Abstract

The Roman Empire’s road system was critical for structuring the movement of people, goods and ideas, and sustaining imperial control. Yet, it remains incompletely mapped and poorly integrated across sources despite centuries of research. We present Itiner-e, the most detailed and comprehensive open digital dataset of roads in the entire Roman Empire. It was created by identifying roads from archaeological and historical sources, locating them using modern and historical topographic maps and remote sensing, and digitising them with road segment-level metadata and certainty categories. The dataset nearly doubles the known length of Roman roads through increased coverage and spatial precision, and reveals that the location of only 2.737% are known with certainty. This resource is transformative for understanding how mobility shaped connectivity, administration, and even disease transmission in the ancient world, and for studies of the millennia-long development of terrestrial mobility in the region.

Background & Summary

The study of roads of the Roman Empire is a centuries-old pursuit1,2. There is a wealth of information about roads that were physically identified in archaeological excavations and surveys, about milestones which were placed at regular intervals along Roman roads, and historical sources like the Antonine Itinerary3,4 or the Tabula Peutingeriana5, describing major connections between settlements, as well as detailed regional summaries on Roman roads6,7. However, finding and locating this diversity of research and the spatially precise locations of the roads themselves is inhibited by a lack of an Empire-wide synthesis and digitisation.

There have previously been very few other open digital representations of our knowledge of roads over the entire Roman Empire. The Barrington Atlas of the Greek and Roman World8 is the main reference atlas for the ancient world, and the roads recorded in it are openly digitally available from the Ancient World Mapping Center (https://awmc.unc.edu) and Mapping Past Societies (formerly the Digital Atlas of Roman and Medieval Civilizations, https://darmc.harvard.edu/). These two digitisations are highly similar and differ only in a few areas where the latter also includes roads from the Tabula Imperii Byzantini for Greece and Asia Minor9,10,11,12,13,14,15,16. They have proven invaluable resources for spatially explicit computational studies of ancient trade and mobility, and as references for the overall structure of the Roman transport system17,18. However, both are of limited spatial detail as they reflect the spatial resolution of the Barrington Atlas (ranging from 1:500,000 to 1:1,000,000), which results in disregarding natural corridors and barriers. Moreover, the sources used to create the roads included in the Barrington Atlas are only available as a short list of references for all information in an atlas map sheet, and not on a road-by-road basis. The sources and methods used to include roads in the Barrington Atlas and trace their particular paths are not documented, and a quantitative assessment of how representative and reliable this resource is is not possible.

Itiner-e (dataset: https://doi.org/10.5281/zenodo.17122148, online platform: https://itiner-e.org/) addresses this gap in resources for studying the ancient world. It presents an open, spatially high resolution digitisation of Roman roads, drawing on published historical and archaeological information, topographic maps, and remote sensing data, citing the sources used to create each road segment, and making each road segment uniquely citable through a URI (Uniform Resource Identifier) that is linked to URIs of nearby ancient places in the Pleiades gazetteer of ancient places (http://pleiades.stoa.org/). Itiner-e covers the area of the Roman Empire at its maximum extent, ca. 150 CE, and includes any terrestrial route with an evidenced, conjectured, or hypothesised location in the sources used for the study. This includes any road predating Roman conquest that continued to be in use during Roman times.

