On the hoof: exploring the supply of animals to the Roman legionary fortress at Caerleon using strontium (⁸⁷Sr/⁸⁶Sr) isotope analysis

Provisioning the Roman army

The Romans’ success as empire builders is in a very large way explained by their capacity to concentrate large numbers of professional, well-equipped soldiers against their enemies for long periods of time. By the middle of the first century AD, the Roman Empire had grown to extend from Britain in the north to Egypt in the south and from Syria in the east to the Atlantic coast in the west. It is estimated that the army at this time comprised some 300,000 men, of whom about 40,000 invaded Britain in AD 43. Over the following decades, all of the native tribes in modern-day England and Wales were incorporated into the new province of the Roman Empire, some by force of arms and others by treaty. The invaders found the natives in Wales particularly troublesome and for some 40 years or so after their final subjugation between 73/74 and 77/78 an estimated 25,000 Roman soldiers were garrisoned in this part of western Britain (estimated to represent between 10 and 17% of the total population) (Mattingly 2006, pp. 101–16; Arnold and Davies 2000, pp. 3–24; Burnham and Davies 2010, pp. 37–48 and 137–45).

The ability to maintain a standing army on this scale and over such a wide area is one of the defining characteristics of the Roman period, and understanding exactly how the soldiers were supplied and provisioned has become an increasingly important question. For many years, the ancient literature was the only source of relevant evidence, but from the 1970s, new information regarding the consumption of animals began to come from the nascent science of zooarchaeology (Davies 1971; Labisch 1975; Dickson 1989; King 1984). As excavators realised the potential of animal bones, the analysis of faunal remains resulted in the correction of oft-repeated misconceptions (for example, that Roman soldiers were largely vegetarian), as well as significant new insights based on species and ageing data. The past 50 years have witnessed a surge in the study of animal bone assemblages from Roman military sites, and today, we have a far better appreciation of the army’s use and consumption of animals, particularly cattle, sheep/goats and pigs (King 1999; King 2001; Stallibrass and Thomas 2008; Stallibrass 2009; Maltby 2016). Remains of the three major domesticated animal species dominate the zooarchaeological record of the army and it is widely accepted that Roman soldiers enjoyed a relatively meat-rich diet, at least when compared to the majority of the population at the time. It should not be forgotten, however, that these animals also produced other foods (milk and cheese), as well as essential resources such as leather, skins and wool.

Despite the accumulation of new material and analysis, we still have only a very rudimentary understanding of the quantities of animals consumed by Roman military units, where and by whom these were reared, or how their supply was organised. The assumption that most cattle, sheep/goats and pigs would have been bred and reared locally, in part by the army itself or requisitioned from the local population, forms the basis of most recent academic discussions, whether historical or archaeological (Carreras Monfort 2002; Herz 2002; Junkelmann 1997, pp. 154–63; Kehne 2007; Mason 1988; Pearce 2002; Roth 1999, pp. 117–55; Thomas and Stallibrass 2008; Whittaker 2002). If this notion is correct, the presence of a sizeable military garrison will have had a presumably significant (though as yet unquantified) impact on the inhabitants of recently conquered territories, most especially in the countryside around the army’s forts and fortresses. Simply feeding the 5500 men in a Roman legion, for instance, must have required the ability to control agricultural resources over a wide area and to organise their redistribution from producers to consumers quickly and effectively. Yet, how the Roman authorities were able to make this happen is almost entirely unknown (though see Thomas (2008), Cavallo et al. (2008) and Groot (2008) for examples of zooarchaeological approaches to this question).

In part, this lack of understanding is because the surviving ancient sources did not generally concern themselves with the day-to-day routines of the army. Meanwhile, even though the archaeological evidence for the eating and drinking habits of Roman soldiers is varied and plentiful, until recently, there have been few attempts to integrate the study of pottery, animal bones and other environmental remains in order to address wider questions of the military diet, or where and how the cereals, vegetables, fruit and meat consumed in Roman forts and fortresses were produced (Cool 2006, p. 1). The significance of this topic has long been recognised and in 2002, Richard Reece stated: ‘The huge quantity of resources needed by the army and the impact this might have had on the local native populations seem to me to be areas which cry out for investigation’ (Reece 2002, p. 187).

