Abstract
This study combines standard dendrochronological analyses with network science and spatial analysis to determine the provenance of wood used to build river barges in the Roman period. The river barges studied were found in the Lower Rhine region and would have carried bulk goods, such as grain, military supplies, and building materials. The importance of these vessels in the supply of the local economy and military units is evident since many were found in the area, including some in the vicinity of military complexes. However, it remains unknown where these ships were built and how and where the raw materials for their construction were obtained. To better understand the provenance of the wood, network science was applied to visualise and understand the complex patterns of similarity between the tree-ring curves. For the interpretation of the networks we have studied the context, the position of the trees in the network and the use of these trees in shipbuilding. In addition, the shape of the converted timber was used to visualise the wood use patterns in this type of Roman-period shipbuilding. For the river barges, we were able to determine several possible regions for wood procurement. Based on the analysed material, we assume that there was at least one shipyard in the Lower Rhine region where two ships, found in separate excavations, were most likely produced at the same time.
1 Introduction
The Roman period in the Lower Rhine region was a dynamic era. The Roman presence starts in the fifties of the first century BC, when Caesar and his legions fought various tribes in the region (van Es 1981; pp. 22–28). During the second half of the first century BC, the Roman presence became more permanent, and people started settling along the Lower Rhine in the area now known as the Netherlands (Kemmers 2005). During the first century AD, building activity in the region increased, and several army camps (Hallewas et al. 1993; Blom & Vos 2008; Polak 2014; Polak et al. 2004) and civilian settlements were built for which large amounts of wood were used, among other building materials (van Enckevort & Thijssen 2001; Eck 2004; Leih 2019). Infrastructure was developed, and various roads (Luksen-IJtsma 2011) and bridges (Fehr 1981; Tegel & Yupanqui-Werner 1999; Vos 2004) were built. In this transforming landscape, the transport of building materials and daily goods was an essential part of the development. Transport over land was probably at least ten times more expensive than transport over water (Duncan-Jones 1982: pp. 366–369; Kunow 1980: p. 23). Large quantities of archaeological objects found in the region were not locally produced or obtained and, most likely, were brought into the area via water transport. Various ship finds in the Lower Rhine region and the surrounding areas provide literal evidence of the importance of ships as a mode of transport. For example, ships were found in Zwammerdam (de Weerd 1988), Pommeroeul (De Boe 1978), and Mainz (Höckmann 1993; Bockius 2006). The town of Zwammerdam, down the Rhine, became synonymous with a certain type of ship — the Zwammerdam river barge. This type is a long, box-shaped cargo vessel built of oak, with L-shaped sides and a flat bottom without a keel. Over the years, several of these river barges were found, in particular, in the Lower Rhine region, for instance in Druten (Hulst & Lehmann 1974), Woerden (Haalebos 1996, 2000; Blom et al. 2008), Xanten (Obladen-Kauder 1995), and De Meern (de Groot & Morel 2007; Jansma & Morel 2007). In addition, similar ships and cargo vessels were found in other Roman territories within Europe, for example, in Switzerland (Bevaix and nearby Yverdon-les-Bains (Arnold 1992a, 1992b: pp. 9–19)), in France (Lyon (Parc Saint-Georges: Guyon & Rieth 2011) and Arles (Marlier 2014; Marlier et al. 2019)), in Croatia (Sisak (Gaspari et al. 2006)), and in Serbia (near Prahovo (Bockius 2003)).
The ships in the Lower Rhine region were found in wet environments, which preserved the wood and other organic materials. Their contexts reveal that some of them accidentally sank (Woerden 1: Haalebos 1997: p. 84; De Meern 1: van Dinter & Graafstal 2007: p. 36; Vos & Vorst 2007: pp. 92–94), while some were perhaps sunk on purpose (Zwammerdam 2/De Meern 4: van Dinter & Graafstal 2007: p. 32; de Groot & Morel 2007: p. 9; Vorst, in press). This clearly indicates that the ships functioned as transport vessels and were later sometimes used as deliberately-placed obstacles to influence, for instance, the water flow. However, we still do not know where these ships were built, since only a very small number of possible shipyards have been identified (Arnold 2009). There are a few indications of possible shipyards, slipways, or places on land (Morel 1987; Höckmann 2003), although some of these remain speculative (de Weerd 1988: pp. 41, 44). Since dendrochronological analyses can help to determine the provenance of wood from an archaeological context, it can perhaps pinpoint the shipbuilding places or regions, or, at least, help to determine the source of the wood used in shipbuilding.
This study aims to combine standard dendrochronological analyses of oak (Quercus spp.) wood from Roman-period river barges found in the Lower Rhine region with network and spatial analyses to determine the provenance of the wood used to build these ships and visualise their wood use patterns.
2 Material and Methods
The methodology applied in this study is visually explained in Fig. 1. Existing dendrochronological data was used and has been described in an earlier study (Visser, 2021a). In addition, data from several river barges was added to this dataset, with extensive datasets from the Zwammerdam barges (numbers 2, 4 and 6) and the Woerden 7 barge (Table 1). To make the research reproducible, all analyses and scripts are available via GitHub and Zenodo (https://github.com/RonaldVisser/NetworkRomanBarges (Zenodo: Visser & Vorst 2022). The methods and scripts applied from Visser (2021a) can be found at https://github.com/RonaldVisser/ProvenanceNetworks (Zenodo: Visser 2022)). The computational analyses were performed using R (R Core Team 2022) using various libraries (as explained later).
