Hillforts in the Post-Roman West

The (re)occupation of hillforts was a distinctive feature of the post-Roman (fifth to seventh) centuries across much of Europe (e.g. Kobylinski, Reference Kobylinski, Austin and Alcock1990; Francovich & Hodges, Reference Francovich and Hodges2003: 61–74; Constant et al., Reference Constant, Ségura, Valenciano and Vicelja-Matijašić2015; Tejerizo-García & Canosa-Betés, Reference Tejerizo-García and Canosa-Betés2018). This post-Roman ‘hillfort phenomenon’ was complex and multi-faceted, coinciding with other changes, including the abandonment and/or reorganization of Roman villas, farmsteads, and urban settlements (Christie, Reference Christie2004; Lewit, Reference Lewit2004; Sami & Speed, Reference Sami and Speed2010). It can be strongly linked to the evolution of power structures and the associated socio-political, economic, military, and ideological transformations that related to the end of Roman imperial control in Europe (Wickham, Reference Wickham2005). In northern Spain, for example, the (re)occupation of hillforts has been associated with the transition from imperial, State-backed ‘extensive’ power structures to more local, elite-focused ‘intensive’ systems of power in the fifth and sixth centuries (Tejerizo-García & Canosa-Betés, Reference Tejerizo-García and Canosa-Betés2018: 21–24). In western and northern Britain, hillforts are seen as loci for the negotiation of patron-client relationships through feasting, manufacture of prestige items, collection of render, and gift-exchange (Seaman, Reference Seaman2013; Noble, Reference Noble, Christie and Herold2016). Excavated evidence comprising fine metalworking residues and imported pottery and glass, in congruence with occasional documentary references, attest to high-status occupation and in some instances ceremonial activity associated with sacral kingship (Lane & Campbell, Reference Lane and Campbell2000; Noble et al., Reference Noble, Gondek., Campbell, Evans, Hamilton, Taylor, Noble and Evans2019). These sites are interpreted as the seats of potentates (land-holding rulers) rather than community settlements and can be seen as theatres of power akin to the great hall complexes of Anglo-Saxon England and Scandinavia (Blair, Reference Blair2018: 114–25; Thomas, Reference Thomas2018: 263).

In western and northern Britain, ‘hillfort’ is used as an umbrella term for a range of enclosed sites including promontory forts, duns, and ringworks (Alcock, Reference Alcock2003: 179–99; Seaman, Reference Seaman, Christie and Herold2016). They are often reoccupied Iron Age sites, but de novo constructions are also well evidenced. Nevertheless, site-based activities have produced limited diagnostic material and only ephemeral remnant settlement features. Alongside scarce historical documentation, this greatly constrains our understanding of the post-Roman period throughout western and northern Britain. Furthermore, such paucity inhibits the application of traditional forms of landscape archaeology, such as fieldwalking and aerial survey. Thus, most research on post-Roman hillforts has been site-focused, and there has been little consideration of their wider context. This contrasts significantly with Iron Age studies where landscape research, such as the Danebury Environs Project, have significantly enhanced understanding of hillforts in some regions (e.g. Cunliffe, Reference Cunliffe2000). Attempts at reconstructing the landscape context of post-Roman hillforts have so far often resorted to the problematic use of settlement evidence relating to prehistoric, Roman, and later medieval periods (e.g. Lane & Campbell, Reference Lane and Campbell2000: 255–58; but see Davey, Reference Davey2005 for an examination of the landscape context of Cadbury Castle in Somerset in the late-Roman and early medieval periods). Together, these issues mean that our understanding of the significance of post-Roman hillforts and the factors that influenced their siting is limited.