The resulting map includes, in total, 299,171.31 km of roads in an area of nearly 4,000,000 km2, which is nearly double the length of other resources (the Digital Atlas of Roman and Medieval Civilizations road dataset is 188,555 km). This increase is due to a higher coverage of roads (Figs. 1b, 2a; e.g., the Iberian Peninsula, Greece, North Africa), but also by the decision to make a spatially explicit dataset that adapts routes to the geographical reality (i.e., to cross a mountain, our roads follow a winding pass rather than a direct line), resulting in a higher number of vertices for the road lines and higher total length (Figs. 1c; 2b). The dataset includes 14,769 road segments and consists of 14 connected components, reflecting the contiguous landmasses (including continental Europe, Britain, Asia–North Africa, and various Mediterranean islands). The core of the Roman terrestrial transport network were those roads classified as main roads (see Main/Secondary Roads) documented via milestones or historical sources. Our dataset reveals these covered 103,477.9 km (34.58%), and this number is unlikely to increase significantly, since these are the most documented and studied Roman roads. The dataset enables studies of how these main roads structured and enabled phenomena including Roman conquest and administration. Our dataset further includes 195,693.3 km (65.42%) of secondary roads which can be studied to understand the structure of more local mobility, and future data collection focusing on understudied areas could significantly increase this number. The dataset reveals for the first time that only 2.737% of the spatial location of the total road length is certain, whilst 89.818% is conjectured and 7.445% hypothesised (see Certainty categories). This shows a discrepancy between our knowledge of the existence and location of Roman roads: we know all of the included roads were used at some point during the Roman period, but their precise locations are not certain. Our confidence maps (see Technical validation) visualise this uncertainty and elaborate on it by expressing differences across the dataset in road density, spatial accuracy of road digitisation, and source reliability.

Fig. 1

(a) The Itiner-e road dataset, and two comparisons with the DARMC dataset: (b) difference in coverage, and (c) difference in spatial resolution as expressed by the number of vertices contained in the datasets. Itiner-e increases the coverage in selected areas throughout the Empire, most notably in the Iberian Peninsula, the Aegean and Egypt, and increases spatial detail throughout with a few exceptions marked in blue in c).

Fig. 2

Comparison of DARMC (in orange) and Itiner-e (in black) datasets, (a) showing an example from France of a region with increased coverage of roads, and (b) an example of increased spatial detail.

A major challenge is the absence of chronological evidence of the creation and change of roads, that is comparable at an Empire-wide scale. This means the current dataset cannot show the growth and change over time of Roman roads and the degree to which they built on and reused previously existing roads. Historical and archaeological sources teach us that transport networks grow organically, new roads are constructed on top of old ones, they change function and physical characteristics, and become disused. Detailed temporal evidence for road construction, use and change is only available for a handful of cases, making an evidence-based reconstruction of how the road system changed throughout the Roman period at an Empire-wide scale currently impossible. This should be the subject of dedicated large-scale efforts in future research.

Itiner-e makes such gaps in our current knowledge of Roman roads explicit for the first time, allowing this information to inform robustness tests and uncertainty quantifications of future studies, and specifying how and where future data collection efforts can improve our knowledge.

Methods

Road digitisation followed three steps (Fig. 3), with regional variation depending on the availability of sources and intensity of research activity, as well as collaborators (see Regional overviews): (1) identifying roads through relevant literature, historical and archaeological sources, gazetteers and milestones; (2) locating roads using modern and historical aerial photographs, modern and historical topographic maps, and modern and historical satellite imagery; (3) digitising roads using GIS, cross-checking with sources, and adding attribute information.

Fig. 3

Workflow summarizing the data collection and digitisation process.

The first step consisted of identifying roads in sources and those revealed in prior studies within the research area, based on previously published atlases, regional summaries, surveys, milestones and excavations. Previous studies identified roads through a set of historical and archaeological parameters19,20, often based on a combination of historical and archaeological sources, diachronic studies of the landscape and the geographical characteristics of possible paths. A key factor in all prior research was the inclusion of roads on ancient itineraries such as the Antonine Itinerary or the Tabula Peutingeriana. The location of milestones, and the location of cities or settlements close to each other also allowed researchers to assume the existence of roads that connected them, since all ancient places must have been reachable by a terrestrial route. But only rarely have long stretches of Roman roads been excavated or survived to the present day. Key overviews of roads of the entire Roman Empire include the Barrington Atlas8, and its digital representation by the Ancient World Mapping Centre and by Mapping Past Societies, as well as the Tabula Imperii Romani (https://tir-for.iec.cat/), and summaries of the Roman road network in specific regions, provinces, countries or extensively surveyed areas (discussed per region in Regional overviews). This phase further included collecting and summarizing relevant archaeological and historical sources and publications for each region. Milestones were used as an important source to determine points where roads passed through, and for chronology, as they were placed along the roads to indicate the distance to ancient places and sometimes named infrastructure investments by emperors. Milestone data were collected primarily using the geocoded database LIRE (Latin Inscriptions of the Roman Empire)[21](https://www.nature.com/articles/s41597-025-06140-z#ref-CR21 “Heřmánková, P., Kaše, V. & Sobotková, A. LIRE (v1.0.0) [Data set]. Zenodo https://doi.org/10.5281/zenodo.5074774

