Radiocarbon calibration and Late Glacial occupation in northwest Europe.

BLOCKLEY, S.P.E ; DONAHUE, R.E. ; POLLARD, A.M. 等

Various methods of analysing the dating of the late Glacial suggest various interpretations. Here, in answer to a paper from 1997, radiocarbon dates are calibrated and used to reconsider the dating of this contentious period.

Key-words: northwest Europe, Late Glacial, radiocarbon calibration, population movement, climatic fluctuation

Introduction

In a recent paper Housley et al. (1997) presented 127 AMS and 14 conventional radiocarbon estimates from Magdalenian, Hamburgian and Creswellian contexts, in eight areas of Europe (FIGURE 1a). The dates were predominantly on bone, most of which had been humanly-modified. Using these dates they attempted to model movements of people in Europe during the last deglaciation. They argued that much of northwest Europe was abandoned around the time of the Last Glacial Maximum, with populations dwelling in refugia. They proposed a re-colonization of northwestern Europe constrained by the retreat of the ice, and outlined a chronological sequence for this process. A `moving sum' was used to count the dates (Housley et al. 1997: 44). They assumed that this method implicitly accounted for the 1 sigma errors on the uncalibrated dates (since the `bin width' was chosen to be roughly the same as the average 1[Sigma] error), and therefore allowed the treatment of the data as point estimates. The method produced a series of histograms (FIGURE 1a) for the areas of Europe, which were interpreted as supporting a model of population movement. The earliest occupied `bin' on each histogram was taken as identifying the initiation of colonization (`pioneer phase'); the mode of the histogram was interpreted as a `residential phase', when populations were fully established.

[FIGURE 1a ILLUSTRATION OMITTED]

There are, however, some difficulties with this approach, centring around two key areas. Firstly, it may be argued that whilst the moving-sum method may take into account the 1[Sigma] errors on the uncalibrated dates, it does not account for errors at 2[Sigma] (required for 95% confidence). A second difficulty is that it is totally based on uncalibrated radiocarbon dates. Because the [.sup.14]C timescale is incorrect and nonlinear, the true chronological relationships between dates and between groups of dates are not known. It is therefore not possible to use uncalibrated [.sup.14]C age estimates as a linear relative chronology. In this case apparent chronological differences between groups of uncalibrated dates from European regions are used incorrectly to imply population movement (Housley et al. 1997: 43).Thirdly, the definition of the geopolitical regions used in the study is not necessarily relevant to Late Glacial geography. Finally, the dating programme was based around bone which was, where possible, humanly modified (Housley et al. 1997). The archaeological logic behind this is sound, but it leaves open the problem that some bone can be notoriously difficult to date (e.g. Pollard & Heron 1996: 288-90; Taylor et al. 1996).

Here we use a stepwise approach to the reanalysis of the [.sup.14]C dates in the original paper. Firstly, we look at the effects of applying the 2[Sigma] radiocarbon errors to the moving-sum method used by Housley et al. (1997). Secondly, we calibrate the dates using the curve published in 1993 (Bard et al.) which was available when the paper of Housley et al. was written, in order to demonstrate that the original conclusions could have been shown to be unsound at the time. Thirdly, we examine the effect, if any, of using the latest calibration curve published in Radiocarbon 40/3 (Stuiver & van der Plicht 1998). By calibrating, it becomes possible to compare the radiocarbon dates in this study with ice-core temperature curves (e.g. Alley et al. 1993), which are independent of the radiocarbon timescale. This allows us to test the relationship between climate and population movement against high-resolution climatic data. Finally, we consider the effect that sample selection may have on the chronological pattern of Late Glacial [.sup.14]C dates in northwest Europe.

