An Example of Geographic Network Analysis: The Case Study of the Fortore Valley (Molise and Apulia, Italy)

1 Introduction

The middle Fortore Valley, located between the provinces of Campobasso and Foggia (central Italy), serves as the administrative boundary between the regions of Molise and Apulia (Figure 1). The human and natural landscape of this extensive area, today as probably in antiquity, is as various and articulated in both geographical aspects and in both settlement dynamics. The orography is characterized by hills that slope towards the coast. However, the mountains are modelled in various forms in the Molise area, while the Dauni Month, i.e. the physical limit between Molise and Foggia, is characteristically Apennines in form (Rizzi, Orsino, & Ingaramo, 2008). On the other hand, population centres, mostly small towns, are arranged on the edge of the flat Tavoliere area along the foothills of the Dauno subalpine or on the Molise hill system. Instead, the waterscape hydrographic basin is characterized by valleys and narrow channels, determined by the many tributaries of the Fortore. Then, in the late 1950s, the earth dam, known as Occhito, was built in this sector of the valley.

Figure 1

Topographical overview of the Fortore Valley.

The key role of the Fortore River, or Fertor (the ancient name), is highlighted already in ancient literary sources. Indeed, these describe the Fortore River as widely navigable and its valley as rich in landings (see Plinius, Naturalis Historia, III, 103). Moreover, modern literary sources highlighted the peculiar nature of the area, which is part of an “Oscan-Daunian” enclave (Corrente, 2016; Torelli, 1992). According to Antonacci Sanpaolo (2000), this term indicates the existence of some groups with an Oscan-Sabellian culture (Stek, 2021; Tagliamonte, 2005) that coexist with the Daunian culture (Mazzei, 2010, 2015). Although this interpretation has been partly revised, recent archaeological research points to a significant trail of Italic (Samnite) groups in the Apulian area (Marchi, 2016). In the sixth century, in Apulia, Samnite presences marked by a different funerary ritual were already detected. During the fifth century BC, the phenomenon becomes more stable, to take on a regular population connotation by the fourth century BC (Marchi, 2021). Indeed, several urban centres located along the Fortore River, which flourished during the eighth to sixth centuries BC. The data provided by the coastal sites (Termoli, Larino, and San Paolo di Civitate) about this chronological horizon suggest a complex socio-economic stratification (Di Niro, 1981). These sites, moreover, have a mixed agricultural and pastoral economy (Di Niro, 2004; Pelgrom & Stek, 2010). However, short- to medium-range trade is also not ruled out, which is attested by the Daunian geometric pottery or bucchero ware (Di Niro, 1991; Faustoferri, 1991).

In the second period, until the Romanization (first century BC) of the area, the Samnite presence was consolidated as the case of Teanum Apulum (San Paolo di Civitate) or Larino testifies (De Benedittis & Santone, 2006, p. 15; Marchi, 2000, pp. 240–241). In this debate, an analysis of the evidence located in Macchia Valfortore (Campobasso) and Carlantino (Foggia) is proposed as a case study, characterized by a typically Oscan-Daunian culture (Gravina, 2007; Muccilli, Colombo, & Santone, 2021; Muntoni, Frangiosa, & La Trofa, 2022; Naso, De Simone, & Esposito, 2022).

This article highlights, through a non-conventional archaeological approach, the role of the Fortore River Valley in the past and how the settlement patterns examined support the cultural enclave theory. Furthermore, the use of several spatial analyses, including the point patterns and spatial networks analysis, aims to identify the type of settlement pattern and clarify the dynamics of interchange between the two districts. Finally, the conclusions suggest how river connections relate to the terrestrial road system, demonstrating the importance of river connections for the economic and social development of the area.

2 Data and Methods

2.1 Data

The data employed in this research are obtained from several sources, as defined legacy data (Witcher, 2008). In particular, the data on the Macchia Valfortore area were collected as part of the “Archaeological Map of the Fortore Valley” project. The project was a systematic field survey of the middle valley of the Fortorer river (near Occhito Dam) (De Simone, 2021; Naso et al., 2022). Instead, concerning the Carlantino area, the sources are based on literature (Gravina, 2007; Muntoni et al., 2022) and the “Archaeological Map of Apulia” (CartApulia).^[1] The data processed in the GIS software (QGIS) were collected in a vector layer of a SQL geodatabase^[2] (Figure 2).

Figure 2

The data employed in this research: (a) the dataset; (b) principal locations listed in the text; (c) the areas about Archaic-Late Classic period (sixth century BC to fourth century B.C. code PRM1); (d) the areas dates to the Hellenistic–Samnite period (fourth century BC to first century BC code PRM2). Base map DEM.