The development of isotope analysis and its application to animal bone assemblages from archaeological contexts has the potential to transform the study of the provisioning of the Roman army. The analysis of isotopic ratios in bones and teeth can be used to reconstruct diet (carbon and nitrogen), geographic origin and palaeoenvironment (oxygen and strontium). Strontium in particular can provide direct evidence for non-local individuals and should prove useful for reconstructing the supply networks that supported military sites. This can in turn provide insights into the nature of the developing relationships between the Roman invaders and the populations of conquered territories.

Isotope provenancing techniques have been commonly applied to a Romano-British material in recent years, although research has principally focused on humans (Chenery et al. 2010; Chenery et al. 2011; Eckardt et al. 2009; Leach et al. 2009; Montgomery et al. 2010a; Montgomery et al. 2011; Müldner et al. 2011; Redfern et al. 2016; Shaw et al. 2016). Studies of faunal remains are very rare, and to date, only a single isotope study that addresses livestock supply in Roman Britain has been published and this dealt with a rural settlement rather than a military site. The study by Minniti et al. (2014) compared patterns in the movement of cattle to the site of Owslebury in Hampshire from the Middle Iron Age to the later Roman period. The research identified a greater diversity of results during the Roman occupation that, it was suggested, indicates an increase in the supply of livestock from a wide range of locations beyond the local area. Although the paper did not explore origins in detail, some cattle from Owslebury must have moved substantial distances to reach the site, perhaps over more than 70 km. This provides important insights into the scale and volume of livestock movement in Roman Britain and hints at the potential of an isotope study for investigating the provisioning of the Roman army.

Feeding the legion at Caerleon

The native tribes of what is now Wales were renowned for their stubborn opposition to Roman expansion, occupying much of the army in Britain from 47/8 to their final conquest in 77 or 78. The Silures of South Wales gained a reputation for their particularly implacable resistance, leading the Roman writer Tacitus to record that ‘on the Silures neither terror nor mercy had the least effect; they persisted in war and could be quelled only by legions encamped in their country’ (Annals 12, 32). Ultimately, however, the Silures and the other tribes of Wales were defeated and their lands occupied. An estimated 25,000 Roman soldiers would be garrisoned in Wales for the next 30 to 40 years before many of the troops were subsequently withdrawn to construct and man Hadrian’s Wall in northern Britain from the 120s (Burnham and Davies 2010, pp. 37–48 and 137–45).

Two full legions formed the backbone of this short-lived frontier in Wales, based on their fortresses at Isca (Caerleon) in the south and Deva (Chester) in the north. Both fortresses were strategically sited on the lower reaches of important rivers (the Usk and the Dee respectively), controlling the coastal routes into and out of Wales while also connecting the legions to the rest of Britain as well as other parts of the Roman Empire. Although both fortresses have been the focus of archaeological excavations for many years, unlike Chester, large parts of the fortress at Caerleon are free of later buildings, thus giving archaeologists more opportunities to explore the site’s history and that of its legionary garrison, the Second Augustan Legion (Legio Secunda Augusta). Mortimer Wheeler’s excavation of the amphitheatre in the 1920s was the first scientific archaeological investigation of the site, after which further excavations, often over large areas, were undertaken in and around Caerleon throughout much of the twentieth century. Recent work has provided a great deal of new information about Isca, particularly the early decades after its foundation in the mid-1970s when the fortress was probably often occupied by only a skeleton garrison, while most of the legion was occupied elsewhere building roads, forts, bridges, new towns and cities, or frontier walls (Boon 1987; Breeze 2002; Jarrett 2002; Evans 2010).