The methodology applied in this study. The various steps show how the data collection and analyses were stored in a database. Further analyses were based on the stored data, resulting in networks that integrate and visualise various attributes of the wood. The research processes are further explained in the paper
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
The ships in the dataset (for the rest of the dataset, see Visser 2021a)
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
The archaeological wood from the ships was analysed by the second author using standard dendrochronological methods (Bräker 2002). The tree-ring widths (TRW) were measured with an accuracy of 0.01 mm. The tree-ring series (TRS) were dated with all available reference chronologies using PAST4.3 (Knibbe 2011). The dates are based on the best statistical match, visual similarity and replication of the date in multiple reference chronologies. To determine if TRS are obtained from wood of the same tree, several similarity measures were used: Pearson’s correlation coefficient (r), t-values, Synchronous Growth Changes (SGC) and the related p-values (Visser 2021b), number of overlapping TRW s, and Euclidean distance (Visser 2015, 2021a). The SGC is an improved measure that replaces the Gleichläufigkeit (GLK), since the latter includes semi-similarities, which can lead to incorrect high values of the GLK between series with a low true similarity (Visser 2021b). The p related to the SGC is the probability of exceedance, that is, the chance that the similarity is incorrect. The similarity measures were calculated using dplR (Bunn 2008) and scripts (see aforementioned Github/Zenodo repositories). Strong visual similarities (t-values above 8 (Vorst 2005: p. 34)) were considered in conjunction with the plank lengths and widths and patterns in the knots to consider the wood as belonging to the same tree. Site chronologies were created after careful statistical and visual comparison of the material from a single ship or site. Only TRS or trees with strong similarities were combined into a site chronology. The grouped material was checked using COFECHA (Holmes 1992). The mean chronologies were created using two methods: first, the arithmetic mean without standardisation (M-chronologies) and second, the standardisation used in COFECHA based on a 32-year cubic spline with a 50% wavelength cut-off (Holmes 1983, 1992; Cook 1985).
To determine the provenance of the wood used to build these ships, networks were created following an earlier developed method (Visser 2021a). All tree ring material was compared, and all similarities calculated in R were stored in a PostgreSQL/PostGIS database (https://www.postgresql.org and https://postgis.net/). Networks were created using iGraph (Csardi & Nepusz 2006) and visualised in Cytoscape (Shannon et al. 2003; Smoot et al. 2011) via RCy (Gustavsen et al. 2019). Networks consist of nodes (or vertices), and the relations between nodes are visualised as edges (see Fig. 2). Tree-ring curves form the nodes in the network, and the edges were created using various similarity measures (r and SGC with p) and the overlap (noverlap). The r and Student’s t-values were calculated after standardisation (Hollstein 1980). The thresholds for edge creation were the same as those used in a previous study (Visser 2021a). The thresholds were based on generally assumed optimal values, and lowering these resulted in connections between tree-ring material that seemed related more to chance than to similar growth patterns (see also the comments on lower thresholds in Gut 2020). Four network types were created:
Schematic explanation of a dendrochronological network. Nodes represent dendrochronological curves and the edges (relations) are represented by the similarity between the curves
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
Type 1: r ≥ 0.5, SGC ≥ 70% with p ≤ 0.0001 and noverlap ≥ 50
Type 2: r ≥ 0.5, SGC ≥ 70% and noverlap ≥ 50
Type 3: r ≥ 0.5, any SGC with p ≤ 0.0001 and noverlap ≥ 50
Type 4: A merged network of types 2 and 3
The different threshold combinations (without p-threshold in type 2 and without SGC-threshold in type 3) created the possibility to study the influence of the various measures on grouping and determine the optimal combination of similarities. If, for example, the thresholds of type 1 for the similarity between tree-ring curve 1 and 2 in Fig. 2, are met, edge A is created in all the network types. However, if the SGC is below 70%, for instance, between curve 2 and 3 in Fig. 2, while all other thresholds are met, edge B is created only in the network of types 3 and 4. Since all relations are reciprocal, the network is undirected, and duplicate edges are removed. This simplification of the networks is based on retaining the edge with the highest overall similarity. The highest similarity for each edge is determined by finding max (SGC + r + (t/10)).
The study uses three levels of testing to create the networks, which result in networks with the following nodes:
(1) TRS with TRS
(2) Trees with trees
(3) Site chronologies with site chronologies
The advantages of these levels have been explained elsewhere (Visser 2021a), and similar levels have been applied in other studies (e.g., Daly 2007). For this study, slightly different levels are used. The first level consists of individual TRS, each representing a single sample. These are all the measurements in the dataset. For each sample, only the best matching radius is used for the creation of edges in the network(s). At this level, the network shows the matches between the different pieces of wood and presents different samples from the same trees in the network(s). The second level of testing consists only of trees. For each tree in the dataset, only one TRS is retained. If a tree is represented by multiple TRS, the mean curve of these TRS replaces all the TRS of this tree. This prevents the overrepresentation of some trees. The third level of testing is used to determine the provenance of the wood since, when carefully created, site chronologies represent a common signal of trees from the same area. To determine the provenance of site chronologies, various factors, such as location in the network, spatial locations, contexts, and the type of wood conversion, were considered (Visser 2021a).
Each network is described using standard measurements. The number of nodes (N), edges (E) and components, and the diameter of the network (
For the interpretation of the networks, various factors are considered, the most important being the different components of the network, the archaeological context, the position of the trees in the network, and the use of these trees in the ship. In addition, the shape of the converted timber is considered. The documentation standard developed by Biax Consult (www.biax.nl) is used to record the timber shapes (Fig. 3).
Codes used for various timber shapes (mostly planks and beams) in the barges. Most codes for round wood or radially split logs (1–6) are not presented, since these are practically absent in the timber used for the barges. Figure based on a documentation standard developed by Biax Consult (www.biax.nl)
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
3 Results and Interpretation
Networks were created for all the ships in the dataset. All the data and all the networks are available as supplementary data (Visser & Vorst 2022). The river barges included in the data are discussed below.