Moreover, more data were added to the GIS project. Specifically, it has digitalized the ancient route system based on published studies that combine archaeological and topographical data (e.g., Alvisi, 1970; Gravina, 2007). In addition, the topography data are collected from the National Institute of Geophysics and Volcanology website, such as the Digital Elevation Model (DEM) (TINITALY model, see Tarquini et al., 2007), with a resolution of 10 m. The data were then processed by means of SAGA GIS tools to achieve the shaded relief map, slope map using 2.5σ clipping as a technique for image enhancement, the hydrographic channel network, and drainage basins.

3 Data Processing

A popular point analysis is nearest neighbour analysis, which calculates the distance separating each point from the nearest neighbour according to the Euclidean distance expressed by a numeric index R (Bailey & Gatrell, 1995, pp. 90–91). A change of scale, however, would put this model in “crisis.” Multi-spectral PPA models, on the other hand, use the K Replay function to represent aggregation and disintegration phenomena at a different scale (Palmisano, 2013). In this way, an accurate assessment of how the settlements are related to each other and how dispersive (e.g., regular distributions of farms) or agglomerative (e.g., villages) processes underlying the settlement system were identified. Consequently, the point distribution of evidence was analyzed both through the R-index and Ripley’s K function.

To improve the PPA analysis, the density cluster analysis is carried out.^[3] Density estimates of different types are considered as the count or frequency of a group of elements in a specific area. The one used here is called the KDE model or Kernel Density Estimation (Conolly & Lake, 2006, pp. 175–177). This is a non-inferential statistical technique in which several mathematical functions return a two-dimensional density estimate (the kernel), which according to fuzzy logic is based on the distance between the centre of the point and its outside (D’Andrea, 2006, pp. 61–75; Nakoinz & Knitter, 2016, pp. 77–81). The Kernel Density Estimation showed 23 areas featuring a significant density. The archaeological evidence in these areas dates them to the Hellenistic–Samnite period (fourth century BC to first century BC code PRM2), while nine of them have yielded evidence dating to the Archaic-Late Classic period (sixth century BC to fourth century BC code PRM1)^[4] (Figure 2). Consequently, the pattern of density analysis output is analyzed both through the R-index and the Ripley’s K function.

If, through PPA, it was defined to explain settlement patterns and territorial organization, the small-scale spatial networks analysis, however, helps to explain the interchange dynamics between the different settlements, and it testifies to the primary role of the Fortore River.

To define the network model, it was first identified based on Delaunay’s graph as a set of nodes.^[5] The graphs, like the defined nodes and edges, are in geometric space, and network topology (the structural arrangement of network elements) is at least partly constrained by the spatial relationships among them (Brughmans & Peeples, 2020).

However, to improve the analysis and have a more challenging model, it was customized as the so-called the minimum spanning tree (MST) graph (De Smith, Goodchild, & Longley, 2007).^[6] In this case, the links are more realistically edited, where “a set of nodes in the Euclidean plane, edges are created between pairs of nodes to form a tree where each node can be reached by each other node, such that the sum of the Euclidean edge lengths is less than the sum for any other spanning tree” (Brughmans & Peeples, 2020, p. 281). In some studies, the MSTs are used for the representation of least costly paths between places (see Herzog, 2013). In addition, to achieve improvement in the information potential of the graph, a weighted MST is produced based on the cost surface processed from the topographic data (Herzog, 2020) (see par. 2.1). The surface obtained takes into consideration the anisotropic frictions, i.e., the cost is established considering as an incident factor in the difficulty of movement some topographical elements. The hydrographic network and its direction are evaluated as a further cost element, with the assumption that swampy routes and frequently flooded areas would be avoided as it would increase the cost of travel (Casarotto, De Guio, & Ferrarese, 2009; Wheatley & Gillings, 2002, pp. 137–147). The movement within the natural environment to find resources or meet a need can lead to the formation of an organized landscape.^[7]

Furthermore, in accordance with the research objective, the structure of the spatial network was analyzed using Freeman’s indices (Freeman, 2004; Nakoinz & Knitter, 2016, p. 189). Freeman’s indices express the notion of network centrality. The measure of centrality can be stated as the importance or prominence of a node in directing and receiving different types of flows through a network. This is referred to as a node-level property, but most centrality measures can also be aggregated to define the centrality of the entire network (a graph-level measure of centrality concentration among actors) (Peeples & Roberts, 2013). In the research, the degree and betweenness centrality are used. The value of degree indicates the extent to which an actor is connected to all other actors in the network and specifies how easily information can reach it. It assumes that the greater the number of ties and neighbours an actor has, the greater its probability of receiving information. The measure of betweenness defines the ability of an actor to act as a mediator in the passage of information between two other actors who are not neighbours, and thus to control their relationship (Fulminante, 2012; Nakoinz et al., 2021).