Several recent studies have, however, demonstrated that GIS-enabled analysis can offer new ways of studying the landscape. For example, in their innovative study of the Pillar of Elise, an early medieval carved cross in Powys (Wales), Murrieta-Flores and Williams (Reference Murrieta-Flores and Williams2017) used least-cost paths (LCPs) and viewsheds to argue that the location and topographic context of the monument enhanced its significance as a place of royal inauguration and a locus of power, faith, and commemoration. Viewshed analysis by Tejerizo-García and Canosa-Betés (Reference Tejerizo-García and Canosa-Betés2018) provided perspectives on the landscape context of northern Iberian hillforts that enabled the authors to propose new models of early medieval hillfort occupation. Finally, Semple and colleagues (Reference Semple, Buchanan, Harrington, Oliver, Petts, Semple, Orsini and Mui2017) have used LCPs to explore the ‘locational strategies’ of the important Anglo-Saxon great hall complexes at Yeavering and Milfield (Northumbria) and contextualize their relationship to long-term routes of movement and communication.

The analytical power afforded by GIS promotes development beyond the traditional cartographic ‘bird's eye view’ way of interpreting the landscape, to what Tejerizo-García and Canosa-Betés (Reference Tejerizo-García and Canosa-Betés2018: 3) dub a ‘God's eye view’, espousing the accessibility to and manipulation of multiple datasets. GIS applications have the potential to significantly enhance understanding of the landscape context, especially when movement and visibility analyses are integrated (Lock et al., Reference Lock, Kormann, Pouncett, Polla and Verhagen2014) and used in combination with data derived from other forms of landscape research (in our case paleoenvironmental and place-name research projects: Seaman, Reference Seaman2017; Davies et al., Reference Davies, Rippon and SeamanIn prep).

The theoretical underpinning of this approach is discussed below, but from the outset we must highlight the two methodological challenges that lie at the heart of this study. First, LCPs and viewsheds largely rely on high-resolution digital terrain models (DTMs) and their derivatives. Datasets created from direct measurement of the modern landscape are readily available but, since lowland districts where most historic settlements were located have often been substantially altered by housing and infrastructure such as roads and railways, the applicability of these datasets to the post-Roman period is questionable. The impact of significant landscape change is seldom acknowledged in archaeological research, which tends to assume that, whilst the precise setting of sites and monuments is irrecoverable, the ‘bones of the land’ (the mountains, hills, rocks and valleys, escarpments and ridges) remain the same over the longue durée (e.g. Tilley, Reference Tilley1994: 73–74). We know this was not always the case and, if landscape changes are not adequately addressed, they could undermine the validity of GIS-derived analysis or constrain the application of these approaches to less developed, predominately upland, landscapes. Second, GIS-enabled analysis requires defining key environmental and cultural parameters, such as historic land-use patterns, palaeovegetation, or the location of river crossings and focal points within the landscape (Herzog, Reference Herzog2014). Without proper definition, we encounter further risks of bias and misinterpretation. Thus, the present study had two primary aims: first, to establish a methodology for the creation of datasets that are appropriate for GIS analysis of pre-modern lowland landscapes. Second, to use these data in a pilot study of the landscape context of Dinas Powys, an important and comparatively well understood post-Roman hillfort in western Britain.

Dinas Powys: A Post-Roman Power Centre in Wales

Dinas Powys (ST148722, Glamorgan, South Wales) (Figure 1A) is the most extensively excavated post-Roman hillfort in Wales; it is also the richest, in terms of the quality and quantity of its material evidence, and an important type-site of the post-Roman Celtic West (Alcock, Reference Alcock1963; Campbell, Reference Campbell1991; Seaman, Reference Seaman2013). This site was chosen for the present study not only because of its archaeological importance, but also because it lies within a lowland landscape that has seen extensive post-medieval development, including construction of major roads and railway lines.

Figure 1. Location map and site plans. B: Dinas Powys: after Alcock, Reference Alcock1963, based on survey by RCAHMW, with additions. C: Llanvithyn. Banks in dark grey, ditches in light grey. © Crown Copyright and Database Right (June 2019) OS (Digimap Licence).