3.1 Ship Woerden 7
The river barge Woerden 7 (WOS7, Fig. 4) was excavated in the centre of Woerden in 2003 (Blom et al. 2008). The ship was about 26 m in length and 4.7 m in width and was constructed of oakwood felled in the autumn/winter of AD 162 and of oakwood with a felling date range between AD 160–164 (Vorst 2005: pp. 12–17; Blom et al. 2008: p. 378). The ship with the characteristic flat bottom amidships had wide open bow and stern areas, which ran up at a gentle angle. Heavy floor timbers ran along the width of the ship, and in the preserved stern area, the first pair of floor timbers formed a cross. The vessel had a rowing installation for propulsion and a (towing) mast placed in a mast step in a keelson set far forward.
The Woerden 7 ship during the excavation
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
Photo: Hazenberg ArcheologieThe dendrochronological analysis, based on the visual and statistical comparison of the TRS, resulted in the distinction of two site chronologies with different growth patterns (Vorst 2005: pp. 20–32; Blom et al. 2008: pp. 378–381).
The current study shows that the networks of TRS and trees pinpoint the same groups with no connection between the two components in the network of trees (see Table 2 for the network statistics). However, in the network of types 2 and 4 of the TRS, there are some connections between the components. These individual connections that exist between the series have high p values and are, therefore, less significant. These connections are most likely the result of promiscuous series, that are series that match with many series (Visser 2021a: p. 241). These connections result in hubs in the TRS networks and reflect the scale-free topology of tree-ring networks (Barabási & Albert 1999; Visser 2021a). The general pattern in the network components shows the different growth conditions of the wood in the two groups and the strongly connected networks within the components.
Network statistics of various ships and networks of site chronologies
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
It is interesting to find that certain ship parts have a strong relation to a component (Fig. 5). This is evident in the network of TRS. The stringers, mast step with keelson, and chine cleats are only present in the small component, whereas inserts and floor timbers are present in both components. However, the bottom planks are only found in the largest component. Similar patterns can be determined for the shapes of the converted timber, since shape and function have a strong relation. Both implicate a conscious choice of wood with certain growth conditions for certain applications. In addition, if we compare the TRS network with the nodes coloured by tree and ship part, it becomes clear that for some parts multiple trees were used. However, some trees were also used to produce different parts. This implies that the shipbuilders had access to complete trees instead of already converted wood delivered to the shipyard.
Network of TRS for the ‘Woerden 7’ (network type 1), showing the two components in the network of nodes: one larger and one smaller component. The visualisations show the nodes categorised by tree, shape, and ship part (from left to right)
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
3.2 The Zwammerdam River Barges
In the early 1970s, the Zwammerdam vessels, three large wooden river barges and three dugout canoes dating to the Roman period, were discovered during building works in the Dutch village of Zwammerdam. The vessels, historically found near the Roman fort of Nigrum Pullum, lay scattered along a section of timber shoring at the edge of the River Rhine, in front of the fort and were located at various depths (Fig. 6). All the vessels were made of oak and the river barges, in particular, were of exceptional size, ranging from 20 to 34 metres in length. At the time an early form of crowdfunding, supported by a former Queen of the Netherlands, allowed the vessels to be salvaged and placed in a conservation process to enable future display in a museum (de Weerd 1988: pp. 27–42). The preserved ships are currently being reassembled at a museum park in the Netherlands, near the site where they were originally found. The restoration project can be followed on the official website, www.archeon.nl/zwammerdamschepen.
The locations of the vessels of Zwammerdam in front of the fort of Nigrum Pullum
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
Drawing: Y. Vorst3.3 Ship Zwammerdam 2
The river barge Zwammerdam 2 (ZWS2), 22.5 m in length and 3 m in width, was constructed of oakwood felled in the winter of AD 205 (Fig. 7). The ship has a (towing) mast placed in a mast step in a keelson set far forward and a rowing installation for propulsion in the bow area.
The Zwammerdam 2 ship during the excavation
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
Photo: F. Gijbels, collection ACASA, University of AmsterdamThe dendrochronological analysis based on the visual and statistical comparison of the TRS resulted in the distinction of one site chronology.
The networks of TRS and trees show the presence of two to three different components (Fig. 8; Table 2 for the network statistics). There are no connections between the components in the TRS networks; however, the components change across the different network types. The TRS in the two smaller components in network types 1 and 3, are found together in the second largest component in network types 2 and 4.
Network of TRS for the Zwammerdam 2 (network type 1), showing two to three different components in the network of nodes. The visualisations categorised the nodes by tree, shape, and ship part (from left to right)
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
The L-shaped timbers are present only in one of the components in network types 2 and 4, whereas the larger beams are present in the other component. The bottom planks are found in all the components; however, the L-shaped chines exist only in one component. The latter are found in the same component as the side strakes. The two large side strakes are made from wood from a single tree. Some trees were used to make the bottom planks; nevertheless, the use of trees seems to be restricted to certain parts.
3.4 Ship Zwammerdam 4
The river barge Zwammerdam 4 (ZWS4), 34 m in length and 4.6 m in width, was constructed of oakwood felled in AD 97 (Fig. 9). Unlike the other Zwammerdam barges, it has bow and stern ends that curve upward relative to the flat midsection. Relatively little material on this ship has been researched, only to verify the date of AD 97 which was already published by Ernst Hollstein in his book on the European oak chronology (Hollstein 1980). Further sampling will be possible in the future since the ship has been preserved and will be rebuilt for museum display.