The research workflow and data processing are as follows:

1. The development of a geodatabase: The archaeological, topographical, and other available data were collected in an SQL geodatabase. It uses the “Spatialite” suite to manage, edit, and query the dataset.

2. PPA: The R indices and the K Replay were calculated about the entire dataset. The analysis is carried out in PAST software (Hammer, Harper, & Ryan, 2001), a powerful set of tools for quantitative and statistical analysis. The output results are the schematic graphs.

3. KDE: With the help of the KDE technique, areas with a significant density were obtained within the GIS software. The pattern of density analysis output is analyzed both through the R-index and Ripley’s K function (see ii).

4. Spatial network design: In particular, Delaunay’s graph and the MST graph are defined. GRASS GIS and the “v.net” tools were used.^[8] This allows users to create, maintain, and analyze the spatial network.

5. MST graph and cost surface: The weighted MST is produced based on the cost surface raster. This is generated in according to the “raster calculator” language on QGIS software. Indeed, the weighted MST model (a vector output) is performed in the Geo-MST plug-in for QGIS (Çalışkan & Anbaroğlu, 2020).

6. Analysis of the structure of networks using Freeman’s indices: In particular, this research considered the degree and betweenness indices. The indices are calculated in “v.net.centrality” tool of GRASS GIS.^[9]

4 Results

Considering the results of the point pattern analysis, a certain observation can be considered. As a result, both the nearest neighbour analysis approach (R Index) and the multi-spectral PPA models (K Replay) revealed over the entire dataset a rough aggregation pattern (Figure 3). Analyzing from a different point of view, for the first period (PRM1), a dispersed pattern according to both the Ripley K-function and the R-index is revealed. For this time, however, the Ripley K-function would indicate a regular distribution above 1,800 m, which would lead one to assume at least for this scale an aggregation-type distribution (Figure 4). Instead, for the second period (PRM2), a tendency towards dispersion would be confirmed according to both analyses (Figure 5).

Figure 3

The PPA on the entire dataset: (a) nearest neighbour analysis approach (R index) and (b) the multi-spectral PPA models (K Replay).

Figure 4

The PPA on the Archaic-Late Classic evidence (PRM1): (a) nearest neighbour analysis approach (R index) and (b) the multi-spectral PPA models (K Replay).

Figure 5

The PPA on the Hellenistic–Samnite evidence (PRM2): (a) nearest neighbour analysis approach (R index) and (b) the multi-spectral PPA models (K Replay).

These results would be confirmed, at least for the first period, by comparing known data in the literature. Indeed, between the end of the eighth century BC and the end of the classical age (fifth to fourth centuries BC), the number of villages in the investigated territory likely contracted, as occurred in the localities of Masseria San Nicola, Monte San Giovanni, and Serra Fullona (Figure 2).^[10] In both the internal and coastal areas of Molise, scholars identify the archaic communities as occupying small villages primarily engaged in an economy focused on the exploitation of primary resources (Di Niro, 2004; Pelgrom & Stek, 2010). As evidenced by a mixed agricultural–pastoral economy and different kinds of ceramics (e.g. “ollae” and “dolia”) (Di Niro, 1991). In addition, the evidence from the cemetery in the Cigno and Santo Venditti localities (Bernardini, Naso, Olivieri, & Raccar, 2008; De Benedittis & Santone, 2006; Marchi, 2016; Muccilli et al., 2021) could suggest an increase in the level of occupation of the territory at the end of the first period (fifth century BC to fourth century BC), which could be indicative of a more capillary occupation of the territory and could be related to a dispersed pattern. This phenomenon becomes clearer in relation to the structuring, from the fourth century BC, of an Oscan-Daunian enclave (Section 1). In respect of this, the constructed spatial network model would help to understand how the different areas interact with the surrounding landscape and the hydrographic network of the Fortore River. Taking into account the MST graph and the cost surface model (cf. Section 3), the least cost paths between Hellenistic and Samnite areas (PRM2) follow the hydrographic network with altitudes above the sea level of no more than 400 m (Figure 6). This would not appear to be valid in the case of the Monte San Giovanni area, where the path would follow a hill ridge 500 m above the sea level.^[11] This pattern does not correspond to the Archaic-Late Classic period (PRM1), where the Fortore crossings would seem to be preferred near the area of Serra Fullona.

Figure 6

The weighted MST based on the cost surface raster: (a) the least cost paths between Hellenistic and Samnite areas (PRM2) and (b) the least cost paths between Archaic and Late Classic areas (PRM2).