Dinas Powys is a small promontory fort enclosed by four sets of banks and ditches on its southern side. It occupies the now very wooded tip of a whaleback ridge, with the ground dropping-off steeply to the north. Three of the four banks (1, 3, and 4) are massive (at least 6 m wide and 4 m high from bank top to ditch base) and cover an area (0.25 ha) far larger than that enclosed by the innermost bank (0.1 ha) (Figure 1B). Leslie Alcock, who excavated the site in the 1950s, initially dated banks 1, 3, and 4 to the eleventh century ad. However, following comprehensive re-analysis of the finds and stratigraphy in combination with targeted radiocarbon dating, all the banks can now be confidently assigned to the fifth and seventh centuries ad (Alcock, Reference Alcock1963; Campbell, Reference Campbell1991; Seaman, Reference Seaman2013). Alcock's excavations revealed extensive evidence for fifth to seventh century occupation, including at least five hearths, drainage gullies associated with two buildings, and rich animal bone assemblages in midden contexts. The artefact assemblage is the largest of its type in Wales and includes fragments of over seventy rare pottery and glass vessels imported from production centres in the eastern Mediterranean and southern France. There were also twenty fragments of composite bone combs and a range of iron objects, such as spear ferrules, knives, and tools. Metalworking debris including crucible fragments, a brooch die, and metal slag attest to on-site manufacture of fine jewellery in copper-alloy, silver, and gold (Alcock, Reference Alcock1963). The site is interpreted as the residence of a local ruler, a status that is illustrated by fragments of a rare blue glass squat beaker of a type also known from ‘princely’ Anglo-Saxon burials such as Sutton Hoo Mound 2 and Prittlewell (Campbell, Reference Campbell1989).

Dinas Powys has been argued to lie within a small ‘petty kingdom’ focused on the eastern Vale of Glamorgan and Cardiff basin (Campbell, Reference Campbell1991: 225; Seaman, Reference Seaman2013: 13, fig. 5). This territory is likely to have formed after the collapse of the Romano-British administration in the fifth century and was subsumed into the larger kingdom of Glywysing around the time that Dinas Powys was abandoned in the later seventh century (Davies, Reference Davies1978: 97, Reference Davies1990: 37). Whilst we can reconstruct some of the site's political background, less is known about its landscape context, the significance of its location, and its role within systems of governance and territorial control.

Dinas Powys is the only well-excavated post-Roman settlement in the region, but it shares many characteristics with an unexcavated bivallate promontory fort near Llanvithyn (see Figure 1C), which will be considered for further investigation.

GIS and Landscape Analysis: Methods and Theories

The ability of GIS to model and integrate complex datasets means we can use it as a ‘bridging concept’ for interpretive frameworks that seek to understand the construction and articulation of social space (Hu, Reference Hu2011; Llobera, Reference Llobera2012). In this study, 3D visualization, least-cost paths (LCPs), and viewshed analysis are employed to examine how the landscape of Dinas Powys was reflected in the exercise of elite power. The analyses undertaken are comparatively simple ‘off the shelf’ forms of cartesian analysis that have long been applied in landscape archaeology (Conolly & Lake, Reference Conolly and Lake2006), but we have adopted an integrated approach that moves beyond ‘typecast’ analysis focused on identifying intervisibility between sites (Lock et al., Reference Lock, Kormann, Pouncett, Polla and Verhagen2014; Gillings, Reference Gillings2017). There is a wide literature on the application of GIS in landscape archaeology (for a recent overview see Howey & Brouwer Burg, Reference Howey and Brouwer Burg2017), and critiques of the tools employed in this study focus on the reductive nature of movement modelling and the primacy given to visual perception over other forms of sensory engagement (Frieman & Gillings, Reference Frieman and Gillings2007; Supernant, Reference Supernant2017). It is necessary, therefore, that we consider how our approach differs and why it is appropriate for exploring socio-political aspects of hillfort siting and development.