The Zwammerdam 4 ship visited by Queen Juliana of the Netherlands during the excavation
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
Photo: F. Gijbels, collection ACASA, University of AmsterdamThe dendrochronological analysis, based on the visual and statistical comparison of the TRS, resulted in the distinction of one site chronology.
The networks of TRS and trees show the presence of one larger component and two (network types 1 and 3) or three (network types 2 and 4) components with two nodes (Fig. 10; Table 2 for the network statistics). There are no connections between the components in the TRS networks. In the TRS network type 2 (and hence 4), there is one extra component. The larger component is nearly a complete network, showing the strong relations of the material.
Network of TRS for the Zwammerdam 4 (network type 1), showing one large component and two smaller components with two nodes. The visualisations categorised the nodes by tree, shape, and ship part (from left to right)
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
The small number of nodes makes this network hard to interpret. There is only one tree that is represented by more than one node in the TRS network. Bottom planks are found in the largest component and one of the smaller components. The L-shaped chines are only present in the largest component.
3.5 Ship Zwammerdam 6
The river barge Zwammerdam 6 (ZWS6) is about 20.5 m long and 3.6 m wide and was built in the third quarter of the second century A.D. (Fig. 11). Similar to the Woerden 7 river barge, it has wide open bow and stern areas, which run up at a gentle angle. At both the ship’s ends, a pair of floor timbers form a cross between which the other floor timbers run along the width of the ship. It has a rowing installation for propulsion and a (towing) mast placed in a mast step in a keelson set far forward. Similar to the Woerden 7 vessel, this ship was constructed of oakwood felled in the winter of AD 162 and of oakwood felled a year later in the summer/winter of AD 163.
The Zwammerdam 6 ship during the excavation
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
Photo: G. Verkuyl, collection ACASA, University of AmsterdamThe dendrochronological analysis, based on the visual and statistical comparison of the TRS, resulted in the distinction of two site chronologies with different growth patterns.
The networks of TRS and trees show the same groups in the material with no connection between the two components in the network of trees, except for types 2 and 4 of the TRS (Fig. 12; Table 2 for the network statistics). The individual connections that exist between series have high p values and are, therefore, less significant. The general pattern in the network components shows the different growth conditions of the wood in the two groups and strongly connected networks within the components.
Network of TRS for the Zwammerdam 6 (network type 1), showing two components in the network of nodes: a large and a small component. The visualisations categorised the nodes by tree, shape, and ship part (from left to right)
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
Similar to the Woerden 7, certain parts have a strong relation to the network component, evident in the network of the TRS. The side strakes, inserts and chine cleats are only present in the smallest component. Notably, this wood was felled a year later than the wood of the larger component. However, the bottom planks, floor timbers, L-shaped chines, mast step with keelson and stringers are only found in the large component. Similar patterns can be determined for the shapes of the converted timber, with conversion codes 10, 13 and 19 only in the large component and the smaller timbers (11 and 15) only in the small component. Some of the bottom planks (timber shapes 15 and 16) are found in both components. The strong relation between the shape and function is evident; therefore, a conscious choice of wood, with certain growth conditions for certain applications, was made when the ship was built. In addition, if we compare the TRS network with nodes coloured by the tree and ship part, it becomes clear that for some parts, multiple trees were used; however, the trees were also used to produce different parts. This implies, once again, that the shipbuilders probably had access to complete trees, instead of converted wood at the shipyards.
4 Combining the Evidence
4.1 A Network of the Trees Used in the Ships
The available dendrochronological dataset contains dendrochronological material from various ships and river barges. A network of all the trees from the ships shows that some of the ships form separate groups, that is, components, most clearly in network type 1 (Fig. 13). This is partly because of different growth patterns and different temporal coverage. However, there are strong connections between various ships (Fig. 13). The largest component shows the strong relationship between ships Woerden 7 and Zwammerdam 6 and strong ties with other vessels, including Zwammerdam 2 and 4, and one of the smaller boats, Zwammerdam 1. The second largest component consists mainly of Woerden 7 and Zwammerdam 6 and a few nodes from Zwammerdam 4. In addition, the latter has connections with the river barge, De Meern 4, in another network component. Another tree from De Meern 4 has a link in yet another network component with a ship found nearby, De Meern 1, which, in turn, has its own component but also some relationship to the river barge, Pommereul 4 (Houbrechts & Zambon 1995). The network of relations is complex, but shows the interconnectedness of the various provenances and probably building locations. The material from the river barge, De Meern 1, is grouped in three site chronologies/network components, showing that, for this ship, wood was used from at least three different regions.
Networks of trees with all available trees from the ships in the dataset. The visualisations have nodes coloured by ship. Network type 1 (on the left) provides the best overview, while network type 4 (on the right) is the most connected
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
The strong interconnectedness of Woerden 7 and Zwammerdam 6 becomes evident when we look at the TRS network of these two barges (Fig. 14). The largest component consists mostly of timber shapes 10 and 13. The ship parts in this component are, generally, bottom planks and floor timbers. The chine cleats of both vessels, however, are only found in the smaller component. These elements are important in construction for absorbing large amounts of impact when the ships are set ashore with their bow or stern ends. The inserts and the futtocks are mostly present in the smallest component, with some exceptions. It seems that the shipbuilders of these two ships made the same choices to use wood, with certain growth patterns and attributes, for various parts of the ship. Together with the fact that these ships have similar tree-felling dates and a similar appearance, it may point to the same shipbuilders and possibly the same shipbuilding location. Further evidence for this, directly related to both the ship constructions, will be discussed in a publication on the ships (Vorst in press).
