate determined by gamma spectrometry measurements, since the brick fragments are more radio-active than the rest of the mortar matrix (Fig. 4). Such discrepancy can explain the final over-estimation of the age of the sample BDX 16304. This example demonstrates the importance of the preliminary characterization of mortar for the correct interpretation of the results.

The samples BDX 16293 and BDX 16299 were very weakly luminescent (less than 0.5% of grains gave the signal). For reasons of time constraints, these samples were analyzed by multigrain technique and a complementary SG-OSL analysis was performed only for the sample BDX 16299. Even if we considered the probability of the presence of more than one luminescent grain per disc as relatively low, the SG-OSL measurement showed the presence of some poorly bleached grains on 2 out of 6 single grain discs (Table 8). This observation explains the over-estimation of the age obtained for the sample BDX 16299. It may be therefore concluded that even for such weakly luminescent samples, an averaging effect occurs when using the multigrain technique (Jain et al., 2004) and its use should be avoided as much as possible if we want to obtain reliable dating results.

Due to a large scatter of ED values, the archaeological dose in the case of poorly bleached samples was calculated by using the following models: –

3-parameter minimum age model (MAM-3, Galbraith et al., 1999)

To run these models, an input parameter σ_b (corresponding to a for IEU) defining the expected scatter of equivalent doses in the well-bleached part of the population is needed. Due to a small quantity of material available for analyses, additional dose recovery tests on bleached but non-irradiated samples could not be performed and the b parameter in IEU model was therefore considered 0. This implies that no over-dispersion for a given dose of 0 Gy is supposed, which may slightly differ from real behavior of the samples. Given that the expected values of archaeological doses are not lower than 1 Gy, such discrepancies should not have significant effect on the age determination. The results are summed up in Table 7. The poorly bleached mortars analyzed in this study can be basically divided into two groups.

The principal difficulty of existing minimum age models is that an input parameter is needed to run the model, and the right estimation of this parameter is very complicated for mortars affected by both poor bleaching and heterogeneous microdosimetry. The MAM-3 and MAM-4 often fail to fit experimental parameters. The IEU gave quite satisfactory results in many cases, however, it seems to work mainly if there is a significant frequency peak in the lowest dose region of ED distributions since the model aims at the lowest Gaussian distribution (provided that the input parameter correctly describes the data). Nevertheless, the model is not adapted for the evaluation of the distributions of extremely poorly bleached mortars when the form of the beginning of the distribution deviates from the Gaussian one. Alternative models that seem to be better adapted for the series of samples studied are being developed and tested on the data acquired in this work (Guibert et al., submitted; Christophe et al., submitted).

Currently, it is not possible to quantify with satisfactory precision the microdosimetric variations to which natural archaeological samples are subjected. Beta autoradiography enables the study of the distribution of beta emissions in the matrix, however, it does not make it possible to distinguish between different emitting radioelements (Ruffer and Preusser, 2009). EDX-SEM cartography allows indirect mapping of the distribution of potassium, but not of uranium and thorium. Such mapping of all three radioelements can be performed by ICP-MS analyses, if available. In addition, the distribution of uranium and thorium could potentially be studied via the distribution of alpha irradiations by using SSNTD detectors (Sanzelle et al., 1986; Wagner et al., 2005; Grainger, 2009). A priori, low background gamma spectrometry measurements of K, U and Th allow the identification of the predominant beta emitting radioelement.

Currently, different approaches aimed at a quantification of microdosimetric variations in archaeological materials by means of computer simulations (Martin et al., 2015; Lebrun et al., in preparation) are being developed. The integration of these approaches for the samples presented in this study might be attempted in future.