Introduction

The concept of central place has traditionally played an important role in the study and interpretation of Roman urbanisation to explain the position and hierarchy of urban centres, secondary settlements, and surrounding provisioning rural sites, such as villae (Woolf 1998; Bintliff 2002). The Lower Rhine region (Fig. 1), however, is an atypical case. The area played a crucial role in protecting the empire’s economic and military interests between 15 BCE and 275 CE as part of the Lower Germanic limes. After the conquest of the region, the local tribes, in particular the Batavians, were required to provide auxiliary soldiers to the Roman army. Furthermore, a string of Roman forts as well as legionary camps with associated towns were established along the Rhine, providing a permanent military presence in the region. The region’s ‘urbanisation’ was thus the result of a deliberate Roman policy to facilitate the military administration of the region and to integrate it into the global Roman political and economic network. However, despite this strong influence of the Roman army, the continuation of tribal identities is observed throughout the Roman period (Roymans 1995; 2004; 2011). Also, it was one of the few border regions of the Empire that witnessed a complete collapse of its ‘urban society’ in the late third century, for which the reasons are still unclear (Heeren 2015; Roymans et al. 2020).

Fig. 1
figure 1

Map of the Lower Rhine region with major towns and tribes, with the study region outlined in red

Full size image

The major urban centres of Cologne, Xanten and Nijmegen are clearly aligned with the transport axes of the rivers Rhine and Meuse and of the road system that developed alongside (Kunow 1988). In contrast, there is hardly any evidence for the existence of larger ‘secondary centres’ or vici that are much more common further south in Gaul and Germania (Hiddink 1991; Woolf 1998; Nüsslein 2016), and large villae settlements are very rare north of the line Cologne—Tongres (Habermehl 2011; Roymans et al. 2015). The Lower Rhine region in the Roman period thus exhibits a weak interaction between the towns and their hinterlands, with ‘urban’ and ‘rural’ settlement development following seemingly separate trajectories. Nevertheless, differences in the socio-economic status of rural settlements can be observed in the archaeological record, in particular through the presence of more or less ‘Romanized’ building styles. It has been suggested that these differences resulted from a mostly autonomous socio-economic development of increasing agricultural specialisation and production (Vos 2009), in which direct Roman interference only played a limited role. The extent of this differentiation, however, is less clear, even when several micro-regional studies have analysed rural settlement patterns in considerable detail (Jansen and Fokkens 1999; Vossen 2003; van Londen 2006; Vos 2009; Heeren 2009; de Bruin 2017).

These studies did not perform a formal analysis of the hierarchical structure and spatial configuration of the rural settlements, making it hard to analyse and compare settlement hierarchy at the regional level. The sometimes assumed ‘standard’ Roman hierarchical structure of towns, vici, villae, large rural settlements and small rural settlements (Bloemers 1983; Kunow 1988; Vos 2009) seems too simplistic for this. Therefore, in this paper, a method is presented and applied to characterise and analyse the hierarchical structure of the rural settlement system quantitatively and spatially, using a model of settlement hierarchy originally developed by Bertoncello et al. (2012). The results of this analysis are then confronted with prevailing theories on the development of the region’s settlement pattern.

Settlement dynamics in the Dutch Lower Rhine region in the Roman period

Settlement numbers

Before the Romans came to the area, the study region was characterised by relatively low densities of small rural settlements. A significant rise in the number and size of settlements has been extensively documented in the Early Roman B and Middle Roman A periods (25–150 CE). Settlements are habitually dated based on expert judgement, following the classification of subperiods in the Dutch national archaeological database ARCHIS (Table 1).

Table 1 Periodization according to ARCHIS
Full size table

The accuracy of these datings is highly variable, depending on the number and nature of the objects found. It can also be questioned whether the datings supplied are always supported by the registered find materials, with many of the finds being poorly dated as well (see also Hiddink and Roymans 2015). For this reason, Verhagen et al. (2016b) applied aoristic sum calculations (Crema 2012) to obtain probabilities of settlements being occupied during a certain period. Each find was attributed a probability of belonging to a specific subperiod. These probabilities were summed and normalised to obtain dating probabilities for the whole settlement. While this method will not lead to more reliable datings of the finds themselves, it quickly identifies the uncertainties in the dating of the settlements. Simulations were then used to obtain probabilities of occupation: finds were attributed to a specific subperiod when their probability of belonging to this period was higher than a random number between 0 and 1. If 10 or more finds were then attributed to a specific subperiod in more than 50% of the simulations, this was considered sufficient evidence for occupation.