(2021).“), which includes 8,388 milestones with Latin inscriptions. While milestones with Greek inscriptions are also included in LIRE, they are not fully represented. The drawbacks of the sources used by LIRE include the absence of anepigraphic milestones, the imprecise or incorrect location of some milestones and missing or unknown chronological data. This necessitated case-by-case corrections by consulting additional region-specific sources of the milestone data (see per region in Regional overviews). Furthermore, milestones were not evenly distributed throughout the Roman Empire and their explanatory potential is limited mainly to the major public roads (viae publicae)22. Additional data about ancient places, sites and settlements was consulted from Pleiades, which is the gazetteer of ancient places, and Vici.org, the archaeological atlas of antiquity (https://vici.org/). The archaeological, milestone, and site data were used to identify the existence of individual roads and the general layout of the road network in each region.

In step two, we proceeded to spatially locate identified roads. The specific information obtained for each road in the previous step (location of settlements, location of extant remains from survey or excavations, location of bridges, milestones, road stations, etc.) was subsequently compared to different planimetric sources such as modern and historical aerial photographs (USAF 1950s photogrammetric flights[23](https://www.nature.com/articles/s41597-025-06140-z#ref-CR23 “de Soto, P. Mercator-e Project. Mercator-e Project https://fabricadesites.fcsh.unl.pt/mercator-e/

(2016).“)), modern satellite imagery (ESRI World Imagery24, Google Satellite25) and historical satellite imagery (Corona mission[26](https://www.nature.com/articles/s41597-025-06140-z#ref-CR26 “Center for Advanced Spatial Technologies, University of Arkansas & U.S. Geological Survey. Corona Atlas & Referencing System. https://corona.cast.uark.edu/

.“)), and modern and historical (19th–early 20th century) topographic maps (see a detailed overview of the sources for each region in Regional overviews). Available topographic maps were scanned and georeferenced as needed. This enabled locating the roads in the real world.

In step three, each road section was manually digitised with a target spatial resolution of 5–200 m on a Geographic Information Systems (GIS) platform and stored as a single data record in a vector layer with an associated attribute table. A number of general principles or observations were followed to guide the digitisation work. In ideal cases, ancient roads can be visually identified on satellite imagery, aerial imagery, or are marked as such on topographic maps (Fig. 4a–g). These usually appear as straight linear features in flat areas (Fig. 4b), or as sharp switchback roads on slopes (Fig. 4c). Roman land divisions (centuriation), which include the construction of roads in an orthogonal layout, are often preserved in modern networks of roads and agricultural fields, especially in northern Italy27 (Fig. 4d), but also elsewhere (e.g., Tunisia28 and Syria29). All such identified features were cross-checked against the primary publications and additional datasets (milestones and site data) where applicable. However, in other cases where roads were not visible in imagery, the digitised road sections follow a line connecting known archaeological road remains, milestones, and ancient sites. The digitised paths of such segments are refined by comparing them to existing contemporary and historical roads on georeferenced topographic maps and satellite imagery. It is then possible to approximate the layout of a Roman road to one of the existing roads to achieve a more realistic road layout that takes into account local topography, especially in areas where more formidable obstacles are present (wetlands, rugged terrain, mountain passes, etc.). In many of these cases then, the conjectured line of a Roman road follows the line of a modern road, which is often corroborated by additional archaeological and historical evidence. The only modern roads which were excluded from consideration for corroboration with Roman roads were modern highways/expressways, which in most cases do not conform to any previous historical roads. This process results in a plausible digitised road course recorded as ‘conjectured’ or ‘hypothetical’ (see Certainty categories), which is informed by Roman remains and prior research whilst conforming to the geographical context. A consistent issue across all study areas was the presence of modern water reservoirs and dams. In these cases, we relied on historical topographic maps or historical satellite imagery (especially Corona satellite photography for the Near East, from ca. 1967–1972) taken before construction of any dams to digitise the road segments (Fig. 4h–i).