As is well established, some hillforts were sited tactically in relation to topography, and monumentality and visual prominence could be exploited to exert control over a landscape and/or define hillforts as ‘special’ places (Bowden & McOmish, Reference Bowden and McOmish1989; Hamilton & Manley, Reference Hamilton and Manley2001: 31; Harding, Reference Harding2012: 15; O'Driscoll, Reference O'Driscoll2017; Driver, Reference Driver2018). Conspicuous sites could reinforce ideological statements, as projected for example by hillfort ‘defences’ (Seaman, Reference Seaman2013: 10–12, Reference Seaman, Christie and Herold2016: 41), and act as permanent visual reminders of elite power (Jones et al., Reference Jones, Jones and Woods2004: 117–18). Such aspects of a hillfort's siting are difficult to assess on the ground, especially where recent woodland regeneration and limited public access constrain fieldwork. We can, however, use 3D visualization and visibility methods to investigate relevant topographic positioning, theoretical zones of visibility, and visual prominence (Lock et al., Reference Lock, Kormann, Pouncett, Polla and Verhagen2014; O'Driscoll, Reference O'Driscoll2017). Spatial analysis of hillfort locations can also illuminate adjacency and oversight of routeways and strategic nodes such as river crossings through the modelling of patterns of movement by LCPs. This makes it possible to conduct a robust debate about how social control was implemented through the monitoring of strategic nodes within the landscape and maximizing elite movement between centres. Patterns of visibility, as reconstructed through cumulative and total viewsheds, can be used together with modelled patterns of movement from derived LCPs to examine how social control was generated through surveillance and visual domination. Foucault (Reference Foucault1977: 195–230), for instance, argued that power over individuals could be achieved through the creation of panoptic surveillance mechanisms. Such ideas of surveillance existed in the nineteenth century, as extolled in Jeremy Bentham's panopticon, a prison designed to ‘induce in the inmate a state of conscious and permanent visibility that assures the automatic functioning of power’ (Foucault, Reference Foucault1977: 201; for archaeological applications of these ideas, see Carlson & Jordan, Reference Carlson and Jordan2014; Pierce & Matisziw, Reference Pierce and Matisziw2018). We do not contend that landscapes could be organized in such a way that they operated precisely like a panopticon but assert that visual domination can operate as a method of social control (Giddens, Reference Giddens1985: 14–15). The importance attributed to the ‘elite gaze’ in the early medieval world is exemplified through the literary motif of the sentinel warrior: a leader buried on the boundary of his kingdom overlooking enemy territory to protect his people against that enemy (O'Brien, Reference O'Brien2008). On a more practical level we see it manifest in the built environment; Ray and Bapty (Reference Ray and Bapty2016), for example, argue that the design of Offa's Dyke facilitated surveillance of strategic nodes on the frontier between Merica and Powys, inferring that the earthwork operated as a tool of Mercian political domination.

Since the significant research potential afforded by GIS-enabled analysis is tempered by the difficulty of defining meaningful parameters and appropriate datasets from which to work, the first step in this project was the creation of topographic datasets that are broadly applicable to the post-Roman period. Our methodology for achieving this was rooted in ‘retrogressive landscape analysis’. This is a key principle of landscape archaeology that involves applying the rules of horizontal and vertical stratigraphy to identify and remove successive layers of landscape alteration and, thus, achieve an understanding of a historic landscape (e.g. Williamson, Reference Williamson1987; Rippon, Reference Rippon2004).