However, by analyzing the structure of Freeman’s index network, it is possible to make some other observations. The nodes are classified based on the centrality index, specifically the betweenness and degree measures. Consider that the sites of the Archaic-Late Classic period (PRM1) near the banks of the Fortore River, in particular at Serra Frullona and Serra della Guardia, play a fundamental role in transit control over the river. However, it is possible that the Serra della Guardia area (370 m a.s.l.) constitutes a control site considering their predominant position on the river. Moreover, the figure also illustrates that the cemeteries located in the Cigno and in Santo Venditti have a marginal role in the interaction system, which is different from what has been highlighted in the published archaeological literature (Figure 7).

Figure 7

The Freeman’s index about the Archaic-Late Classic data (PRM1). The picture highlights the (a) betweenness value and (b) degree value. Base map DEM.

The dataset of the second pre-Roman period (PRM2) is also processed according to Freeman’s indices (Figure 8). The centrality index provides evidence of good organization in the locality of Masseria De Marzia.^[12] However, the role of farms located on the banks of the Fortore, such as Masseria Don Lupo, Difesa delle Valli or Cigno, is more visible. Moreover, the agglomeration of Serra della Guardia and Serra Fullona declined from the prominent role it had during the early pre-Roman period while conversely gaining importance as an interaction node of the Piano San Leucio or Santo Venditti sites.

Figure 8

The Freeman’s index about the Hellenistic–Samnite data (PRM2) (PRM1). The picture highlights the (a) betweenness value and (b) degree value. Base map DEM.

Between the fourth and second centuries BC, in fact, there was likely a widespread and significant occupation of the territory, as demonstrated by an increase in farms in the area (Figure 2). An intensive field survey carried out near Cigno identified a large building divided into several areas that would normally be flooded by the Occhito Dam (Babbi & Naso, 2008). The artefacts found allow suggesting both service and public functions for the area. Furthermore, the site has considerable importance, as it was identified as the find place of an extraordinary eaves drip, representing two humans and likely derived from theatre masks; it is datable to the second century BC (Känel & Naso, 2008; Liberatore, 2011). The excavation near Carlatino also highlighted two villas (Mazzei, 1997; Muntoni et al., 2022): the first is characterized by a complex terraced layout facing the river; the latter, on the other hand, is similar in construction. In this one, two kilns were found, one of which contained ceramic material in the firing position. They are indicative of kilns, which would testify to the presence of productive district and commercial trade, called the “Fortore system” (Marchi et al., 2021). As in nearby areas (e.g., Apennines-central Italy), these housing units are associated with places of cult worship (Stek, 2009; Van Wonterghem, 1999). Evidence for this includes an Oscan inscription on a lithic slab found in the territory of Macchia Valfortore (Benelli, Monda, & Naso, 2008; Marchi, 2016, p. 56) and figurines of Hercules from Masseria Linciotti and Masseria Bellini (see Gravina, 2007).

5 Discussion and Conclusion

While literary sources highlight the peculiar nature of the area as part of an “Oscan-Daunian” enclave (Antonacci Sanpaolo, 2000), recent research indicates the middle valley of the Fortore River was part of a productive district called the “Fortore system” until nearly the Hellenistic period (Marchi, Forte, Frangiosa, La Trofa, & Savino, 2020; Marchi et al., 2021). This hypothesis is based on archaeological findings, including kilns and waste products, as well as an analysis of the environmental features of the area. The system would have facilitated trade in products typical of both the Oscan and Daunia cultures. The models and network analysis of the second pre-Roman period may support the existence of such a system. This model might be valid for the Hellenistic–Samnite period, where there is more archaeological data, but it is less clear for the early pre-Roman period. The proposed spatial analyses demonstrate the role of the Fortore as more of a link than an obstacle, in accordance with ancient literary sources (cf. Section 1). On the other hand, the interaction between river and overland routes is relevant. Starting from studies on the Apulian area, such as Giovanna Alvisi’s research (Alvisi, 1970; Gravina, 2007), it is possible to suggest an ancient pathway system (Figure 9). If the network links of MST produced during the network analysis are considered a pathway, it is possible to achieve a “complete” road system of the middle Fortore Valley and suppose an integrated local transportation system characterized by fluvial and terrestrial connections. In the future research, the current observations will be integrated and verified with the improvement of the proposed models and the addition of relevant elements (e.g. studies of material culture and specific paleo-environmental research).

Figure 9

A proposal of local transportation system based on literature data (Alvisi, 1970; Gravina, 2007) and spatial network analysis results. Base map DEM.