Moving sum distributions as measures of population movement

The moving sum method (FIGURE 1a) produces a histogram of the dates which consist of a series of `bins', where the value in a bin is determined by the number of dates that fall into it. The moving-sum distributions produced by Housley et al. (1997) appear to be interpreted as indicating `pioneer phases' (the first bin) and `residential phases' (the modal bin). From this, a map of pioneer phases across northwest Europe was produced, which displayed the central values of the earliest `bins', and taken to imply a movement of pioneers re-colonizing a deserted north European landscape (Housley et al. 1997: 46). It is our contention that this is an oversimplification, and may not be statistically valid. Because the errors on uncalibrated radiocarbon dates are assumed to have a Gaussian distribution, a 1[Sigma] range only encompasses 68% of the possible values. Working to 2[Sigma], 95% confidence, is a more realistic assessment of the uncalibrated ranges for the dates in this study. We have re-drawn the original moving sum distribution of uncalibrated dates taking into account the 2[Sigma] errors (FIGURE 1b). From this it is immediately apparent that the original map using the central point of the earliest bin (Housely et al. 1997: 46) is not an accurate representation of the chronology of the proposed re-colonization phases. It is also evident that, in many of the cases, the 2[Sigma] ranges of the earliest dates could cover a period of up to 600 years. This level of uncertainty suggests that the proposed `residential' phases of Housley et al. (1997) are at best poorly defined. FIGURE 1b demonstrates the difficulty in assigning any real population movement using data of this resolution. If the earliest date is taken to represent a specific pioneer phase then, in effect, population movement is being constructed from one, or at most two, uncalibrated dates in each region, which is statistically unsound.

[FIGURE 1b ILLUSTRATION OMITTED]

Radiocarbon calibration in the Late Glacial

Since 1993 it has been possible to calibrate [.sup.14]C dates in the Late Glacial (Bard et al. 1993) using data derived from paired uranium series and AMS determinations on Barbados corals. This produces a marine curve which requires a 400-year offset to be applied for calibration of terrestrial samples, due to the amounts of old carbon in the deep oceans (the marine reservoir effect). It is highly unlikely that the marine reservoir effect has remained constant for the whole of the Late Glacial, due to the dumping of `old' carbon into the ocean during times of interstadial deglaciation. The 1993 curve has, nevertheless, been shown to provide a more realistic assessment of chronological relationships than uncalibrated dates (e.g. Batt & Pollard 1996; Taylor et al. 1996). The dates used in this study have initially been calibrated with the 1993 curve using the program OXCAL (Bronk Ramsey 1999). Once dates have been calibrated they can no longer be expressed as a point date with a Gaussian error, because the probability distribution of the date is a function of the shape of the calibration curve. This means that, once calibrated, dates must be expressed as a range. FIGURE 2 shows the 2[Sigma] calibrated ranges for the dates defined as the pioneer phases in Housley et al. (1997). This figure demonstrates convincingly that, once the dates are calibrated, a simple movement of `pioneers' is difficult to accept.

[FIGURE 2 ILLUSTRATION OMITTED]

As mentioned, the probability distribution of a calibrated date is partly a function of the shape of the calibration curve. Because of this, a moving sum which gives no weight to the actual probability distributions of dates is unlikely to be a good assessment of their true distribution. It is more appropriate to look at the summed probability distributions of the calibrated dates in this study. Summed probability distributions give a best estimate of the chronological distribution of events (Bronk Ramsey 1999). FIGURE 3 shows the summed probability distributions for seven of the eight regions in the study (the dates for Denmark and north Germany were not reproduced in the original paper). These distributions indicate that it is not possible to separate the onset of occupation in the defined study areas. Moreover, it becomes apparent that, in most cases, it is not appropriate to infer separate `pioneer' and `residential' phases; many of the calibrated probability distributions, for each region, do not demonstrate a greater concentration of dates later on in the sequence, and most of the dates for `pioneer' phases overlap with those of the `residential' phases.

[FIGURE 3 ILLUSTRATION OMITTED]

It has been argued in the past that the Late Glacial section of the 1993 curve is not yet adequate to use archaeologically. It has now been updated by the addition of data from marine varves (Stuiver & van der Plicht 1998). The improved chronological resolution provided has altered the shape of the curve, which now features sinusoidal fluctuations. FIGURE 4 is the summed probability distributions for the study areas using the new curve. Again these distributions confirm that population movement is difficult to argue, and there is no clear evidence for a 'pioneer' and 'residential' phase. All of this shows that, when 2[Sigma] errors and calibration are taken into account, there is neither evidence for a phased re-colonization of northwest Europe, nor for any identifiable 'pioneer' and 'residential' phases.

[FIGURE 4 ILLUSTRATION OMITTED]

Dating evidence and site distribution

From FIGURE 2 it is only possible to conclude that there might be a difference between the onset of the Magdalenian in the Upper Rhine and the Creswellian in the British Isles. This illustrates two key points when studying date distributions between regions. Firstly, the majority of what is now the British isles was under ice. There was, however, much land which is now under the English Channel and North Sea, and therefore unavailable for archaeological sampling. This is not the case with the Upper Rhine, where all of the region occupied in the Late Glacial is available for modern sampling. If the number of potential sites is greater, then the amount of potential data is also larger. Each site will have its own earliest date and sampling theory suggests that the greater the number of sites sampled, the more likely it is that an earlier date will be obtained. Of course, the larger the sample size, the greater the probability of selecting an `earlier site'. This means that as our samples increase through further archaeological research, the date for the earliest occupation is likely to become earlier.