Network of TRS for the Woerden 7 and Zwammerdam 6 (network type 1). The visualisations have nodes coloured by timber shape (left) and ship part (right)
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
4.2 Networks of Site Chronologies
The networks of site chronologies are slightly different than those published earlier (Visser, 2021a) due to the addition of new ship chronologies. The number of nodes and edges, and the clustering have increased (Table 2). The network is better connected and less sparse, making it better to draw conclusions (see below). The topology does not show any major changes, and there are no large-scale changes in the communities. This indicates the robustness of the method.
The provenance of the ship chronologies can be easily determined by the location in the network and the spatial relations (Figs 15 and 16). For instance, the chronologies that form the smallest components in the TRS and tree networks of Zwammerdam 6 and Woerden 7 can be found in the network of site chronologies near those that have provenance in the Western Netherlands (NL_West). The site chronologies of these two ships, which reflect the largest components in the TRS and tree networks, end up in the network with material from Alsace and Gallia Belgica; however, they are also linked to material from Germania Superior. These two provenance regions are roughly similar to an earlier study on the provenance of Woerden 7 (Vorst 2005); nevertheless, the provenance of the largest group/component is now better defined. Zwammerdam 4 material connects to Woerden 7 material in this part of the network, in the network of types 1 and 2. In network types 3 and 4, there are more connections, including the addition of Zwammerdam 2. The lower thresholds to form these network types result in a more complete network with new connections, yet no structural changes. Both Zwammerdam 2 and 4 seem to share the same provenance region, although the wood was obtained in possibly different locations.
The position of the various barges in the networks of site chronologies (network types 1 and 4). This network shows the common provenance of the wood of some of the ships and some of the variations in provenance
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
Map showing all site chronologies. For each barge, there are coloured lines with their first neighbours in the network (type 1). It reflects the spatial connections and the movement of material. The map was made with QGIS (rivers and boundaries are originally from Geodan)
Citation: International Journal of Wood Culture 3, 1-3 (2023) ; 10.1163/27723194-bja10014
The Pommeroeul ships are found in the same part of the network, although less well connected. In addition, there is a link to some of the material from De Meern 1, indicating a provenance in or around Western Belgium, the region with sandy soils in Belgium, and the southern Netherlands. This is more or less the same region, both spatially and in the network, where some of the wood was obtained for De Meern 4. The wood used to build these ships was, therefore, most likely obtained from regions that could be reached by the Meuse and Rhine rivers. It is important to emphasise here that the area surrounding the Meuse and Rhine has a high density of shipping-related inscriptions (Schmidts 2011, 82–83).
The combined evidence points to the possibility that Woerden 7 and, at least, Zwammerdam 6 were built on the same shipyard in the Lower Rhine region with wood from different provenances. The ships found in the other locations are more difficult to relate to a building location due to limited dendrochronological sampling. It seems that the shipbuilders of Zwammerdam 6 and Woerden 7 were provided, or provided themselves, with complete trees and converted these to the timbers they needed. Based on the small number of ships, we can only speculate on the number of possible shipyards in the region. The different provenances of the various ships and their wider common provenance region point to either a more centralised organisation, or a region that could supply wood suitable for shipbuilding. The former option was suggested in relation to the four (military) inscriptions of vexillationes agentis in lignarii in the early third century (Herz 1985; Nenninger 2001). It is equally possible that the shipbuilders (faber navalis, CIL XI, 139) and carpenters (fabri tignarii) were civilians and organised in collegia or corpora. However, there is some evidence that shippers (nautae) had strong ties with the shipbuilders (Schmidts 2011, 20–42).
5 Discussion and Conclusions
The network analyses illustrated that the TRS network types 2 or 3 (and, thus, 4) can result in strange connections between nodes. This is mostly due to the scale-free topology of the networks and general-matching series. This was less pronounced in the networks of trees or site chronologies. Therefore, network types 2 or 3 (and 4) must only be used to analyse data with trees or site chronologies. It appears that network type 1 is best for analysis of the TRS or trees. In addition, the differences between the network types were noticed in the analyses of the networks of site chronologies; however, here, they seemed to be less disturbing in the network topology. In general, it seems that the best interpretations can be based on network type 1.
The networks of TRS and trees are suitable for the interpretation of ships and archaeological structures. The networks can help understand and visualise complex relations between the pieces of wood within a structure. The TRS or trees, that function as a hub in a network, can be useful in dating undated material; however, these should not be used to determine the provenance of the material. The provenance of the material should be based on the networks of site chronologies, bearing in mind various factors (see Visser 2021a, 244).
In general, this study demonstrates how network analyses can contribute to the interpretation of dendrochronological material. The networks of TRS and trees can be used to understand past material use from a single structure or even an excavation and visualise the complex patterns of tree ring material and wood use. The network types with the strictest thresholds to create edges (r ≥ 0.5, SGC ≥ 70% with p ≤ 0.0001 and noverlap ≥ 50) are best suited for these levels. To determine the provenance of the material, it is best to create networks of site chronologies; nevertheless, the variation of thresholds to create the networks can be helpful in discerning meaningful patterns.
Roman-period river barges from the Lower Rhine region, sampled to varying degrees, show oakwood of different provenances used in the vessels and, those sampled extensively, a pattern of particular shapes in terms of timber conversion used in the vessels and associated with different provenances. This points to deliberate choices for the use of oak with certain growth patterns and attributes for specific parts of the ships. For reasons not exactly known, the shipbuilders specifically used wood from the Lower Rhine region for the impact-bearing chine cleats in both Zwammerdam 6 and Woerden 7. This Lower Rhine oakwood ended up being used for inserts and some of the futtocks, while the wood from further upstream was used for bottom planks and floor timbers. It is highly likely that they selected their wood for specific purposes. Further ideas regarding this will be discussed in a publication on the construction of the ships (Vorst in press).