From this analysis, it was concluded that the number of settlements in the region rose by approx. 70% in the Early Roman B and Middle Roman B periods, followed by a decline of about 40% in the Late Roman A period, which is usually linked to the collapse of the Lower Germanic limes as a fixed frontier (Heeren 2015). For this paper, this analysis was re-run with additional data collected up to 2019. In total, 230 settlements (21.2%) have less than 10 finds registered, so they have not been included. Another 243 settlements do not have sufficient finds per subperiod to meet the criterion applied, leaving us with a total of 612 for which we can be more precise (Fig. 2; the full data analysis results can be found in Appendix A).

Fig. 2
figure 2

Development of settlement numbers in the study area, based on aoristic analysis (after Verhagen et al. 2016b)

Full size image

The absolute increase and decrease in population, however, must have been larger than the settlement numbers indicate (see also Groenewoudt and van Lanen 2018). First of all, the average size of rural settlements increased during the Middle Roman period, even when they generally did not grow beyond the size of three households (Vossen 2003; Verhagen et al. 2019). Secondly, non-rural settlements (forts, vici and towns) started to appear during the Early and especially Middle Roman period. Their appearance and growth are closely tied to the military occupation of the limes area, with the vici appearing alongside the forts mainly from 70 CE onwards. It seems that the non-rural population may have made up 20–30% of the total, a much larger proportion than what is normally assumed for the core areas of the Roman Empire (Verhagen et al. 2016a; 2019).

The observed demographic trend is clearly reflected in rural settlement density measures. Nearest neighbour distances of settlements decrease from a mean of 1289 m in the Late Iron Age to 968 m in the Middle Roman period, to increase again to 1158 m in the Late Roman period (see Table 2/Fig. 3).

Table 2 Mean nearest neighbour distances and nearest neighbour index (NNI) of rural settlements
Full size table
Fig. 3
figure 3

Distance to nearest rural settlement

Full size image

Settlement distribution

Analysis of the distribution of settlements on the palaeogeographical map of the region shows a clear preference for levees, but during the Middle Roman period, settlement expands into the floodplains as well (Groenhuijzen 2018, 217–247). The total area settled increases from the Late Iron Age into the Roman period: approx. 30% of the new sites dated as Early and Middle Roman appear in areas that are more than 1500 m away from previous settlement (Fig. 4). This is most evident in the west of the region and in the central river area. In the Late Roman period, settlement clearly contracts again. About 23% of the deserted Middle Roman settlements are at more than 1500 m distance from the nearest Late Roman settlement location. This effect, however, is more evenly spread over the area.

Fig. 4
figure 4

Distance at which new settlements appear and disappear (cumulative frequencies)

Full size image

A conspicuous characteristic of the settlement pattern is its high degree of spatial clustering. This is a direct consequence of the geography of the area, with a concentration of habitable zones on the river levees and near the coast (see also Verhagen et al. 2016b; Groenhuijzen 2018, 217–247). Percolation analysis (Maddison and Schmidt 2020) of the dataset reveals a clear separation of the eastern and central river area from the coastal zone (Fig. 5). Only in the Middle Roman period do we see some filling in of the settlement pattern along the western course of the Rhine and Meuse. The boundary between the two clusters is placed at the castellum of Woerden and possibly represents the extent of the Batavian and Cananefatian civitates.

