Cap a una poesia del fer

The previous sub-section argued that there is a relationship between measurement precision, calibration curve precision and magnitude of offset to be expected in a dating model. However, there are different kinds of offsets, each of which requires a different management strategy. While not pertinent to the problem at hand, they are relevant to the application of wiggle-match dating in places with different seasonal growth patterns and geographic separation from the origin of the calibration data.

Three provisional kinds of offsets can be defined: analytical, geographic and archive-dependent.

1. Analytical offsets are introduced by measurement errors and statistical effects of the calibration curve construction algorithms, such as the autocorrelation in the T-947 decadal averages. They can be resolved through increasing curve precision.

Another form of analytical offset would come from oversmoothing the data by the calibration curve algorithm. As the IntCal algorithm makes its estimates of the calibration curve based on a large number of data surrounding any point of interest (see Chapter 3), there does exist the possibility that, in points of sudden increases in short-term variability, oversmoothing will take place. Indeed, one of the archaeological case studies discussed in Chapter 6 herein may be affected by such offset (see section 6.5.1).

2. Geographic offsets emerge due to a persistent presence of a radiocarbon source or sink pulling the atmospheric concentration at the sampling location away from the global average. In this category inter-hemispheric offsets are of widest importance (Hogg et al. 2009, 2013), albeit in some cases localized offsets can emerge, for example through proximity of vents emitting geologic CO₂ (Capano et al. 2013).

Geographic offsets will also include transient offsets, such as the intermittent shifts of atmospheric radiocarbon to the southern hemisphere trend in Japan (Imamura et al. 2007; Nakamura et al. 2013), given that they persist over the time-scale of the analysis. They can be resolved either through calibration against local archives, or through modelling the causal processes and updating the curve. Both these strategies have been applied in developing the SHCal curve for the southern hemisphere (Hogg et al. 2013).

3. Archive dependent offsets stem from the nature of the samples themselves. They emerge because different archives develop through different mechanisms, drawing upon different carbon reservoirs and being subject to different mixing rates. The resolution of these offsets can be achieved either through understanding and mod-elling the effects of their underpinning mechanisms (this procedure was applied in the development of the Holocene marine calibration curve; Reimer et al. 2013), or through taking paired measurements of samples whose carbon derives from dif-ferent reservoirs (e.g. dating terrestrial animal bone points embedded in remains of humans who had a significant aquatic diet component to difference between atmospheric and aquatic reservoirs in a given location (Cook et al. 2001).

The key difference between analytical, geographic and archive-dependent offsets is that analytical offsets are a function of our technical capacity to produce accurate and precise data and develop appropriate statistical models for it. In other words, with a greater number of more precise calibration measurements and algorithms that limit the artefacts (such as oversmoothing) in the calibration curve, the number of instances in which analytical offsets would have a noticeable effect on the interpretation of data will diminish. Geographic and archive-dependent offsets are objective: they are inherent in

the objects studied and their effects become greater as the precision of the calibration curve improves.

Of the two objective offsets, the archive-dependent ones pose the greatest challenge to calibration and force a limit on how far we can improve a wiggle-match or any dating model for that matter. A simple example of an archive dependent offset is dating animal tissues. We know from the human body regenerative map that different tissues have different carbon turnover times (Salephour et al. 2013): blood will be equilibrated with the diet, bone collagen will have several years of delay relative to the diet and neurones, with the exception of a small proportion of cells in the hippocampus, will date the birth of an individual. Hence, the animal organism will consist of a set of different archives with different carbon turnover rates, some of which will no longer equilibrate with the atmosphere. Besides some of the more speculative implications, such as the potential to wiggle-match human remains if sufficient tissue diversity has been preserved, this is important in understanding the resultant models. For example, when building a site model using animal bones, it can be expected that the parameters will be offset by several years. As long as parameter HPD areas are wide, this does little to affect interpretation. However, as the HPD areas are reduced to several decades or less, the offset introduced by the different cycling rates, will have an increasing effect on interpretation.

