Testing feldspar luminescence dating of young archaeological heated materials using potshards from Pella (Tell Tabqat Fahl) in the Jordan valley

The archaeological materials came from three deposits in Trench XXXIIFF, locus 10.6, locus 15.1 and locus 15.2, all excavated in January 2013 (Bourke, 2013, 2014). All were drawn from thick Iron Age destruction debris layers, which sealed all earlier construction and occupation within the Iron Age Civic Complex. Based on typology, almost all the pottery from this horizon dates from the later Iron I and early Iron IIA periods (ca. 1050–900 BCE).

From each of the loci, around 20 potsherds were collected for luminescence dating (Fig. 2). Because of the destructive nature of this method, we were restricted to using samples of little or no use in any further archaeological study; as a result no meaningful diagnostic studies could be undertaken on these potsherds individually and the associated ages can only come from the ages assigned to the layers as a whole. The samples were taken in January 2013 in such manner as to avoid any direct sunlight exposure.

Figure 2

Any soil attached to the potsherds fragments was kept as a light-protection cover and later used for external dose rate measurement. In addition, ∼ 30 g of soil from each layer was collected directly for dose rate measurement.

The potsherds were processed under subdued red light following procedures described in Aitken (1985). Water content measurements were undertaken on each of the fragments before any chemical treatment. After removing around 1 mm of the surface, a double face vice was used to break the fragments and ease the process of crushing with the agate mortar. Two fraction of sand size were extracted; 90–300 μm for coarse grains preparation and <90 μm for fine grains preparation.

For both grain sizes, material was cleaned with 10% of HCl for 3 hours, followed by a overnight rinsing in 10% of H₂O₂ to remove any organic matter, rinsing in 0.01 N sodium oxalate to separate clay grains; repeated washing was carried out after each step. For coarse grains, samples were treated with 10% HF before separating K-feldspar and quartz using a heavy liquid solution (ρ < 2.58 g/cm³), the quartz-rich fraction was then etched for 40 minutes with 40% of HF to remove the outer alpha-irradiated shell from the grains.

Fine grains in the size 4–11 μm were extracted by applying Stokes’ settling (Frechen et al., 1996), around 1 mg to 3 mg of sample in acetone suspension was settled on 9.7 mm diameter aluminium disks and dried at 50°C.

All luminescence measurements were carried out using Risø TL DA 20 reader, equipped with IR diodes emitting at 875 nm, blue diodes emitting at 470 nm, and a calibrated ⁹⁰Sr/⁹⁰Y beta source. Blue-stimulated UV OSL was measured through a U-340 glass filter, and IR-stimulated blue light was measured through a BG-39/7–59 filter combination (Bøtter-Jensen et al., 2003).

External dose rate: Sediment samples from each layer were collected for gamma dose rate measurement. Radionuclide concentrations were measured using high resolution gamma spectrometry (Murray et al., 1987) and these concentrations were converted into infinite-matrix dry dose rates using the factors given by Guérin et al. (2011), and the assumption that there was (20 ± 10)% escape of ²²²Rn during burial (this covers all reasonable scenarios). Typical shards are 5 to 10 mm thick; according to Aitken (1985) this will result in an external gamma dose rate to the shard of ∼90% of the infinite matrix dose rate (with a further contribution of ∼ 10% of the internal infinite matrix gamma dose rate), for these potsherds, a typical dose rates in the sediments are similar to those in the shards, and so the error involved is <<5%. The radionuclide concentrations and derived external infinite-matrix dose rates are summarised at the end of Table 4.

Internal beta dose rate: because of the small size of the potshards samples, it was not possible to use gamma spectrometry to determine the internal dose rate. Instead, a Risø low-level gas flow beta multi-counter was used to estimate the total beta activity (Bøtter-Jensen and Mejdahl, 1985, 1988). This approach has the advantage of taking into account any disequilibrium in the U series. About 3 g from each potsherd was ground and mixed with wax (to retain ²²²Rn) before measurement. We adopted the counting geometry and sample preparation developed by Ankjærgaard and Murray (2007) and compared the observed beta count rates (counts per kilo seconds, cks^-1) from Uranium series, Thorium series and ⁴⁰K standards with the predicted infinite-matrix beta dose rates from the same standards, derived from gamma spectrometry analysis as discussed below.