Fig. 5
figure 5

Results of percolation analysis at 3 km (top) and 9 km (bottom) distance for the Middle Roman period. Clear clusters can be distinguished around Noviomagus and Forum Hadriani. The central river area remains highly clustered as well

Full size image

This clear clustering of rural settlement in two zones only starts to break apart at cluster distances of 5 km and less. Separate clusters first appear in the western area, whereas the eastern cluster remains almost completely intact until the cluster distance drops below 3 km. The river courses clearly influence the spatial configuration of the clusters, although it can be suspected that uneven data collection also has an effect at smaller distances. The clear cluster in the Midden-Delfland region (between The Hague and Rotterdam) can be associated with the hinterland of the Cananefatian civitas capital of Forum Hadriani (current Voorburg), and for Nijmegen (Noviomagus), clustering is obvious as well. Also, the location of the forts and their vici does not seem to play a role in cluster configuration.

Analysing settlement hierarchy

A classical central place model would assume a highly regular and hierarchical system of settlement, but the available rural settlement data make it hard to distinguish such a hierarchy, and it is even more challenging to trace the development of individual settlements through time. In order to better understand the potential hierarchical structure in the dataset, the available settlement data (see Appendix A) were classified according to the concepts originally defined for the Archaeomedes and ArchaeDyn projects (Bertoncello and Gandini 2005; Bertoncello et al. 2012; Nüsslein 2016; see also Hiddink 1991), based on the assumption that certain characteristics of settlements reflect differences in socio-economic status.

Building material

The first criterion used is building material. Most settlements in the region only show evidence for post-built, wooden structures. It is assumed that ceramic and stone building materials are signs of higher status (see also Jeneson 2013), and that specific find categories such as tubuli, hypocaust tiles or window glass constitute evidence of the highest possible status. An analysis of the registered finds per site thus results in four major hierarchical groups (Table 3/Fig. 6), with most rural settlements showing no finds of durable building materials and only a small group displaying high-status elements that are typically associated with villa-like settlements.

Table 3 Hierarchical classification of rural settlements according to building material
Full size table
Fig. 6
figure 6

Hierarchical classification of rural settlements according to building material

Full size image

Imported ceramics

Similarly, we can look at the provenance of imported ceramics as an indicator of wealth. Ceramic provenance ranges from local, hand-made ceramics to regionally produced wares and high-status pottery imported from Gaul and further south. Table 4 lists the main pottery types and their approximate dating. More detailed typologies and chronologies are available for these (e.g. Hiddink 1991; Geerts et al. 2020), but the subdivision presented here is the maximum level of detail that can be derived from the datasets. The pottery can be characterised according to the exchange networks involved as being from local (trade between settlements), regional (within the civitas), interregional (between civitates) or interprovincial origin (empire-wide trade; Van Kerckhove 2015). The difference between regional and interregional origin is, however, hard to establish on the basis of the level of detail in the dataset.

Table 4 Main pottery imports distinguished in the settlement data
Full size table

The overall distribution of pottery types in rural settlements is given in Table 5/Fig. 7. The category ‘no ceramics’ corresponds to sites known through metal detection only or where the ceramic finds have not been identified, so these sites do not constitute a separate hierarchical category. Also, the ‘unspecified imports’ category might contain pottery from all the import categories distinguished. The distribution is also partly influenced by the dating of the pottery (see also Nüsslein 2016).

Table 5 Hierarchical classification of rural settlements according to the type of ceramic finds
Full size table
Fig. 7
figure 7

Hierarchical classification of rural settlements according to the type of ceramic finds

Full size image

Alternatively, we can look at the variety of pottery types as an indicator of status (see Nüsslein 2016; Table 6/Fig. 8). High variety and the presence of ‘high status’ ceramics are clearly correlated (Table 7).

Table 6 Hierarchical classification of rural settlements according to the variety of ceramic finds
Full size table
Fig. 8
figure 8

Hierarchical classification of rural settlements according to the variety of ceramic finds

Full size image
Table 7 Correlation between the type and variety of ceramic finds. Contingency coefficient = 0.861, Cramer’s V = 0.757
Full size table

The large variety in the number of types present suggests differential access to ceramics. However, compared to building materials, we can see a more evenly spread hierarchical range with a higher proportion of ‘high status’ finds. The terra sigillata pottery, for instance, which is often considered a luxury item, is present in 43.0% of all rural settlements. This points to a much wider circulation of imported pottery than of building materials. This is also clear when cross-tabulating building materials and ceramics: 78.0% of the rural settlements without building materials have imports registered.