As the precision of the calibration curve improves, archive specific offsets may become the greatest challenge to wiggle-match dating due to variability in growth seasons.

From observation we know that the radiocarbon content of the atmosphere may be subject to seasonal fluctuations, with amplitudes of up to 7h∆¹⁴C (Currie et al. 2011;

Graven et al. 2012a,b; Levin et al. 2013, 2010). This is driven by a number of fac-tors including, but not exclusive to, the oceanic carbon exchange rates (Siegenthaler and Sarmieto 1993), circulation of air from the tropics into the extratropical lower stratosphere and towards the poles (Andrews et al. 2001), or the down-welling of high-altitude air from the stratospheric overworld (Boering et al. 1996; Holton et al. 1995), all of which are coupled with climate in different ways. Hence, the inter-annual changes of radiocarbon discussed so far are supplemented by an intra-annual variability (Fig-ure 5.29). As different tree species will have different growth rates and seasons, and are subject to different environmental stressors (Vaganov et al. 2006), the section of intra-annual variability recorded in tree rings will be dependent on all of the factors mentioned above (Figure 5.30). Until now this has not been a challenge as variability even as high as 0.5h¹⁴C (4¹⁴C years BP) would have been lost in the uncertainties of measurements and the calibration curve. However, the recent results from the eastern Mediterranean demonstrate that we are already capable of detecting archive-specific offsets in plants (see Chapter 3.2.4). Given that some laboratories now consider ana-lytical uncertainty of less than 20 ¹⁴C years to be routine (Szidat et al. 2014), we can expect the effects of these archive specific offsets to become more visible in the coming years.

All of these offsets create a Red Queen-like paradox for wiggle-match dating. As the analytical offsets caused by curve uncertainties are resolved with improvements in the calibration programme, the part played by objective offsets will increase; on a calibra-tion curve with a standard deviacalibra-tion of 12 ¹⁴C years, an archive-specific offset of 3¹⁴C years will be a lesser issue than on a curve with a standard deviation of 6 ¹⁴C years.

Hence, as the technical capacity to identify the individual Schwabe cycles develops, wiggle-matches will be at risk of producing more biased results due to increasing input of divergences between different archives. So, to take full advantage of the increases in analytical capacity (understood as sample throughput and measurement precision), the calibration curve will need to be intensified and sourced from more diverse loca-tions. But, if the increase in analytical capacity required for calibration improvement is reached, it can also be used for measurements on unknown samples. However, as these will match the precision of the underlying curve, they will become sensitive to any of the underpinning offsets, both objective and analytical. Therefore, taking full advan-tage of the analytical capacity would then require another extension of the calibration programme that would be possible only through increasing analytical capacity to tackle the greater demands on calibration. But once this increase is achieved, the situation resumes. Hence, wiggle-match dating research design has to make a conscious effort to maintain the accuracy of the results, which can be lost if measurement frequency and precision are increased beyond what is feasible under a given calibration curve. In other words, it is risky to conduct wiggle-matches to the highest possible precision and measurement frequency, as the resulting modelled date ranges may be inaccurate.

Figure 5.29: Schematic representation of the intra-annual radiocarbon variability (continuous line) over an 11-year solar cycle (dashed line). If an archive is deposited on a seasonal basis, as is the case for tree-ring growth in the temperate zone, there exists a chance that it will overestimate or underestimate the decadal average of a given year on a persistent basis.

Figure 5.30: Schematic representation of the effects of Intra-annual radiocarbon variability on comparing radiocarbon determinations. Samples collected from archives with different deposition seasons, for example because of differences between tree species, will contain records of different atmospheric radiocarbon concentrations.

If the difference between these concentrations is large enough relative to the measurement uncertainties, a systematic offset may develop.