Creating a Digital Terrain Model

Our area of interest was initially defined as the entire Dinas Powys proto-kingdom, as reconstructed in Seaman (Reference Seaman2013: fig. 5). However, it became clear that recent urban and industrial development in and around the city of Cardiff was too extensive to warrant inclusion, and so the boundary of the study was shifted west, from the River Rhymney to the River Taff (Figure 2). The basis for the Digital Terrain Model (DTM; all analyses were undertaken in Ersi ArcGIS 10.5) used in this project was the 2 m-resolution LiDAR data freely available from Natural Resources Wales (http://lle.gov.wales/Catalogue/Item/LidarCompositeDataset/?lang=en). This provided coverage of most of the study area. Several gaps in the data were filled using the Ordnance Survey 5 m-resolution terrain data provided by Edina Digimap (https://digimap.edina.ac.uk/). These datasets had previously been processed to remove most of the surface features such as buildings and trees, thus introducing a level of abstraction. In addition, the cell resolution within the 5 m data was re-gridded (with no value change) to match that of the 2 m LiDAR. This process will have introduced an additional level of abstraction, but the resulting composite DTM provides a representative picture of the modern topography and is of sufficient resolution for the generation of a cost-surface analysis (Verhagen et al., Reference Verhagen, Nuninger, Groenhuijzen, Verhagen, Joyce and Groenhuijzen2019).

Figure 2. The Dinas Powys proto-kingdom as defined in Seaman (Reference Seaman2013). The area between the Rhymney and Taff, which includes Cardiff and its suburbs, was excluded from the study.

This DTM was processed to remove above-ground structures but was still heavily influenced by extensive post-medieval landscape alterations associated with the construction of railway lines, major roads, and quarries. These features, largely represented by substantial linear cuttings, have dramatically altered patterns of visibility and the natural ‘cost’ of moving across the terrain and, hence, also had to be removed for our analysis to be meaningful. The next step was, therefore, to regress the DTM to a pre-modern state, achieved through retrogressive analysis using contemporary and historic Ordnance Survey maps. Polygon shapefiles were created around all post-medieval features that cut into the natural topography. Once all features judged to have significantly altered the topography had been identified, they were merged into a single shapefile that was used as a clip template for the raster DTM. The resulting holes were then ‘patched over’ by using Triangulated Irregular Network (TIN) models, to draw the true surface values across the gaps and form a cohesive surface. This created a DTM devoid of intrusive post-medieval earthworks, although the interpolation process will have introduced a further degree of abstraction. It was not possible to reintroduce lost topographical features, but analysis of historic mapping suggests that no major features have been removed. A series of test LCPs were run across this surface (see below). These showed that the patched features were not unduly influencing the route of the LCPs. This contrasted with LCPs derived from the unaltered DTM, which tended to run along roads and rivers or follow unreasonable diversions along contours rather than enter modern cuttings. At this stage, the coastline and river courses were adjusted to their earliest recorded positions and the DTM clipped accordingly. This process provided us with a DTM that could be used as the basis for 3D visualization and visibility analysis. However, further steps were required before we could create a cost-surface for use in a meaningful LCP analysis.

Deriving an Accumulated Cost-Surface

The relationship between power centres and contemporary routeways through the landscape is potentially informative. However, the origins of the earliest recorded network of footpaths surrounding Dinas Powys is poorly understood and appears to have developed over an extended period, whilst the modern road network is largely a creation of the post-medieval period. Our methodology, therefore, focused on creating a network of LCPs that would accurately reflect natural patterns of movement through the landscape (White, Reference White2015). The LCPs were defined in relation to the speed of travel between defined nodes (considered below) across an accumulated cost-surface. The cost-surface was created by combining three environmental factors thought to influence the speed of pedestrian travel: slope, ruggedness, and land-cover (Herzog, Reference Herzog2014). Naismith's rule (a calculation of the time taken for a hill walk) was adopted for determining the speed of travel across different slope increments (Carver & Müller, Reference Carver and Müller2014: 23). Relative Topographic Position (RTP) was used as the ruggedness metric, acting as a proxy for the physical challenge posed when traversing an area (Riley et al., Reference Riley, DeGloria and Elliot1999), and was considered suitable because the terrain of the study area does not have extensive elevation or slope diversity. Land-use was defined in relation to four broad categories: ‘open ground’ (for which there is no impediment to travel), followed by ‘woodland’, ‘moorland’, and ‘marsh/mudflat’ representing an increasing scale of impairment. These land-use units were identified through a Historic Landscape Characterization exercise that included a comprehensive analysis of fieldnames and land-use data recorded in nineteenth-century tithe surveys, in addition to a limited programme of geoarchaeological survey (Rippon, Reference Rippon2004; Seaman, Reference Seaman2017; Davies et al., Reference Davies, Rippon and SeamanForthcoming). Once the metrics for slope, ruggedness, and land-use had been defined (see Table 1), they were united using a mean operator and represented in the resulting accumulated cost-surface as seconds per (2 m) cell (s/c) × 1000.