Hierarchical classification

In order to arrive at a consistent and meaningful hierarchical classification of materials found, seven groups were defined (Table 8/Fig. 9). This subdivision is, apart from the need to create groups of reasonable size, based on the consideration that investments in building materials will have been more costly than obtaining fine table wares and imported foodstuffs. The pottery criterion therefore only is distinctive for sites where no building materials were found. However, a small fraction of sites with ceramic building materials (roof tiles) but only low-status pottery is grouped in the lower hierarchies, considering that these tiles could be obtained and transported easily and therefore do not necessarily indicate a higher status. The classes are relatively evenly distributed, which is confirmed by the calculation of Hill’s index of 0.95 (Hill 1973; see also Nüsslein 2016).

Table 8 Hierarchical subdivision of sites based on building materials and ceramic finds
Full size table
Fig. 9
figure 9

Hierarchical subdivision of sites based on building materials and ceramic finds

Full size image

Settlement size and duration

The hierarchical classification scheme developed by Bertoncello et al. (2012) used two more indicators of status: settlement size and duration. These are both problematic in the context of the Dutch limes. When looking at the number of house plans and other built structures found in excavations, we can get a rough indication of settlement sizes (Table 9), but these figures cannot be extrapolated with any confidence to sites that are only known through survey since there has been no systematic registration of the spread of artefacts in a surface survey. Also, factors like erosion, sedimentation and urbanisation can make it hard to determine the approximate size of settlements during a survey. This factor can therefore not be reliably operationalised in the context of the current study, but from the excavated sites, almost 60% have only 1 or 2 house plans registered.

Table 9 The number of registered house plans in excavated settlements
Full size table

For duration, we must deal with the uncertainty in the dating of the sites. In order to control for this, the analysis was re-run with all poorly dated finds removed (see ‘Settlement dynamics in the Dutch Lower Rhine region in the Roman period’), leaving a total of 612 settlements (Table 10). A small number of these (35) would seem to have been occupied in the Late Iron Age only and thus to have been classified erroneously as Roman. The rest of the short-lived sites are almost exclusively placed in the Middle Roman period. It can be suspected that this is, apart from reflecting larger quantities of finds from this period, also an effect of the nature of the finds material itself, that is easier to date for this subperiod. In the categories of ‘medium’ and ‘long’ duration, we mainly find sites that continue into the Roman period from the Late Iron Age. The category of sites of ‘extended’ duration also contains sites that have large amounts of poorly dated finds.

Table 10 Duration of reliably dated settlements
Full size table

Given that a large portion of rural settlements cannot be reliably dated, it is problematic to include this criterion in the hierarchical classification. Combining the information on duration with the hierarchical classification of finds also shows that sites with no ceramics are poorly dated (Table 11/Fig. 10). For the rest, however, there does not seem to be a very strong relationship between find materials and settlement duration. It has therefore not been included in the analysis.

Table 11 Cross-tabulation of hierarchical classification according to duration and ceramic finds
Full size table
Fig. 10
figure 10

Duration of sites according to the hierarchical classification of ceramic finds

Full size image

Spatial analysis of settlement hierarchy

As discussed in ‘Settlement dynamics in the Dutch Lower Rhine region in the Roman period’, zones of high settlement density are found in the Kromme Rijn area and to a lesser extent in the western Midden-Delfland region and the central and eastern river area. The high densities in these areas are partly the consequence of higher research intensities, but it seems unlikely that the spatial distribution of settlements is extremely skewed because of that (however, see the debates in Vossen 2003; Vos 2009; and van Lanen et al. 2018). Settlement is also strongly correlated with the natural environment (see Verhagen et al. 2016b; Groenhuijzen 2018, 217–247), with the large majority of settlements concentrated on natural levees. The high-density areas also seem to show concentrations of higher rank settlement (Fig. 11), but this pattern is not very pronounced.