5.4 Chapter conclusions

The analyses presented in the current chapter had their origin in the geometric in-tuition that single-ring data are best suited to wiggle-match dating short tree ring sequences. Hence, 50 consecutive rings of a known-age timber T-947 were dated to evaluate whether this is true for the Scottish Iron Age. After resolving the technical difficulties relating to the measurements themselves, it became clear that the single-ring data contains more short-term variability than what is included in the calibration curve based upon decadal and bi-decadal blocks of wood. These results demonstrated that, as far as archaeological applications of wiggle-match dating are concerned, there is no clear gain in increasing the sampling resolution to more than that of the underlying calibration data (hence decadal sampling is optimal for the Scottish Iron Age). What also became clear, however, is that for several decades in the early fifth century cal BC the mean of the calibration curve is lower than the past trend of radiocarbon, leading to the risk of small-scale offsets in the results. These biases do not, however, preclude improving the performance of wiggle-match dating: they only point out that there is a point beyond which precision will be gained at the cost of accuracy. Therefore, there ought to exist a sample space of measurement frequencies and analytical precisions within which offset magnitude is negligible. This was illustrated in Figure 5.27 where wiggle-matches constructed from simulations of routine precision measurements have covered the target date within their calibrated date ranges, even though some of the bias is retained, much the way it has been present in the decadal measurement series.

However, because the number of measurements involved is smaller, and because their precision is much lower, the measurements on the hypothetical unknowns should no longer “float” within the calibration curve, and so the extent to which they are pulled onto the mean of the calibration curve should also be lower. The resulting wiggle-matches will have lower precision, however they also ought to be more reliable. A further benefit is that these less precise wiggle-matches will also be less expensive and hence it becomes possible to date a larger number of timbers within the same budget.

This larger number of more reliable (but less precise) wiggle-match dates can then be included within Bayesian models to obtain a more holistic description of archaeological sites and, in some instances, to regain some of the lost precision. Such an approach is, of course, nothing more than a practical mitigation of a deeper problem of offsets from the calibration curve, a problem which, in the long run, ought to be addressed by improved methods of statistical analysis, better data sets, and the improvements in the understanding of production and cycling of radiocarbon, so as to develop a systematic way of dealing with archive-dependant and geographic offsets. Nevertheless, such com-prehensive solutions are far beyond the sccope of this project and the research capacity of any single archaeological funding agency and hence, at this point in time, research into the chronologies of Scottish wetland sites needs to accept that there are limits to how precise short-span wiggle-matches can be before they begin losing accuracy.

dating to Scottish wetland sites

Chapter 5 argued that wiggle-match dating can produce reliable results, even if small-scale offsets from the calibration curve are present. Nevertheless, this notion alone cannot provide suitable guidelines for the implementation of wiggle-match dating to a specific group of sites this can only be achieved through the discussion of actual case studies. The current chapter discusses six such case-studies: the analyses of legacy dates from loch Arthur and Dormans Island and the new wiggle-match dating projects at Black loch of Myrton, Cults Loch 3, Dumbuck and Erskine Crannog (Figure 6.1).

From these analyses three main conclusions emerged: 1) for most cases (but not all), wiggle-match sampling ought to take place within a feature-oriented framework; 2) research design ought to assume that complications related to timber re-use and other site-formation factors will appear and hence it should retain flexibility to allow for this fact; and 3) the relationship between site formation processes and archaeological questions is paramount and has to be reconsidered at every step of project design.

Figure 6.1: Locations of sites discussed throughout this chapter. Modified from:

https://commons.wikimedia.org/wiki/File:Scotland location map.svg.

Feature-oriented sampling is concerned first and foremost with establishing a set of modelled date ranges for an associated group of material to a precision sufficient for the archaeological question at hand. An example of such an approach would be distribut-ing the available measurements across two timbers from a sdistribut-ingle structural feature, rather than focussing them on a single log, or dispersing them to two timbers from different parts of the site. Without this procedure, timbers with unusual dates may skew interpretations of sites and create the illusion of multiple phasing.