Caerleon is one of the most studied and best understood legionary fortresses anywhere in the Roman Empire, yet we have very little idea about how the Second Augustan Legion was supplied. Some foodstuffs came from other parts of the Empire (notably wines from Italy and Spanish olives and olive oil), but it has always been assumed that the daily staples must have originated locally (Davies 2004; Mason 1988). However, this is little more than educated guesswork and, in reality, where and how grain for bread and fodder was grown, or the cattle for beef, milk and leather, and the sheep for mutton and wool were reared, are questions to which we did not have the means to provide reliable answers (although reclamation of the Gwent Levels is thought to have possibly been for the pasturing of the legionary cattle herd, see Fulford 2002, pp. 85–86).

Samples

A total of 44 animals were analysed (23 cattle, 10 caprines and 11 pigs, Table 1). Given the complexities of identifying caprine dentition to precise species (Zeder and Pilaar 2010), all sheep/goat specimens have been assigned to the taxonomic category of caprine. Ideally, sampling would repeatedly target the same-sided tooth for all individuals of each taxon to ensure that all samples represent separate individuals. However, this was rarely possible due to the limited numbers of suitable samples securely assigned to a single phase (0–4). Measures were taken to reduce the chance of repeat sampling of the same individual when available specimens allowed, by assessing morphology, context, dental attrition and element side. On the balance of probability, the likelihood of repeat sampling at a legionary fortress like Caerleon is very low, but it cannot be entirely discounted.

These plants were supplemented by seven dentine specimens (paired to the CPF31–CPF37 enamel samples) analysed to characterise the local diagenetic signal of the soil in Caerleon (CPF38–CPF44, Table 2).

Methods

Strontium (⁸⁷Sr/⁸⁶Sr) isotope analysis was employed to identify animals that were not locally reared and to explore potential areas of origin. The approach has been used extensively on humans, but in recent years, methodological research (e.g. Madgwick et al. 2012; Montgomery et al. 2010b) and case studies (e.g. Bendrey et al. 2009; Towers et al. 2010; Viner et al. 2010) have demonstrated the value of the method for examining the movement of domestic animals in Britain.

The approach has been explained in detail elsewhere (e.g. Bentley 2006; Buzon et al. 2007; Price et al. 2002) and is therefore only briefly described here. Strontium isotope analysis relies on the premise that the bioavailable strontium from the region where a human or animal is sourcing their diet is incorporated into their bodily tissues (including bones and teeth) through the food chain. Strontium substitutes for calcium in the hydroxyapatite of bone and teeth (Bentley and Knipper 2005). Bone and dentine are relatively porous and have low crystallinity and are therefore not suitable for analysis, as biogenic signatures will be contaminated by the burial environment (Price et al. 1992; Trickett et al. 2003; Hoppe et al. 2003). However, dental enamel, which is used in this study, has been shown to be more resistant to diagenetic alteration and to retain biogenic strontium (Budd et al. 2000). However, it is still subject to some diagenetic alteration (Madgwick et al. 2012). Underlying geology has the greatest effect on bioavailable strontium isotope values, with the geological ⁸⁷Sr/⁸⁶Sr controlled by the rubidium (Rb)-to-strontium ratio and the age of the rock. However, precipitation also has an impact, especially in areas with high rainfall such as on the western seaboard of Britain (Evans et al. 2010). The strontium isotope value of rainfall, which derives from seawater, is 0.70918 (McArthur et al. 2001; Pearce et al. 2015), and thus falls within the most common biosphere package in Britain (Evans et al. 2010). In saturated soils, this rainfall-derived strontium will have a diluting effect on geologically derived values, complicating the interpretation of biogenic strontium isotope signals. In addition, interpretation is hindered by the relatively coarse biosphere maps that are generally available. The work by Evans et al. (2010) has greatly improved the situation in Britain, but biosphere mapping remains far from comprehensive.