Fig. 11
figure 11

Distribution of the hierarchical classes over the study region. While some clustering is evident, the general distribution of classes is relatively even

Full size image

In order to better detect possible spatial hierarchy, hierarchical range and hierarchical diversity (Bertoncello et al. 2012) were calculated within a specific spatial range. For each rural settlement, the mean hierarchical rank of surrounding rural settlements and its standard deviation were determined for distances of 1500, 3000 and 9000 m. The difference between maximum and minimum rank was then used as a proxy for hierarchical range and the standard deviation for hierarchical diversity.

Because of the limited temporal resolution of the dataset, this was only done for the total dataset of rural Roman-period settlements, not for the individual subperiods. At 1500 m distance, the mean hierarchical rank of the surrounding settlements increases from 3.27 for HCLASS1 sites to 4.13 for HCLASS7 sites (see Fig. 12).

Fig. 12
figure 12

The mean hierarchical rank of settlements within 1500 m per class

Full size image

This is only a modest increase, but the differences are statistically significant between HCLASS1-2 and HCLASS4-7 (Table 12), indicating that the lower-ranked sites, on average, have less high-ranked sites in their direct surroundings, even when the number of surrounding settlements varies considerably regardless of hierarchical rank (ANOVA p = 0.278). Also, there are no significant differences in hierarchical diversity between the different ranks (ANOVA p = 0.186).

Table 12 Tukey’s HSD test of mean site ranks at 1500 m distance
Full size table

At 3000 m distance, this picture is largely confirmed, with a statistically significant difference in mean rank between the lowest (HCLASS1-2) and highest ranks (HCLASS5-7). Differences with HCLASS3-4 are less significant (Table 13). Due to the larger number of settlements within the 3000 m neighbourhood, differences in hierarchical diversity and the number of surrounding settlements are more significant. HCLASS1-2 have less neighbours than the other groups. The lowest diversity is found in HCLASS4, which is perhaps not surprising given that it is at the middle rank.

Table 13 Tukey’s HSD test of mean site ranks at 3000 m distance
Full size table

At 9000 m, the differences are smaller due to the smoothing effect of incorporating a larger number of sites. Within each settlement’s neighbourhood, we now find on average 8.2% of the total number of rural settlements. Nevertheless, the trends described remain discernible also at this highest spatial aggregation level.

However, we should be careful when interpreting this analysis. The classification used is ordinal, not linear, so the use of means and standard deviations is not ideal. Also, the spread of values is considerable within each class. Furthermore, the pooling together of all Roman period sites implies that not all sites analysed may have been occupied at the same time, and it can be expected that the hierarchical differences will have increased from the Early Roman into the Middle Roman period.

What is well reflected in the analysis results is the spatial patterning. When mapping the means and standard deviations of hierarchical rank in GIS (Figs. 13 and 14), the Kromme Rijn area, the eastern river area and Midden-Delfland region have the highest mean rank of settlements at 3 km, but at 9 km, the Midden-Delfland region clearly loses importance. At 3 km, the strongest hierarchical differences are found in the Midden-Delfland area and in the southeastern river area. At 9 km, this pattern is even clearer, although the spread of standard deviations is smaller. In general, the lowest hierarchical differences in settlement are found in the south of the central and western river area, located on the outer margins of the limes zone. However, we should keep in mind that, especially at larger distances, there is an edge effect. Settlements south of the line Cuijk (Ceuclum) – Oss were not included in the analysis, and the neighbouring area of Germany was not included either.

Fig. 13
figure 13

Mean class ranks at 3000 m (top) and 9000 m distance (bottom)

Full size image
Fig. 14
figure 14

Standard deviations of class rank at 3000 m (top) and 9000 m distance (bottom)

Full size image

In order to better understand what factors may have influenced this pattern, the mean and standard deviation of the rank of settlements within a 1500 m radius were calculated for all palaeogeographical units. Except for the marginal zones (peatlands), these are all similar, indicating that the natural environment is not an important factor in determining hierarchical rank (Fig. 15).