Two considerations were taken into account before using the conversion factors from Ankjærgaard and Murray (2007). First, they neglected the contribution of beta dose rate from other nuclide contamination in their ²³⁸U and ²³²Th standards. In order to define the beta count rate from each nuclide chain individually, we used the gamma spectrometry analysis of each standard to derive the beta dose rate contribution fraction of ²³⁸U, ²³²Th series and ⁴⁰K to the total beta dose rate. For the ⁴⁰K standard 100% of the dose rate was derived from ⁴⁰K, for ²³²Th standard, 88.8% was derived from ²³²Th series, and for the ²³⁸U standard, 88.7% was derived from the ²³⁸U series. The observed beta count rate from each standard was multiplied by these factors before deriving the conversion factor from beta count to beta dose rate.

In addition, Ankjærgaard and Murray assumed 20% loss of radon when calculating the conversion factor in their U-series calibration; here we make the more correct assumption of full radon retention in these standards (mixed with wax). The revised conversion factors are summarised in Table 2 together with the original conversion factors of Ankjærgaard and Murray (2007).

Table 2

Activities derived from gamma spectrometry for the standards used in Ankjærgaard and Murray (2007) together with a comparison of their conversion factors and those derived here using the modifactions discussed in the text.

In order to test the reliability of the new conversion factors, the beta count rate was measured for the soil samples from Pella site and several other samples that are not related to the current study, the average of the ratio of the predicted beta count rate derived from gamma spectrometry to the observed count rate was 1.02 ± 0.03 (n = 20); this average included ²³⁸U and ²³²Th standards, and the measurement of a composite of 14 potsherds of each layer.

Internal alpha dose rate: Ankjærgaard and Murray (2007) did not consider the possibility of deriving an alpha dose rate from beta counting. We have used the known activity concentrations in their U and Th- series radionuclide standards to derive infinite matrix alpha dose rates (allowing for ²³²Th contamination in ²³⁸U standard and vice versa). The observed beta count rates, after correcting for the real contribution (see above) to beta count rate from each decay series in the standard sample, were divided by these infinite matrix alpha dose rates to give the conversion factors given in Table 2. The observed beta count rate from a shard includes some contribution from ⁴⁰K, which does not contribute to the alpha dose rate. Rather than assume some global average fractional contribution to beta count rate from ²³⁸U and ²³²Th, we derive a more sample specific conversion factor, excluding the beta count rate contribution of ⁴⁰K, by estimating the average activity ratios from measurement of the average ²³⁸U, ²³²Th and ⁴⁰K activities in our shards; these were estimated by combining the beta counting samples of 14 potsherds from each layer into a single gamma spectrometry sample (Table 3). The resulting activities suggest that, in our samples, 45.1 ± 1.6% of the beta counts are derived from the U-series, and 5.5 ± 0.3% from the Thorium-series from Layer L10.6, 54.5 ± 3% from the U-series and 4.5 ± 0.5% from Thorium-series for the middle layer L15.1, and finally the bottom layer L15.2, the fractional contribution from Uranium series was 45.4 ±1.9% and for Thorium series it was 5.4 ± 0.5% (Table 3). On average, the likely beta count rate derived only from the U- and Th-series for each shard were then derived by multiplying the total beta count rates by these fractional contributions, and these predicted beta count rates from the ²³⁸U and ²³²Th- series were converted to alpha dose rates by multiplication by the conversion factors of Table 2, assuming an ‘a’ value of 0.09 ± 0.01 (see Aitken, 1985, appendix K).