For the Zwammerdam ships and Woerden 7, their wider common provenance region seems to indicate either a more centralised organisation or a region that could supply wood suited to shipbuilding. Combined with wood that is strongly connected in the network of site chronologies and material from the western Netherlands, it is likely that the shipyards were located downstream in the Lower Rhine region since this wood, which was used for specific parts in Woerden 7 and Zwammerdam 6, was cut just one year later. The river barges found in the other locations in and around the Lower Rhine area were more difficult to relate to possible shipbuilding regions due to limited dendrochronological sampling. Therefore, we recommend that vessels and other interconnected structures be sampled as completely as possible to better understand past activities.
Acknowledgements
We would like to thank Marta Domínguez-Delmás, Aoife Daly and Kristof Haneca for organising the inspirational ‘Forests to Heritage’ conference in Amsterdam in April 2022 and editing this special issue. We thank the anonymous reviewers and the editors of IJWC for their valuable help and useful comments. The research was partly funded by NWO (the Dutch Research Council) and supported by the Cultural Heritage Agency of the Netherlands (RCE) and the University VU Amsterdam as part of the NWO-research program (Humanities) 380-60-004. Saxion University of Applied Sciences enabled the first author to work on the research. Special thanks to former colleagues of RCE Lelystad and to Maarten de Weerd, Jaap Morel (†) and Tom Hazenberg for their continued support of our work. Apart from most of the river barge measurements recorded by the second author and the data from the Woerden 1 barge (kindly shared with the second author by M. Neyses), the dendrochronological data of other find locations was provided to the first author by several people from different tree-ring laboratories: S. van Daalen, M. Domínguez-Delmás, T. Frank, K. Haneca, P. Hoffsummer, B. Schmidt and W. Tegel. We are very grateful that they allowed the use of their data. The majority of the open data can be found in the DCCD (https://dataverse.nl/dataverse/dccd), and references can be found elsewhere (see Visser 2022, 2021a and https://github.com/RonaldVisser/NetworkRomanBarges (Zenodo: Visser & Vorst 2022)).
References
Arnold B. 1992a. Batellerie gallo-romaine sur le lac de Neuchâtel, Volume 1. Editions du Ruau, Saint-Blaise.
Arnold B. 1992b. Batellerie gallo-romaine sur le lac de Neuchâtel, Volume 2. Editions du Ruau, Saint-Blaise.
Arnold N. 2009. A gallo-roman naval building yard at Avenches / En Chaplix. In: R Bockius, M Schönfelder, AG Brown (eds). Between the Seas. Transfer and Exchange in Nautical Technology. Proceedings of the Eleventh International Symposium on Boat and Ship Archaeology, Mainz 2006: 167–175. Verlag des Römisch-Germanischen Zentralmuseums, Mainz.
Barabási A-L, Albert R. 1999. Emergence of Scaling in Random Networks. Science 286 (5439): 509–512. DOI: 10.1126/science.286.5439.509.
Blom E, Vorst Y, Vos WK. 2008. De ‘Woerden 7’: Een Romeinse platbodem. In: E Blom and WK Vos (eds). Woerden-Hoochwoert. De opgravingen 2002–2004 in het Romeinse Castellum Laurium, de vicus en van het schip de ‘Woerden 7’: 349–401. ADC Archeoprojecten, Amersfoort.
Blom E, Vos WK. 2008. Woerden-Hoochwoert. De opgravingen 2002–2004 in het Romeinse Castellum Laurium, de vicus en van het schip de ‘Woerden 7’. ADC Archeoprojecten, Amersfoort.
Bockius R. 2003. A Roman River Barge (?) Found in the Danube near Prahovo, Serbia. In: C Beltrame (ed). Boats, Ships and Shipyards: Proceedings of the Ninth International Symposium on Boat and Ship Archaeology, Venice 2000: 169–176. Oxbow Books, Oxford.
Bockius R. 2006. Die spätrömischen Schiffswracks aus Mainz. Schiffsarchäologisch- technikgeschichtliche Untersuchung spätantiker Schiffsfunde vom nördlichen Oberrhein. Verlag des Römisch-Germanischen Zentralmuseums, Mainz.
Bräker OU. 2002. Measuring and data processing in tree-ring research — a methodological introduction. Dendrochronologia 20 (1–2): 203–216. DOI: 10.1078/1125-7865- 00017.
Bunn AG. 2008. A dendrochronology program library in R (dplR). Dendrochronologia 26 (2): 115–124. DOI: 10.1016/j.dendro.2008.01.002.
Cook ER. 1985. A time series analysis approach to tree ring standardization. University of Arizona, Tucson, AZ.
Csardi G, Nepusz T. 2006. The igraph software package for complex network research. InterJournal Complex Systems: 1695: 1–9.
Daly A. 2007. Timber, Trade and Tree-rings. A dendrochronological analysis of structural oak timber in Northern Europe, c. AD 1000 to c. AD 1650. PhD Thesis. University of Southern Denmark, Odense.
De Boe G. 1978. Roman boats from a small river harbour at Pommeroeul, Belgium. In: J du Plat Taylor and H Cleere (eds). Roman shipping and trade: Britain and the Rhine provinces: 22–30. The Council for British Archaeology, London.
de Groot T, Morel JMAW. 2007. Het schip uit de Romeinse tijd De Meern 4 nabij boerderij de Balije, Leidsche Rijn, gemeente Utrecht. Waardestellend onderzoek naar de kwaliteit van het schip en het conserverend vermogen van het bodemmilieu. Amersfoort.
de Weerd MD. 1988. Schepen voor Zwammerdam. Bouwwijze en herkomst van enkele vaartuigtypen in West- en Middeneuropa uit de Romeinse tijd en de Middeleeuwen in archeologisch perspectief. Dissertation. Universiteit van Amsterdam, Amsterdam.