Fig. 15
figure 15

Boxplots of mean hierarchical rank within 1500 m distance per palaeogeographical unit

Full size image

Discussion

The spatial analysis of patterns of rural settlement hierarchy shows that higher-status settlements are relatively evenly spread within zones of both lower and higher settlement density, reflecting a somewhat regular but weak hierarchical structure. The surroundings of the civitas capitals of Noviomagus (in the eastern river area) and Forum Hadriani (in Midden-Delfland), however, exhibit a higher hierarchical rank and diversity of rural settlement. For the castella and other civil settlements, this is not observed. The Kromme Rijn area is the only area that shows higher hierarchical rank and diversity without the presence of an urban centre.

Two counteracting processes have been suggested in literature that have influenced the patterning of rural settlement in the area. On the one hand, the Roman authorities interfered directly to better integrate the region into the larger empire by developing urban settlements and infrastructure. On the other hand, the local population maintained a strong local identity throughout the Roman period and seems to have resisted a complete socio-economic integration.

Roman interference started with the building of the military forts in the region and went hand in hand with the establishment of new towns and military vici (Polak 2009). The inhabitants were mainly immigrants from Gaul who initially relied on external, supra-regional networks for much of their provisioning. During the first century CE and especially from 70 CE onwards, the Romans increasingly tried to integrate the province of Germania Inferior into the larger empire. This is witnessed by the expansion of both Noviomagus (Willems and van Enckevort 2009) and Forum Hadriani (Driessen and Besselsen 2014), by ambitious building programmes, especially under emperor Hadrian, and by the overall increase in regionally produced goods, in particular ceramics. It is also clear that the region could and did produce surplus crops and animals (Groot 2008; van Dinter et al. 2014; Joyce 2019), in particular, emmer wheat and barley (most probably used as fodder and for the production of beer), cattle (for meat, dairy products and leather), sheep (for wool) and horses (as riding animals for the military) that were not only intended to pay taxes in kind, but also to trade locally and regionally. The observed evidence for rural settlement hierarchization can thus be linked to the increase of local provisioning and strengthening of local supply lines to the forts and towns from 70 CE onwards, as part of the socio-economic integration of towns, forts and countryside.

This is in particular relevant for the town of Nijmegen, which is surrounded by a relatively high concentration of settlement and has a higher proportion of high-rank settlements, including a few villae and temples. Van Enckevort (2005) suggested that this stronger hierarchical division of settlement was the consequence of the direct involvement of the Roman military. The stationing of 5000 soldiers of the Legio X in Nijmegen between 70 and 104 CE and a possibly equal or even larger amount of followers in the canabae surrounding the legionary camp (Haalebos 2001) must have placed an immediate and heavy burden on the productive capacity of the region. Van Enckevort (2005) thus sees the installation of villae and the building of temples in the area as a conscious attempt to increase productive capacity and to strengthen the bonds between the Romans and the local population. This idea of deliberately increased surplus by the military is, however, contested by Habermehl (2011) who considers the villae to have been established by the local elite rather than to have been the result of a ‘planned military economy’. In either case, it suggests that highly organised surplus production in the area around Nijmegen was pursued and thus profitable.

The area around Forum Hadriani, on the other hand, has a less marked hierarchical settlement structure. There are only a few villa-like settlements in the area, and given the low number of inhabitants of the town, which by most authors has been estimated as no more than 600–1000 people (de Bruin 2017), there probably was little need for many highly productive rural settlements. We should also consider that the town only started to grow in earnest after around 120–125 CE, with a harbour built around 160 CE. In comparison, Noviomagus had been part of the supra-regional Roman network since at least 10 CE. The time between the installation of the military-administrative infrastructure and the falling into disuse of the harbour of Forum Hadriani around 230 CE may therefore have been too short to have had a more marked effect on rural settlement development.