Table 3

gamma spectrometry analyses of mixtures of 14 potshards from each layer, and the fractional contribution from each nuclide to beta count rate factors.

Table 4

Derived beta and alpha dose rates using the observed beta count rate from each shard. The gamma dose derived from gamma spectrometry analyses of sediment samples from each layer is given at the end of the table.

A summary of the internal alpha and beta dose rates for the potsherds is given in Table 4.

Water content: to estimate the water attenuation factors, measurements of the field and saturated water contents were undertaken for all soil (for gamma dose rates) and shard (for alpha and beta dose rates) samples. Field water contents were chosen as most closely representing the long-term burial average. Water content corrections used the equations given by Aitken (1985)

The cosmic dose rate contribution was evaluated following Prescott and Stephan (1982) and Prescott and Hutton (1994) taking into account the depth of the samples, and combined with the total external dose for each layer to give the total dose rate (Table 5).

Table 5

Summary of ages and related data for 52 potshards. The samples rejected because of low luminescence intensity (test dose response minus background, T-Bg, < 1000) are identified by italics, and the single underlined sample was rejected because of poor recycling ratio. The 3 outliers discussed in the text are highlighted in grey.

The majority of the samples provided few sand-sized grains of K-feldspar and quartz; in the few potsherds that yielded quartz; the quantity was insufficient to prepare enough aliquots for efficient equivalent dose measurement. As a result, all further studies were undertaken using poly-mineral fine grains.

In total, fifteen potsherds from deposit L10.6, nineteen from L15.1, and eighteen potsherds from L15.2 were used for estimation of equivalent dose (D_e). At least eight aliquots were prepared for each potshard.

A modified single aliquot regenerative dose SAR protocol for K-feldspars (Buylaert et al., 2009) was used to estimate the D_e; the preheat temperature before any IR stimulation was fixed at 320°C and held for 60 seconds before cooling and subsequent IR stimulation. Two stimulations were used: IR at 50°C for 100 s to remove any unstable signal followed by post IR- IRSL stimulation at 290°C for 100 s. The same parameters were used for measurement of signals induced by the test dose. An illumination step using the IR source for 100 s at 325°C was used at the end of each SAR cycle to reduce the charge transfer after repeated thermal and stimulation treatment (Murray and Wintle, 2003).

The observed pIRIR₂₉₀ luminescence signals were relatively weak for most of the samples (Fig. 3) but still stronger than the corresponding IR₅₀ signals. Dose response curves were constructed using integration of the first 20 seconds of the stimulation curves, after subtracting a background based on the final 80 seconds.

Figure 3

The performance of the measured pIRIR₂₉₀ signal was tested by means of the recycling and recuperation ratios; the average of both values was taken only for aliquots which gave an error <30%. For samples used in age calculations (see section 5 - Equivalent Dose D_e), 69% of the measured aliquots were accepted following this criterion, and the recycling ratio average is 1.07 ± 0.02 (n = 240). For the average recuperation; 46% of the measured aliquots was used to estimate the average (7.0 ± 0.3% of natural signal, n= 161). Both the recycling and recuperation results are considered acceptable.

The reliability of the dose estimates was first investigated by comparing the pIRIR₂₉₀ D_e values integrated after first IR stimulation at various temperatures; samples L10.6.12, L15.1.4 and L15.2.10 were chosen for the measurements. Eighteen aliquots were prepared from each sample, and our SAR protocol applied to groups of 3 aliquots with the first IR stimulation temperature ranging from 50°C to 250°C in 40°C steps. The results are summarised in Fig. 4, and the equivalent dose is considered independent of first stimulation temperature in the temperature region 50 to 210°C, for all three samples. The same variables were applied to the test dose which chosen to be 70% of the expected natural dose (Fig. 4). The ratio between the equivalent doses from the IR signal and pIRIR₂₉₀ varied between the three samples. For sample L10.6.12 both equivalent doses were similar (mean ratio 0.995 ± 0.042, n = 18). On the other hand, for sample L15.1.4 the D_e ratio of first IR to pIRIR₂₉₀ was 0.89 ± 0.04 (n= 18), and the last sample L15.2.10 gave a ratio of only 0.68 ± 0.04 (n=18). The estimates of equivalent doses based on the pIRIR₂₉₀ appear to be more reproducible for first stimulation temperatures in the region 50–170°C. We chose to use a first stimulation temperature of 50°C in all subsequent measurements because this temperature has been used by others in pIRIR protocols (Buylaert et al., 2012; Thiel et al., 2011) and because almost all previous IRSL measurements on pottery have been made at or close to room temperature.