Duncan-Jones RP. 1982. The economy of the Roman empire. Quantitative studies. 2nd ed. Cambridge University Press, Cambridge.
Eck W. 2004. Köln in römischer Zeit : Geschichte einer Stadt im Rahmen des Imperium Romanum. Cologne.
Fehr H. 1981. Eine Rheinbrücke zwischen Koblenz und Ehrenbreitstein aus der Regierungszeit des Claudius. Bonn. Jb. 181: 287–300.
Gaspari A, Erič M, Šmalcelj M. 2006. Roman river barge from Sisak (Siscia), Croatia. In: L Blue, F Hocker, and A Englert (eds).Connected by the Sea, Proceedings of the Tenth International Symposium on Boat and Ship Archaeology, Roskilde 2003: 284–289. Oxbow Books, Oxford.
Gustavsen JA, Pai S, Isserlin R, Demchak B, Pico AR. 2019. RCy3: Network biology using Cytoscape from within R. F1000Research 8 (1774) DOI: 10.12688/f1000research. 20887.3.
Gut U. 2020. Assessing site signal preservation in reference chronologies for dendro- provenancing. PLoS ONE 15 (9): e0239425. DOI: 10.1371/journal.pone.0239425.
Guyon M, Rieth É. 2011. Les chalands gallo-romains du Parc Saint-Georges. In: G Boetto, P Pomey, and A Tchernia (eds). Batellerie gallo-romaine. Pratiques régionales et influences maritimes méditerranéennes: 91–101. Centre Camille Jullian, Aix-en- Provence.
Haalebos JK. 1996. Ein römisch Getreideschiff in Woerden. Jahrb. Röm.-Ger. Zentralmuseums Mainz 43: 475–509.
Haalebos JK. 1997. Een Romeins graanschip in Woerden. Jb. Oud Utrecht: 67–96.
Haalebos JK. 2000. Woerden, Oranjestraat. In: D Kok, K van der Graaf, and F Vogelzang (eds). Archeologische Kroniek Provincie Utrecht 1998–1999: 202–207. Matrijs, Utrecht.
Hallewas DP, van Dierendonck RM, Waugh KE. 1993. The Valkenburg-Marktveld and Valkenburg-the Woerd excavations, 1985–1988; a preliminary report. In: RM van Dierendonck, DP Hallewas, KE Waugh (eds). The Valkenburg Excavations 1985–1988. Introduction and detailed studies: 11–46. ROB, Amersfoort.
Herz P. 1985. Zeugnisse römischen Schiffbaus in Mainz — die Severer und die expeditio brittanica. Jb. Röm.-Ger. Zentralmuseums Mainz 32: 422–435.
Höckmann O. 1993. Late Roman Rhine vessels from Mainz, Germany. Int. J. Naut. Archaeol. 22 (2): 125–135. DOI: 10.1111/j.1095-9270.1993.tb00401.x.
Höckmann O. 2003. An early Roman boatyard at Mainz, Germany. In: C Beltrame (ed). Boats, ships and shipyards: Proceedings of the Ninth International Symposium on Boat and Ship Archaeology, Venice 2000: 109–112. Oxbow Books, Oxford.
Hollstein E. 1980. Mitteleuropäische Eichenchronologie. Trierer Dendrochronologische Forschungen zur Archäologie und Kunstgeschichte. Verlag Philipp von Zabern, Mainz am Rhein.
Holmes RL. 1983. Computer-assisted quality control in tree-ring dating and measurement. Tree-Ring Bull. 43: 69–78.
Holmes RL. 1992. Cofecha. Data quality control for crossdating and measurement. Published by the author.
Houbrechts D, Zambon JM. 1995. Bernissart/Pommeroeul: analyses dendrochronologiques des embarcations mises au jour à Pommeroeul. Chron. L’Archéologie Wallonne 3 (1994): 37–38.
Hulst RS, Lehmann LTh. 1974. The Roman Barge of Druten. Ber. Van Rijksd. Voor Het Oudheidkd. Bodemonderz. 24: 7–24.
Jansma E, Morel JMAW. 2007. Een Romeinse Rijnaak, gevonden in Utrecht-De Meern; resultaten van het onderzoek naar de platbodem “De Meern 1”. Rijksdienst voor Archeologie, Cultuurlandschap en Monumenten, Amersfoort.
Kemmers F. 2005. Coins for a legion. An analysis of the coin finds from the Augustan legionary fortress and Flavian canabae legionis at Nijmegen. PhD Thesis. Radboud Universiteit, Nijmegen.
Knibbe B. 2011. PAST4 — Personal Analysis System for Treering Research — Instruction Manual. SCIEM — Scientific Engineering & Manufacturing, Brunn.
Kunow J. 1980. Negotiator et vectura: Händler und Transport im freien Germanien. Marburg.
Leih S. 2019. Holz ohne Holz …?? Die Spuren des Holzes in archäologischen Ausgrabungen in Xanten. In: J Meurers-Balke, T Zerl, and R Gerlach (eds). Auf dem Holzweg … Eine Würdigung für Ursula Tegtmeier: 223–230. Propylaeum, Heidelberg.
Luksen-IJtsma A. 2011. De limesweg in West-Nederland. Inventarisatie, analyse en synthese van archeologisch onderzoek naar de Romeinse weg tussen Vechten en Katwijk. Cultuurhistorie Gemeente Utrecht, Utrecht.
Marlier S. 2014. Arles-Rhône 3, un Chaland Gallo Romain du Ier Siècle après Jesus- Christ. Centre national de la recherche scientifique, Paris.