The majority of analysis was undertaken at the Memorial Applied Archaeological Sciences Laboratory of Memorial University of Newfoundland (CPF01–CPF30), with seven paired enamel and dentine, and all plant samples analysed at the Bristol Isotope Group Laboratory at the University of Bristol. Enamel and dentine samples (10–50 mg) were cut from teeth and cleaned using a diamond saw and burr to remove all adhering dentine from enamel samples and at least 10 μm of the enamel surface. Sample chemistry at the Memorial University of Newfoundland and University of Bristol follows a common and well-defined methodology. Enamel samples were cleaned with high-purity (99%) acetone followed by deionised water (DI H₂O, > 17 MΩ) and dried down. Plant samples were washed in DI water, dried and crushed to tea leaf consistency following Evans et al. (2010). Samples were then weighed into clean perfluoroalkoxy (PFA) vials.

Enamel samples were dissolved in high-purity HNO₃ on a hot plate. Plant samples (0.1 g) were sequentially treated with high-purity HNO₃, HNO₃:HCl, HNO₃:H₂O₂ and HNO₃:HClO₄ on a hot plate. Strontium extraction in both laboratories used Sr-Spec™ resin (Eichrom, Horwitz et al. 1992) and followed a common procedure for strontium purification. Approximately 100 μl of pre-cleaned Sr-Spec resin is loaded into extraction columns, and samples were then loaded on to the resin in nitric acid, either 0.5 ml of 3 M HNO₃ or 1 ml of 6 M HNO₃ in the University of Bristol and Memorial University of Newfoundland, respectively. Matrix elements were then eluted in 3 column volumes of HNO₃, and purified strontium fractions were then collected in 2 column volumes of DI H₂O ready for mass spectrometry.

At the Memorial University of Newfoundland, strontium isotope ratios were measured using a Thermo Finnigan Neptune multi-collector inductively coupled mass spectrometer (MC-ICP-MS) in the CREAIT facility. All data was first corrected for on on-peak blank intensities and then mass bias corrected using the exponential law and a normalisation ratio of 8.375209 for ⁸⁸Sr/⁸⁶Sr (Nier 1938). Residual krypton (Kr) and rubidium (⁸⁷Rb) interferences were monitored and corrected for using ⁸³Kr (⁸³Kr/⁸⁴Kr = 0.20175 and ⁸³Kr/⁸⁶Kr = 0.66474; without normalisation) and ⁸⁵Rb (⁸⁵Rb/⁸⁷Rb = 2.5926), respectively. Analysis of NIST SRM 987 during the analytical session gave a ⁸⁷Sr/⁸⁶Sr value of 0.710292 ± 0.000014 (2 SD, n = 11), and all data is corrected to a NIST SRM 987 value of 0.71024 (Copeland et al. 2008). Total procedural blanks are typically < 1% of typical sample ⁸⁸Sr voltages, and measurement of ⁸⁷Sr/⁸⁶Sr in NIST SRM 1400 (bone ash) gave a corrected ⁸⁷Sr/⁸⁶Sr value of 0.71310, which is identical to published values for this material (Galler et al. 2007).

At the University of Bristol, strontium isotope ratios were measured on a Thermo Finnigan Triton thermal ionisation mass spectrometer (TIMS). Samples were loaded on to rhenium filaments with a TaCl₅ activator (Birck 1986). ⁸⁷Sr measurements were corrected for isobaric ⁸⁷Rb by monitoring the ⁸⁵Rb mass and using the ⁸⁵Rb/⁸⁷Rb value of 2.59265. Samples were corrected for mass fractionation using an exponential mass fractionation law (Russell et al. 1978) and a ⁸⁸Sr/⁸⁶Sr value of 8.375209. All data was corrected to a NIST SRM 987 value of 0.710248 (Avanzinelli et al. 2005). Long-term reproducibility of NIST 987 analyses is 0.710228 ± 0.000034 (2 SD, n = 156) and a modern seal tooth 0.70918 ± 0.000033 (2 SD, n = 25). Total procedural blanks are on the order of 100 pg.