In contrast, the forts and associated vici only seem to have played a marginal role in rural settlement development. Distance analysis to the closest castella and associated civil settlements shows that these have significantly less than average the number of settlements within their surroundings (Fig. 16). This confirms the analysis by Weaverdyck (2019) who could only detect a very limited effect of their ‘market potential’. Most of the forts are peripheral to both the regional transport system and rural settlement. Weaverdyck reasoned that since the rural settlements produced and traded mostly livestock, which is a product with low transportation costs, there was no need for farmers to settle close to the forts, or to strongly concentrate the flow of animals in larger, intermediate centres. This is, however, a hypothesis that warrants further investigation, especially since various sites with storage facilities for grain have been identified by archaeologists (Vos 2009). Local network analysis (Groenhuijzen 2018, 187–198), taking into account the 25 nearest neighbours of each settlement, showed that almost all identified intermediate centres have a lower total path length to neighbouring sites than to the nearest forts, implying that they were relatively easy to reach compared to the forts. This however does not fully explain why some sites developed into such centres, whereas others that were in equally advantageous positions in the network did not.

Fig. 16
figure 16

Boxplots of the mean number of rural settlements within 1500 m distance for ‘urban’ and rural settlements

Full size image

In order to further increase production, more large rural settlements would have been necessary. However, there is no development of the region into a ‘cash-crop’ producing area with large villae like we see in the loess region further south (de Groot 2006; Jeneson 2013). This has been attributed to the response of the local population to Roman interference. Instead of fully embracing the Roman way of life as is observed in the south of the province around Cologne and Tongres, the population of northern Germania Inferior maintained a strong regional identity with only limited involvement in the ‘urbanized’ economy. This is reflected in the persistence of a pastoral culture focused on cattle and horse breeding and in the warrior culture of the Batavians as auxiliaries in the Roman army (Roymans 1995, 2011; Schalles 2001). Also, goods obtained via long-distance trade are very rarely found in the countryside (Haalebos 2001), pointing at parallel markets that did not interact or compete (Schalles 2001). However, the analysis presented here also shows that regional trade must have been widespread (see also Roymans and Derks 2015), suggesting that a strong settlement hierarchy was not a precondition for at least some level of socio-economic integration. This may in part be attributed to the role of returning veteran soldiers as agents in spreading Roman culture in the region, as has been argued by Roymans (2011) for the case of Italian terra sigillata. The veterans, however, do not seem to have been involved in the development of villa-type settlements in the region.

Conclusion

In this paper, the hierarchical structure of Roman rural settlement of the Dutch Lower Rhine region was characterised and mapped using the available settlement data. By applying a formalised hierarchical classification of building materials and ceramic imports, it was possible to quantify potential differences in socio-economic status of rural settlements and analyse their spatial patterning at different spatial scales. The methodology presented can be transferred to other study regions and is open to many more analytical approaches, such as network analysis. It is however limited by the level of detail of the archaeological observations, in particular where it concerns chronology and find materials. Metal finds, for example (Heeren and van der Feijst 2017; Kars and Heeren 2018), would have been very useful to include, but are unevenly recorded in the available dataset.

The analysis has confirmed the assumed weak hierarchy of rural settlement in the region, but the pattern is more varied than has been suggested in literature. High-ranking settlements are relatively rare but also relatively evenly spread, and there is a widespread presence of imported ceramics and roof tiles over the area, pointing to a relatively easy access to ‘urban’ markets.

Some areas show a higher proportion of high-ranking settlements, in particular around the towns of Noviomagus and Forum Hadriani, which thus may have acted as drivers of local economic growth, whereas such a role seems less evident for the military forts and associated vici. The Kromme Rijn area seems to be the only area that may have independently developed a stronger settlement hierarchy, thus confirming the assumptions made by Vos (2009). The marginal areas, lastly, do not show much evidence for hierarchical differences. These areas were mostly settled in a later stage and thus may simply not have had enough time to become more integrated into the regional socio-economic network.

The analysis presented consists of a series of complex workflows moving from SQL queries to statistical analyses and GIS mapping. Because of this complexity, it is crucial that the data and manipulations involved are conforming to the principles of Open Science. Appendix A therefore presents the statistical analyses in Jupyter Notebook format that can also be accessed at https://github.com/LimesLimits/Settlement_hierarchy (https://doi.org/10.5281/zenodo.7716160), and all data used is stored on https://github.com/LimesLimits/Archaeological-data (https://doi.org/10.5281/zenodo.7652532).