Marlier S, Greck S, Djaoui D, Viviés P de, Bayle M, et al.2019. L’ épave Arles-Rhône 5, un nouveau chaland gallo-romain. In: D Djaoui, J Piton (eds). Archéologie et Histoire en territoire arlésien. Mélanges offerts à Jean Piton: 349–494. Mergoil.
Morel JMAW. 1987. Frührömische Schiffshäuser in Haltern, Hofestatt. Ausgrabungen Funde Westfal.-Lippe Münst. 5: 221–249.
Nenninger M. 2001. Die Römer und der Wald. Untersuchungen zum Umgang mit einem Naturraum am Beispiel der römischen Nordwestprovinzen. Franz Steiner Verlag, Stuttgart.
Obladen-Kauder J. 1995. Das römerzeitlichen Plattbodenschiff von Xanten-Wardt. Ein Land macht Geschichte. Archäologie in Nordrhein-Westfalen: 220–222.
Polak M. 2014. An early Roman naval base at Vechten (prov. Utrecht / NL): Facts and fiction. In: C Nickel, M Röder, and M Scholz (eds). Honesta Missione. Festschrift für Barbara Pferdehirt: 69–98. Verlag des Römisch-Germanischen Zentralmuseums, Mainz.
Polak M, Kloosterman RPJ, Niemeijer RAJ. 2004. Alphen aan den Rijn — Albaniana 2001–2002. Nijmegen.
R Core Team. 2022. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna.
Schmidts T. 2011. Akteure und Organisation der Handelsschifffahrt in den nordwestlichen Provinzen des Römischen Reiches. Schnell & Steiner, Regensburg.
Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, Amin N, Schwikowski B, Ideker T. 2003. Cytoscape: A Software Environment for Integrated Models of Biomolecular Interaction Networks. Genome Res. 13 (11): 2498–2504. DOI: 10.1101/gr.1239303.
Smoot ME, Ono K, Ruscheinski J, Wang P-L, Ideker T. 2011. Cytoscape 2.8: new features for data integration and network visualization. Bioinformatics 27 (3): 431–432. DOI: 10.1093/bioinformatics/btq675.
Tegel W, Yupanqui-Werner M. 1999. Neue römische Bauholzfunde aus Offenburg. Römische Brücke, Hafen oder Uferverbauungen. Nachrichtenblatt Arbeitskreis Unterwasserarchäologie 5: 59–61.
van Dinter M, Graafstal EP. 2007. Landschap en militaire infrastructuur rond het schip. In: E Jansma and JMAW Morel (eds). Een Romeinse Rijnaak, gevonden in Utrecht- De Meern. Resultaten van het onderzoek naar de platbodem “De Meern 1”: 17–36. Rijksdienst voor Archeologie, Cultuurlandschap en Monumenten, Amersfoort.
van Enckevort H, Thijssen J. 2001. Der Hauptort der Bataver in Nijmegen im 1. Jahrhundert n. Chr. Von Batavodurum und Oppidum Batavorum nach Ulpia Noviomagus. In: G Precht and N Zieling (eds). Genese, Struktur und Entwicklung römischer Städte im 1. Jahrhundert n. Chr. in Nieder- und Obergermanien: Kolloquium vom 17. bis 19. Februar 1998 im Regionalmuseum Xanten: 87–110. Verlag Philipp von Zabern, Mainz am Rhein.
van Es WA. 1981. De Romeinen in Nederland. Unieboek, Bussum.
Visser R, Vorst Y. 2022. Analyses, data and figures related to: “Connecting ships: using dendrochronological network analysis to determine the wood provenance of Roman-period river barges found in the Lower Rhine region and to visualise patterns of wood use.” DOI: 10.5281/zenodo.7243539.
Visser RM. 2015. Imperial timber? Dendrochronological evidence for large-scale road building along the Roman limes in the Netherlands. J. Archaeol. Sci. 53: 243–254. DOI: 10.1016/j.jas.2014.10.017.
Visser RM. 2021a. Dendrochronological Provenance Patterns. Network Analysis of Tree-Ring Material Reveals Spatial and Economic Relations of Roman Timber in the Continental North-Western Provinces. JCAA 4 (1): 230–253. DOI: 10.5334/jcaa.79.
Visser RM. 2021b. On the similarity of tree-ring patterns: Assessing the influence of semi-synchronous growth changes on the Gleichläufigkeitskoeffizient for big tree-ring data sets. Archaeometry 63 (1): 204–215. DOI: https://doi.org/10.1111/arcm.12600.
Visser RM. 2022. Dendrochronological Provenance Patterns. Code and Data of Network Analysis of Tree-Ring Material. DOI: 10.5281/zenodo.7157744.
Vorst Y. 2005. De constructie en herkomst van de Romeinse platbodem “Woerden 7”. Een studie van jaarringpatronen en bewerkingssporen. MA Thesis. Universiteit van Amsterdam.
Vorst YE. in press. A way through the woods. A dendroarchaeological study of Roman period river barges from Zwammerdam and Woerden. Thesis.
Vos AD. 2004. Resten van Romeinse bruggen in de Maas te Maastricht. Rijksdienst voor het Oudheidkundig Bodemonderzoek, Amersfoort.
Vos PC, Vorst YE. 2007. Geolandschappelijk onderzoek. In: E Jansma and JMAW Morel (eds) Een Romeinse Rijnaak, gevonden in Utrecht-De Meern; resultaten van het onderzoek naar de platbodem “De Meern 1”: 36–94. Rijksdienst voor Archeologie, Cultuurlandschap en Monumenten, Amersfoort.
Watts DJ, Strogatz SH. 1998. Collective dynamics of “small-world” networks. Nature 393 (6684): 440–442. DOI: 10.1038/30918.