Figure 4

The estimation of equivalent doses was also tested for dependence on preheat temperature. Twenty one aliquots of each of the samples L10.6.12, L15.1.8 and L15.2.13 were prepared; for each sample aliquots were divided into 7 groups of 3 aliquots, and each group was measured using our SAR protocol but with different preheat temperature ranging between 100 and 310°C, increasing by 35°C in each step. The first IR stimulation was fixed at 50°C and the pIRIR stimulation temperature was kept 30°C less than the preheat temperature. The same conditions were fixed for the test dose signal measurement (Fig. 5).

Figure 5

The equivalent dose does not vary significantly over the entire temperature range for sample L10.6.12, for both IR₅₀ and pIRIR signals. For sample L15.2.13, the equivalent doses obtained from the IR₅₀ signal appears not to change significantly over the entire range, but the pIRIR signals are only independent of temperature between 205 and 310°C. On the other hand, sample L15.1.8 behaves differently for the IR₅₀ signals — the IR₅₀ D_e increases steadily with temperature, although again the pIRIR signals are independent of temperature between 205 and 310°C.

Based on the results of Figs. 4 and 5, we chose to continue measurements using a 320°C preheat and a pIRIR stimulation temperature of 290°C.

Finally the overall performance of the chosen measurement protocol was tested using a dose recovery test. Murray (1996) first tested the ability of a regenerative protocol to measure a dose applied before any thermal treatment by giving doses in addition to the natural dose using ⁶⁰Co gamma irradiation to test his modified SARA protocol. Later Murray and Roberts (1997) exposed different aliquots to daylight prior to dosing. This was justified as an attempt to imitate the natural signal bleaching process for their samples; the approach was later adopted by others (Wallinga et al., 2000). Because the luminescence signals of our samples were initially reset by heat rather than light, we used three approaches to dose recovery tests.

In the first, 3 different samples (L10.6.1, L15.1.3 and L15.2.2) were heated at 500°C for 20 minutes. Aliquots then were prepared and irradiated with about 21 Gy. The ratio of the measured dose to the given was 1.23 ± 0.03 (pIRIR; Fig. 6a.1) and 1.07 ± 0.03 (IR₅₀; Fig. 6a.2).

Figure 6

In the second test, 18 aliquots were prepared from each of the same samples and were exposed to light using solar stimulator for 4 hours before giving a laboratory dose of 21.3 Gy. For the same set of samples a further 18 aliquots were prepared and exposed to light for 4 hours, the residual dose was then measured and subtracted from the average dose for each sample from the first set of measurements. After residual subtraction the dose recovery ratio for the pIRIR₂₉₀ signal was 1.02 ± 0.08 (n = 18; see Fig. 6b).

In the third test, samples were given an additional large beta dose in addition to the natural dose. Thirty two aliquots were prepared from 3 different samples for which the D_e had been estimated earlier, ∼150 Gy additional beta dose was given to all aliquots before the total dose (given + D_e) values were estimated. For the pIRIR₂₉₀ signal, the ratio between the measured and the given dose (after subtracting the measured natural dose from the total measured dose) was in the range 0.75 to 1.35, with average ratio of 0.99 ± 0.04 (n = 32). For the same measurement; the ratio of the measured IR₅₀ dose to the given dose was 1.04 ± 0.03 (n = 29), both ratios are indistinguishable from unity (Figs. 6c.1 and 6c.2).