Fig. 4

Examples of the road locating process. (a) A Roman road indicated as ‘Voie Romaine’ on a topographic map (Levant 1:50,000) leaving ancient Bostra (Bosra esh-Sham, Syria) due west; (b) Via Hadriana in the Eastern Desert of Egypt appears as a wide light linear feature on a darker soil on satellite imagery; (c) the Klimax pass between Argos and Mantinea (Greece) can be identified as a sharp switchback track on the mountain, partially overlaid by a modern wider track; (d) orthogonal Roman land divisions are retained by modern roads north of Padua, Italy; (e) a road drawn on the Tabula Imperii Romani Iudaea/Palaestina sheet; (f) the same area on the modern 1:50,000 topographic map of Israel, indicating ‘remains of the Roman road’ (in Hebrew), and (g) the same area on a modern orthophoto, where the Roman road appears as a straight line with sharp turns running roughly southwest to northeast across the picture. (h) An example of employing Corona satellite imagery for locating roads around the ancient city of Samosata (Samsat, Turkey) before flooding (Corona mission image DS1107-2138DA061-62), and (i) its current state.

Several parameters were calculated for each road segment and added to the attribute table. Geodesic length in metres of each segment was calculated in GIS using the WGS 1984 World Mercator projected coordinate system (EPSG:3395). Average slope in degrees along the length of each segment was calculated over SRTM[30](https://www.nature.com/articles/s41597-025-06140-z#ref-CR30 “Digital Elevation - Shuttle Radar Topography Mission (SRTM) Void Filled. https://doi.org/10.5066/F7F76B1X

.“) Global Digital Elevation Model with a resolution of 3-arc seconds. The dataset was finally cleaned to ensure consistency in terminology and topological continuity between line segments.

The work was performed as a collaborative effort between the co-authors of this paper undertaken between 2020 and 2024. The MINERVA (https://pure.au.dk/portal/en/projects/minerva-understanding-the-centuries-long-functioning-of-the-roman) project led the digitisation of the Roman road system in North Africa, the Near East, Asia Minor, and Southeastern Europe. The Viator-e (https://viatore.icac.cat/) project led the digitisation of the roads in the Western Roman Empire, excluding North Africa. The Viator-e project is a continuation of an earlier Mercator-e project, that focused on the Iberian peninsula (https://fabricadesites.fcsh.unl.pt/mercator-e/). These two projects’ efforts were supplemented with road datasets for the Eastern Desert of Egypt (ERC Desert Networks, https://desertnetworks.huma-num.fr/), Bithynia (Weissova), Rough Cilicia (Şahin), Pamphylia (Massa), Campania (Renda), the Dutch part of the Roman frontier (Verhagen), the Meuse and Scheldt basins (Bongers), and Britannia and Sardinia (Lewis). The data collection per region is described in Regional overviews.

Research area

The geographical limits of the data collection are broadly defined by the extent of the Roman Empire during the Antonine dynasty, ca. 150 CE, when the Empire reached its largest extent (Fig. 1a). The inclusion of some regions does not strictly follow this chronological framework and is rather founded on available source material and natural connections of these regions to the Roman provinces in the 2nd century CE. Therefore, the road system on the left bank of the Euphrates between Samosata and Edessa (modern Samsat and Şanlıurfa, Turkey) is included, since the exact boundary between the Roman Empire and the Edessan kingdom is impossible to define exactly. Dura–Europos in Syria is included, although it was permanently annexed around the year 165 CE. Further regions to the east (Armenia, Mesopotamia) are not included as they were annexed by the Romans only for a short time in the early 2nd c. CE. A string of north Saharan oases (Jaghbub, al-Wahat, Marada, Zillah, Waddan and Fezzan in Libya, and Touggourt in Algeria) are included – even though they were not part of the Roman Empire they were part of the Roman trade network in North Africa. The region between Hadrian’s Wall and the Antonine Wall in Britain is also included, even though Roman control there was only ephemeral (ca. 142–182 CE).