Defining the locally bioavailable ⁸⁷Sr/⁸⁶Sr range

Bioavailable strontium in Britain represents a combination of weathered strontium from underlying geology and precipitation (Montgomery 2010). The impact of strontium from rainfall is particularly substantial in areas of high precipitation on the western seaboard. Caerleon is located in an area of high rainfall and is also on the floodplain of the river Usk, and therefore, meteoric and river water would affect bioavailable strontium. In this study, the local ⁸⁷Sr/⁸⁶Sr range was estimated through the analysis of six modern plant samples, each from different locations surrounding Caerleon; seven dentine samples from the Priory Field excavations; as well as published biosphere data. The following discussion of the local range focuses principally on the plant results and is supported by published biosphere data. Dentine values from paired enamel samples are presented and have variable ⁸⁷Sr/⁸⁶Sr. This is not surprising, given the range of initial ⁸⁷Sr/⁸⁶Sr of their enamel and that diagenetic uptake of strontium into dentine is likely to be an additive process (Nelson et al. 1986; Conlee et al. 2009; Madgwick et al. 2012). Therefore, even dentine which is at equilibrium with the burial environment Sr will still have ⁸⁷Sr/⁸⁶Sr values that are a mass balance of the biogenic and diagenetic strontium and need not be isotopically equivalent to the diagenetic strontium. The dentine samples show a reduced range in ⁸⁷Sr/⁸⁶Sr relative to their initial enamel values, and all dentine samples lie on a vector in the ⁸⁷Sr/⁸⁶Sr space towards a diagenetic Sr value in the region of ~ 0.7103, which would be consistent with the measured biosphere ⁸⁷Sr/⁸⁶Sr of the site (0.710205, plant sample 4). In any case, simply establishing the signal for the precise location of the site is of negligible value in an area with such diverse geology as the south coast of Wales.

The geology around Caerleon is complex, and an interpretation of the range of the locally bioavailable ⁸⁷Sr/⁸⁶Sr must take account of this. The most useful source of data on the local range is the six analysed plant samples. These provide a bimodal distribution between less radiogenic values of approximately 0.7102–0.7106 (n = 4, plants 1, 3, 4 and 5) and more radiogenic values of 0.7133 to 0.7135 (n = 2, plants 2 and 6, see Fig. 2). The wide range in ⁸⁷Sr/⁸⁶Sr defined by these samples broadly matches what should be expected for locally bioavailable Sr. The plant samples do not cover all of the diverse lithologies in the vicinity of Caerleon. None were analysed from the Carboniferous limestone deposits to the east of the site, which are marine-derived sediments and represent a relatively well-known lithology. Evans et al. (2010) quote an average of 0.7092 ± 0.0002 (1 SD, n = 11) from plants on Carboniferous limestone in Britain. This is likely to represent the lower limit of bioavailable ⁸⁷Sr/⁸⁶Sr in the vicinity of Caerleon along with the marine-dominated tidal flats deposits on the coast to the south. The upper limit for bioavailable ⁸⁷Sr/⁸⁶Sr in the Caerleon area is defined by the plant measurements from the Silurian and Devonian Old Red Sandstone (ORS) lithologies surrounding the site. ⁸⁷Sr/⁸⁶Sr values of 0.713296 (plant 2) and 0.713548 (plant 6) have been measured for the Brownstones Formation and Raglan Mudstone, respectively. However, the ORS can give a range of bioavailable ⁸⁷Sr/⁸⁶Sr with less radiogenic values of approximately 0.7105 for the St. Maughans Formation and a paired measurement on the Raglan Mudstone. The upper limit of bioavailable ⁸⁷Sr/⁸⁶Sr is therefore estimated to be 0.7136. The Evans et al. (2010) biosphere map suggests a range of 0.712–0.713 for the ORS lithology, but this is extrapolated from the sampling of comparable geological areas, rather than being based on local analyses and may underestimate the true range of the locally bioavailable strontium. This represents a cautionary note to an over-reliance on published maps that can rarely achieve comprehensive coverage. It is essential for local biosphere mapping analyses to be combined with archaeological analyses.