Another point regarding sampling is that the choice of particular samples for dating may bias the study. There are sound archaeological reasons for choosing to sample only humanly modified bone. That is to say, the event being dated, the modification of bone, is clearly evidence of human presence. This strategy does, however, bias the study towards sites where bone is most likely to be present. There is a bias against sites where preservation conditions are poor, or where bone is likely to be disturbed. This means that areas with fewer cave sites and more open air sites may be under-represented in the data.

Sampling theory and the earliest dates in a region

As discussed above, it is statistically unsound to use the earliest date from a site or region as an estimate of the earliest human occupation, especially when the total number of dates available is small. If we assume that, in a defined region, a number of sites representing the earliest occupation have been sampled, then we can use a combination of the earliest dates from each site to estimate the earliest human occupation. More robustly, we can simply use some combination of all the dates from each site to estimate the earliest date for human occupation of the region. The first approach would be more valid if there was some alternative (e.g. stratigraphic) evidence to suggest that the earliest date from each site was more representative of the earliest occupation than other dates form the site. If all dates are deemed to be equally valid (for whatever reason), then it is appropriate to use the second approach.

The issue is complicated by the fact that uncalibrated radiocarbon dates are assumed to have a normal distribution, whereas calibrated dates in general do not. It is, however, valid to use normal statistics to combine calibrated radiocarbon dates, providing they can be assumed to represent multiple estimates of the same phenomenon, since the Central Limit Theorem in elementary statistics states that the distribution of means of non-normally distributed estimates will in itself tend towards a normal distribution (Fletcher & Lock 1991: 67-8). We can therefore use the means of each calibrated date distribution in order to estimate the 95% confidence interval of the earliest date for that region using the usual Student's-t statistic for small numbers. We can estimate the means of each calibrated range if the individual probability distributions of the dates are unimodal and approximately symmetrical, as is the case with the dates in this study when calibrated using the 1993 curve.

The equation for relating the mean of a sample to the 95% confidence interval of the `true' mean ([Mu]) is:

[Mu] = [bar]x [+ or -] ts/[square root of n]

where [bar]x is the mean calculated from n estimates and s is the standard deviation of that mean. The parameter t is the 95% confidence value of the t distribution with n-1 degrees of freedom, and is derived from standard tables (e.g. Fletcher & Lock 1991: 180).

To illustrate the procedure, we take the dates from the three sites (Kniegrotte, Olknitz and Teufelsbrucke) in the Thuringian region as published by Housley et al. (1997: table 1: 29-30). We present two ways to combine these dates. Firstly, we may take the oldest date from each site and combine them to give us an estimate of the oldest date for the region. Secondly, we can combine all the dates from all the sites to give us an estimate of the earliest date of occupation in the region (FIGURE 5). From this we see that combining the three oldest dates gives us a very broad confidence interval (approximately 17,000-14,200 BP) which corresponds approximately to the age range obtained by summing all the calibrated age ranges using OXCAL. This is narrowed considerably if all dates are combined (15,300-14,750 BP). This illustrates that, somewhat counter-intuitively, it is better to use as many dates as possible to reconstruct these estimates, rather than trying to isolate only the earliest dates, because of the properties of the Student's-t distribution. Clearly this also strongly reinforces the idea that it is inappropriate to use a single date.

[FIGURE 5 ILLUSTRATION OMITTED]

Using this procedure, we have taken all of the dates used in the moving sum for seven of the eight regions defined by Housley et al. (1997) to estimate the 95% confidence interval for the early Magdalenian (or equivalent) occupation in each region. The results are shown in FIGURE 6, and are discussed below.