Table 1. Metrics used in the calculation of the accumulated cost-surface.

The final step was to account for the impact of the rivers and streams that cross the study area. Except for historic crossing points, identified through the analysis of place-names (for example ffordd ‘road, ford’ or pont ‘bridge’) and historic maps such as Yates's 1799 map of Glamorgan (Walters & James, Reference Walters and James1984), rivers were considered a total barrier to lateral movement unless they were deemed readily passable on foot. These crossings were defined as the point where footpaths marked on the first-edition Ordnance Survey map did not deviate from their course to cross rivers or streams. This is obviously an approximation, but it was considered more appropriate than simply removing rivers from the cost-surface. Movement along rivers was not considered, since none within the study area is likely to have been navigable during this period (Oksanen, Reference Oksanen2019).

Defining Nodes for LCPs

Once the cost-surface had been created, the next stage was to identify start and end points for a network of LCPs that would provide a fair approximation of long-term movement patterns within and across the study area. This was achieved by using the principal points of entry/exit around the edge of the study area (Figure 3). In the south, nodes were determined at the historic landing places on the Bristol Channel coast. The location of landing places in this region was restricted by tides and the nature of the coastline, and nineteenth-century evidence suggests that different locations were chosen to suit certain conditions (Bedford, Reference Bedford1872), for example, it is likely that pilots used a network of landing places in the region as opposed to a single point of entry. We can be confident that these would have been transhipment points for a range of commodities during the post-Roman period, including prestigious imported pottery and glass (Campbell, Reference Campbell2007). Landing places were identified by examining entries in sixteenth-century port books, nineteenth-century navigational texts, and published archaeological surveys (Bedford, Reference Bedford1872; Lewis, Reference Lewis1927; Dunning & Howell, Reference Dunning and Howell2005). In the east and west of the study area, entry/exit points were defined as historic crossing points on the rivers Taff and Thaw. Entry/exit points in the north were defined as the early ridgeway tracks leading north into the Glamorgan uplands, some of which are associated with early medieval stone monuments and linear earthworks. These are thought to have been transhumance tracks associated with the movement of people and livestock between the northern uplands and the fertile lowlands of the Vale of Glamorgan (RCAHMW, 1976b: 2–5). It is important to emphasize that Dinas Powys was not an active node within the network, since our intention is to understand how the site was imposed upon pre-existing patterns of movement.

Figure 3. The network of LCPs with viewsheds from Dinas Powys and Llanvithyn (in grey). Entry/exit nodes referred to in the text are labelled. The Roman road is highlighted.

Once nodes were identified, Cost Back Link and Cost Distance surfaces were created, and LCPs calculated between all nodes in both directions. Supplementary paths were also created between the nodes and the Gloucester to Carmarthen Roman road that bisects the study area from east to west, since this is likely to have been a major routeway in post-Roman centuries (Burnham & Davies, Reference Burnham and Davies2010: 93–96). The resulting network was then edited to remove redundant or nonsensical paths (Figure 3). We do not claim that the final network of linear polylines represents a map of post-Roman routeways. We must expect that there was a degree of ‘fuzziness’, that rough and un-made tracks would have become ‘braided’ over time and that some of the routes may not have been used at all. The network nevertheless provides a reasonable approximation of natural movement corridors through this landscape. Indeed, there are some correlations between the network and footpaths recorded in early maps. Some validation of the network is provided by considering it in light of the distribution of later medieval castles and moated sites, many of which are thought to have been located next to important routeways (Creighton, Reference Creighton2002: 35–65; Liddiard, Reference Liddiard2005: 24). Of the twenty-four castles, mottes, ringworks, and moats within the study area, seventeen (71 per cent) were located within 300 m of one or more of the LCPs. This figure rises to twenty-one (87 per cent) with an increased threshold of 500 m.