Open image in new window

The plant results suggest that the ORS is a variable geological unit that can yield a range of bioavailable ⁸⁷Sr/⁸⁶Sr values including radiogenic values in excess of 0.713, and these very radiogenic values are worthy of further consideration. Squirrel and Downing (1969) give the source for the ORS beds as being the erosion of the Caledonian Mountains, and critically, they note that there is an abundance of mica minerals throughout the ORS succession. Everett et al. (2003) and Kinnaird et al. (2002) have argued, based on Pb isotope data and detailed mineralogical analysis, that the Caledonian source of the ORS has some input from a granitic material but that this source may not be spatially and temporally consistent. Where bioavailable ⁸⁷Sr/⁸⁶Sr for these granites has been measured in situ, they yield some of the most radiogenic biosphere ⁸⁷Sr/⁸⁶Sr values in Britain (Evans et al. 2010). The presence of some granitic input to the ORS is significant as this material will have abundances of mica and potassium feldspar minerals. Both of these have high Rb contents and elevated ⁸⁷Sr/⁸⁶Sr (for a given geological age), leading to weathered products (soils) that could lead to the more radiogenic bioavailable ⁸⁷Sr/⁸⁶Sr we observe for ORS sediments.

The plant analysis provided a bimodal distribution, but intermediate values between 0.7105 and 0.7135 may also be produced on ORS sediments in the vicinity. Even if the bimodal distribution is truly representative of the locally bioavailable ⁸⁷Sr/⁸⁶Sr, this would not mean that animals with intermediate values could not derive from the Caerleon area. In such a geologically diverse location, it is likely that locally raised cattle and sheep will have grazed on pasture that overlies different lithologies throughout their lives. Similarly, pigs are likely to have been foddered with resources from different areas and driving pigs to wooded areas to exploit seasonal beech mast and feeding them meal scraps would mean their biogenic ⁸⁷Sr/⁸⁶Sr would not derive from a single zone. Additionally, in an area of relatively high precipitation such as South East Wales, rainfall could dampen the more radiogenic biosphere values, generating ⁸⁷Sr/⁸⁶Sr values that are intermediate to the bimodal ranges. For these reasons, animal values would be expected to extend across the local range, depending on management and foddering regimes.

Overall, the locally bioavailable ⁸⁷Sr/⁸⁶Sr is estimated to have a lower limit of 0.709 (limited by marine strontium). The upper limit analytically places to 0.7136, which is conservatively revised to 0.714. These values can be obtained within a 5-km radius of the site, but the range is likely to encompass a considerably wider local area. It is not possible to define the geographical limits of what is defined as ‘local’ in this paper, but the range is considered representative of the extended hinterland of the fortress based on the diversity of values and the range of geological units sampled (Fig. 2). This is based on six plant samples and available comparative data, but more comprehensive mapping programmes may mean it will require future revision, particularly given the uncertainty over the ultimate source of the ORS deposits.

The Caerleon fauna

Results from isotope analysis are presented in Tables 2, 3 and 4 and Fig. 3. The dataset shows a high degree of variation, and ⁸⁷Sr/⁸⁶Sr values consistent with all of the major biosphere packages in Evans et al. (2010) are represented, with the exception of the least radiogenic zone (0.707–0.708) that is present in only very small pockets in western Britain. All three domesticates show a considerable variation, with pigs tending towards more radiogenic values. There is no significant difference in ⁸⁷Sr/⁸⁶Sr isotope values between cattle and sheep (Mann-Whitney U = 69.0, p = 0.422) or between cattle and pigs (Mann-Whitney U = 68.0, p = 0.230). Pig values were generally more radiogenic than caprines, but the difference narrowly missed statistical significance at 0.05 level (Mann-Whitney U = 27.5, p = 0.053). In any case, this difference cannot be considered significant, as multiple pair-wise testing runs the risk of type 1 error, the erroneous rejection of the null hypothesis, and therefore, a Bonferroni corrected significance level (Rice 1989) should be employed (0.017).

Table 4

Mean and 1 SD values for each class of the samples

	Mean 87Sr/86Sr	SD	Number
Bos	0.7106	0.0019	16
Sus	0.7123	0.0022	11
Caprine	0.7101	0.0019	10
Dentine	0.7118	0.0019	7
Plants	0.7114	0.0016	6

Open image in new window