[FIGURE 6 ILLUSTRATION OMITTED]

Late Glacial climate change and human occupation

Until recently most models for climate change in Europe were derived from pollen data (e.g. Mangerud et al. 1974), with the chronology being derived from radiocarbon. The publication of the Greenland GISP-2 ice-core data (Alley et al. 1993) has provided us with climatic information that is both independent of radiocarbon dating and which has a much higher chronological resolution, having a 25-year running mean. Late Glacial temperature curves are constructed from the oxygen isotope ratios within the ice and the chronology is provided by the annual laminations. The curve indicates that until around 14,700 years BP the Late Glacial climate was in a stadial (Oldest Dryas), with no discernible climatic amelioration. At this point the Late Glacial interstadial began with a sudden sharp peak; the rapid nature of this Late Glacial climate change might be expected to have important implications for archaeologists, if it is manifest in northwest Europe. Using the `pioneer phase' dates produced by Housley et al. (1997), and calibrating them to allow comparison with the ice core data, we see that all of these `pioneer' phases fall before this crucial event. However, using the 95% confidence estimate for the early occupation in each region as defined above (FIGURE 6) we see that they now cluster almost exactly around the time of the Late Glacial Interstadial peak in the GISP-2 data. It is possible to argue that there may be some link.

Conclusion

We have shown that it is not possible, when errors and calibration are fully taken into account, to demonstrate the movement of peoples across Europe during the last deglaciation. It has also been demonstrated that, when dates are calibrated and their probability distributions are summed, separation of early Magdalenian dates into `pioneer' and `residential' phases on radiocarbon evidence alone is inappropriate. It is also clear that incorporation of climatic information has an important impact for Late Glacial archaeology, but radiocarbon dates must be calibrated in order to compare with ice core temperature curves. When this is done using the same data analysed by Housley et al., it becomes clear that there is a remarkable synchronicity between the floruit of the Magdalenian cultures and the rapid onset of climatic amelioration.

We have also argued that the pattern of dates seen in Europe during the Late Glacial may be influenced by site preservation and sampling strategy, and that we are as yet far from having a detailed understanding of human settlement in the Late Glacial of north west Europe. It may well be that the sudden temperature rise at 14,700 BP allows a change in settlement patterns. The pattern of site use in the cold phases may alter with interstadial conditions; not in fact a northern migration at all, but the increased use of sites which may yield humanly modified bone.

References

ALLEY, R.B., D.A. MEESE, C.A. SHUMAN, A.J. GOW, K.C. TAYLOR, P.M. GROOTES, J.W.C. WHITE, M. RAM, E.D. WADDINGTON, P.A. MAYEWSKI & G.A. ZIELINSKI. 1993. Abrupt increase in Greenland snow accumulation at the end of the Younger Dryas event, Nature 362: 527-9.

BARD, E., M. ARNOLD, R.G. FAIRBANKS & B. HAMELIN. 1993. [.sup.230]Th-[.sup.234]U and [.sup.14]C ages obtained by mass spectrometry on corals, Radiocarbon 35: 191-200.

BATT, C.M. & A.M. POLLARD. 1996. Radiocarbon calibration and the peopling of North America, in M.V. Orna (ed.), Archaeological chemistry: organic, inorganic, and biochemical analysis: 415-33. Washington (DC): American Chemical Society.

BRONK RAMSEY, C. 1999. OxCal: radiocarbon calibration and stratigraphic analysis program. Oxford: Oxford University Research Laboratory for Archaeology.

FLETCHER, M. & G. LOCK. 1991. Digging numbers: elementary statistics for archaeologists. Oxford: Oxford University Committee for Archaeology.

HOUSLEY, R.A. , C.S. GAMBLE, M. STREET & P. PETTITT. 1997. Radiocarbon evidence for the Lateglacial human recolonisation of northern Europe. Proceedings of the Prehistoric Society 63: 25-54.

MANGERUD, J., S. ANDERSEN., B.E. BERGLUND & J.J. DONNER. 1974. Quaternary stratigraphy of Norden, a proposal for terminology and classification, Boreas 3: 109-28.

POLLARD, A.M. & C. HERON. 1996. Archaeological chemistry. Cambridge: Royal Society of Chemistry.

STUIVER, M., A. LONG & R.S. KRA (ed.). 1993. Calibration issue, Radiocarbon 35.1.

STUIVER, M. & P. REIMER. 1998. CALIB. Washington (DC): Quaternary Isotope Laboratory, University of Washington.

STUIVER, M. & J. VAN DER PLICHT. 1998. Intercal 98, Radiocarbon 40/3.

TAYLOR, R.E. , M. STUIVER & P.J. REIMER. 1996. Development and extension of the calibration of the radiocarbon time scale: archaeological applications. Quaternary Science Reviews 15: 655-68.

S.P.E. BLOCKLEY, R.E. DONAHUE & A.M. POLLARD, Department of Archaeological Sciences, University of Bradford, Bradford BD7 1DP, England.

Received 3 February 1999, accepted 2 November 1999.