Llanvithyn: A Post-Roman Promontory Fort?

We believe the siting of Dinas Powys, which is currently the only secular settlement within the petty kingdom with firm excavated evidence for post-Roman occupation (but see Davis & Sharples, Reference Davis and Sharples2016: 48 and below for possible post-Roman activity at Caerau), makes most sense if it formed part of a wider network of elite residences. There is, however, a strong case for suggesting that an unexcavated bivallate promontory fort near Llanvithyn (ST054718) may have performed a role similar to that of Dinas Powys. Llanvithyn occupies the eastern tip of a spur protruding into the valley of the Nant Llancarfan (Figure 1A). Its ramparts are much denuded by ploughing, but appear to have enclosed an area of around 0.1–0.2 ha (Figure 1C). The site has not been excavated, but was classified as a ‘small fort with close-set multiple defences’ by the Royal Commission (RCAHMW, 1976a: 50). Dinas Powys was included in this group of sites but, despite their similarity, Llanvithyn has been ascribed an Iron Age date (RCAHMW, 1976a: 14; Davis, Reference Davis2017). Nevertheless, Dark (Reference Dark1993: 20–21) has made the case that these small, inland promontory forts were a post-Roman type-site, and the parallels with Dinas Powys run deeper than morphology. First, Llanvithyn is close to Llancarfan (1.6 km to the south), which, like Llandough, is known to be an important monastic centre from at least the seventh century and possesses a fragment of early medieval stone sculpture (Davies, Reference Davies1978; Redknap & Lewis, Reference Redknap and Lewis2007: 565–67). Second, Llanvithyn lies immediately next to the junction of natural routeways leading to/from coastal landing places at Barry and Aberthaw, and historic river crossings near Cowbridge and Llan-fair. Indeed, viewshed analysis suggests that at least 10 km of routeways were easily observable from the site (Figures 3). Third, Llanvithyn also seems to reflect the same compromise between defence, visual prominence, and proximity to routeways that we have observed at Dinas Powys (Figures 3 and 4). Until the site is excavated, we cannot be certain of its date, and the setting could apply just as much to the Iron Age as to the post-Roman period. The similarities with Dinas Powys are, nevertheless, striking and we may hypothesize that the sites were contemporary and operated within the same network of elite sites.

The Iron Age hillfort at Caerau (ST134750) (Figure 3) is another site worthy of consideration. This large (5 ha) site is also located next to the confluence of natural routeways, has commanding views across a large area to the east and north, and would have dominated routeways leading from historic crossing points on the rivers Taff and Ely. Interestingly, the site has more limited views to the south, where Dinas Powys is located, and it would not have been able to oversee the natural routes leading north from the coast. Thus, in terms of visual surveillance, Dinas Powys and Caerau are complementary. The directors of an ongoing excavation campaign at Caerau have suggested there may be evidence for a post-Roman phase of fortification there (Davis & Sharples, Reference Davis and Sharples2016: 48). This could date to the fifth to seventh centuries, but no post-Roman material has been recovered so far. A later phase of refortification associated with a Norman ringwork on the eastern promontory of the hill is perhaps more likely. The area enclosed by the Iron Age ramparts would have been difficult to defend with a small warband and would place Caerau amongst the largest post-Roman hillforts in Britain; occupation centred on the site of the later ringwork is nevertheless plausible and has previously been suggested (Dark, Reference Dark1993: 132).