Complexities of wetland site formation processes trigger the need for flexibility in re-search design. During the dating of both Black Loch and Cults Loch 3, the original simulation-based designs had to undergo significant alterations, either through the need to better understand the relationships between specific timbers, or through the emer-gence of new subsidiary objectives that had to be achieved before the project aims could be fulfilled. In practice this calls for iterative sampling, whereby only a part of the available dating resources is spent at any one time, as the key characteristics of the site are being explored.

Both of these concerns, feature-oriented dating and design flexibility, stem from the relationship between archaeological questions and the formation process of a given site.

In essence, it is impossible to answer a question that requires information rendered unavailable by the sampling pattern applied, the excavation methods used, or the loss of contexts through erosion and other processes. This relationship, however, also takes subtler forms; for example, there always exists a possibility that some of the timbers comprising a part of a structure have been sourced from an older building and not felled for the purpose of new construction. It is because of such instances that feature-oriented sampling is so important to a comprehensive understanding of the dating of wetland sites. At the same time, there are situations where the relevant information can be obtained without such sampling designs, or when a particular question does not require feature-level information. Hence, utmost attention needs to be given to the relationship between the postulated questions and the site formation processes to avoid the risk of falling into a routine sampling pattern that may not be optimal for resolving the archaeological problem at hand. It is also the archaeological questions that constitute what classes as good precision; for some questions date ranges of several centuries may be sufficient, while for others annual precision might prove essential.

The chapter consists of six main sections. The first section introduces the concepts out-lined above in more detail through exploring them in the legacy data from Dormans Island in Galloway and Loch Arthur in Dumfriesshire. The limited extent afforded by the legacy data permits a clearer exposition and also stresses the need for adjusting the questions asked to the data available and achievable model precision. The second section introduces aspects of wiggle-match stability and systematic bias in decayed wood, as well as shrinkage; these are the main technical issues that recurred in the wiggle-match based case studies. The third section reports on Block Loch of Myrton in Wigtonshire, Galloway, where the archaeological aims were to derive a date for the construction of the site and hence place it within the regional sequence. The case study

highlights the advantages of iterative sampling and demonstrates the need to be ready to go beyond original research design in the number of samples commissioned, as the sampling designs that work in simulation might not be sufficient to resolve ambiguities of applied research. The fourth section summarizes the wiggle-match dating project conducted at Cults Loch 3, also in Galloway. Here more ambitious aims were pur-sued, with the focus on establishing relative chronologies in a context with instances of possible timber re-use during the Hallstatt plateau. The fifth section reports on the results of the wiggle-match dating of Dumbuck and Erskine Crannog, the two ex-posed intertidal platforms in the firth of Clyde. These case studies demonstrate that off the Hallstatt plateau it becomes possible to use wiggle-match dating to not only address overall dating, but also site formation processes. The final section concludes the chapter.

6.1 Legacy dates from Loch Arthur and Dormans island

The introductory section of this chapter outlined the notions of feature-oriented sam-pling, design flexibility and the relationship between the data and the archaeological questions. The analysis of legacy dates from Loch Arthur and Dormans Island puts feature-oriented sampling and the question-data relationship into context. At Loch Arthur, the limited dating evidence, consisting of single un-stratified radiocarbon dates from a period of calibration ambiguity, as well as the lack of feature-oriented dating, meant that specific questions regarding the site formation process could not be an-swered. However, the distributed nature of the sampling justifies an assertion that the crannog mound had accreted sometime in the second half of the first millennium cal BC. At Dormans Island, where almost all of the dating effort focussed on a sin-gle structural element with several precise dendrochronological sapwood estimates, the opposite is true: a very clear picture of the chronology within this structural grouping was possible, alongside the identification of a timber that may have been re-used. At the same time, the focussed nature of the dating precluded providing answers as to the history of the site as a whole.