We conclude that optical bleaching is unable to completely reset the IR signals, but nevertheless we are able to measure a dose accurately after residual subtraction. In addition a dose given in addition to the natural dose was also measured accurately. This latter experiment confirms the ability of our chosen protocol to measure a dose given to our samples before any laboratory optical or thermal treatment.

The integration of the signal was for the first 20 seconds of stimulation, and the background was integrated over the last 80 seconds for both IR₅₀ and pIRIR₂₉₀ signals. One sample (L15.2.13) was rejected because of a poor recycling ratio. In addition, the signals from many shards were very weak; we have chosen to reject all samples for which the average pIRIR₂₉₀ summed response to the natural test dose was <1000 counts. This resulted in the rejection of 13 shards out of a total of 52 (Table 5). Thomsen et al (2011) have shown that the IR₅₀ signal is significantly less thermally stable than the pIRIR₂₉₀ signals. Thus if a shard was not heated to a high enough temperature to fully reset the more stable pIRIR₂₉₀ signal, this should be revealed by an unusually low IR₅₀/pIRIR₂₉₀ ratio. However the IR₅₀ signals were even weaker than the pIRIR₂₉₀ signals, and only 3 shards passed the intensity rejection criterion of 1000 counts when the IR₅₀ signals were considered. Even relaxing this criterion to accept IR₅₀ signals >500 counts only accepted 14 shards (out of 52). Fig. 7shows a plot of these few accepted IR₅₀ equivalent doses against the corresponding pIRIR₂₉₀ equivalent doses and it can be seen that all are consistent with a straight line of slope 0.72 passing through the origin. There is no evidence here for inadequately heated outliers (these would plot significantly below the line) and so we assume that all shards were heated sufficiently to fully empty the IRSL signals.

Figure 7

Although it can be safely assumed that the pIRIR₂₉₀ signal does not fade significantly (Buylaert et al., 2011; Thomsen et al., 2011), especially at such low doses, we have nevertheless measured fading rates for 3 aliquots each, of two samples, one from layer L15.1 and one from L15.2. The average g_2days value measured over 12 hours storage was –1.6 ± 0.4 % per decade (n = 6; data not shown); there is no evidence here for any significant loss of signal confirming our expectations that no fading correction is necessary.

Age estimates have been derived from the equivalent doses and the total dose rates for a total of 52 samples (Table 5). In Table 5 we list ages with both random and total uncertainties in ages. Random uncertainties arise solely from counting statistics calculated, based on Duller (2007). The total uncertainties include systematic uncertainties arising from external gamma dose rates calculated using gamma spectrometry (common to all shards from a given layer), instrumental uncertainty for Risø OSL reader, beta source calibration (± 2%), water content (± 25%) and Risø Low level beta counter calibration (shared by all samples from the three layers).

We expect, based on archaeological evidence; that the majority of the shards within a layer should be within the same broad age grouping. However it is clear from Table 5 that the random uncertainties in the individual shard ages do not explain the dispersion of the ages from each layer. There are three samples out of the 38 accepted results that are markedly older in age than the remainder (highlighted in Table 5). Two of these outliers were included in the data set of Fig. 7; there is no evidence there for insufficient heating as an explanation for the unusually large equivalent doses of these two samples. In any case, even if these outliers are rejected, the ages within each layer remain over-dispersed. We conclude that our calculated random uncertainties underestimate the true variability in our analyses.

The top half of Table 6 summarise the average age for each layer with the associated standard error, the over dispersion (the degree to which the random uncertainties underestimate the true dispersion) and the total error. The lower part of Table 6 repeats this information but excluding the three outliers discussed above and highlighted in Table 5. This information is repeated for the site average age of 2.85 ± 0.12 ka at the bottom of Table 6.

Table 6

Summary of average ages of each layer and the site average age.