Chronological boundaries

CE (Common Era) and BCE (Before Common Era) are used throughout this paper and in the dataset. The lower chronological limit of the data collection is broadly defined by the first incorporation of each territory into the Roman dominion as a province (be it under the Republic or the Empire). Thus, the earliest dated road in the dataset is the Via Appia between Rome and Capua, built in 312 BCE. The lower boundary then varies on a regional basis (Table 1).

However, the Romans usually incorporated older roads into their road system, which they then improved, paved, or equipped with bridges and milestones20. Therefore, evidence for the pre-Roman road network in each region was taken into account during data collection, based on available published data. Only in rare specific cases is it possible to establish a precise date for the construction of a completely new road. For this reason, the start date of most roads in the dataset is not fixed, and in such cases the date fields are filled with the value 9999, to indicate missing data (see the next section, Data field description). The upper chronological boundary for the data collection was set at 400 CE, as there were minor changes in the Roman road system until the early part of 4th century CE. That means we also only used milestones dated up to ca. 400 CE.

Data field description

Each data record represents a road section between two ancient places or road crossings, and additional information associated with it (Table 2). Due to limited sources about many roads and limits to the scope of data collection, it was not always possible to assign all values to every data record. As a minimum, each data record contains ID, Name, Type, Citation, Bibliography, and Segment Certainty, in addition to Shape Length and Average Slope.

Each record is identified by a unique numerical identifier and a descriptive name. The name is constructed from the two origin–destination place names of the digitised section, connected by a hyphen ‘-’. The ancient place name is used when known. The present-day toponym was used instead in cases where the name of the ancient settlement was unknown or did not exist. A characteristic geographical element was chosen as a place name in cases where the road ended without the existence of a settlement.

Further attributes are included for each road section related to the chronology of its creation and abandonment and, if known, the name of an emperor or a magistrate under whose mandate the road was built. Few roads can be dated with certainty, and it was outside of the scope of our data collection to obtain chronological information for all roads. These fields will be especially relevant to capture future new findings about road chronology, when the attributes will enable us to visualise the evolution of the Roman road system, or to compare the infrastructure investment of each Roman emperor.

Additional fields describe the type of road in the road system hierarchy (‘Main Road’ or ‘Secondary Road’), the creator of each data record, bibliography, field identifying the ancient name of the road or itinerary it is part of, and segment certainty (‘Certain’, ‘Conjectured’, and ‘Hypothetical’). Additional fields describe physical properties of the road segments – length (in metres), and average slope (in degrees). Each entry also has a URI on the live itiner-e.org version of the dataset, which enables linking to other online resources.

Definition of road section/record

For the purposes of the data collection, a road was defined as any line of terrestrial communication connecting sites existing within the defined geographical and chronological boundaries. This definition includes both formal (built, engineered) and informal (non-built) roads, i.e., both paved and unpaved roads which are in regular use as an accepted line of communication (e.g., desert camel tracks). Every inhabited place in the Roman world must have been reachable within its continental or island terrestrial context by a path of some sort, but road digitisation was not attempted in cases where no information was available for tracks and pathways leading to an inhabited place.

In the dataset, all roads are represented as line (vector) segments, where each segment corresponds to a single data record. A road segment spans from an intersection with another segment to the next intersection with a different segment, or from a starting point to the nearest intersection. This means that a single named road (e.g., Rome–Capua) will be split into multiple segments as the road is intersected by other roads along its course. These multiple segments will share many attributes (name, type, author, etc.) but not all of them – Shape Length and Average Slope are almost always unique, and the FID is a unique identifier for each record.

Main/Secondary Roads

A basic road hierarchy is provided by designating each road segment as either a ‘Main Road’ or ‘Secondary Road’. The assignment of these categories to the road segments is for the most part based on prior designations by archaeological and historical publications used in the digitisation process (see further below). We used the following principle in cases where prior designations were lacking. A road segment is defined as a ‘Main Road’ if it has more than one of these characteristics: (a) the presence of milestones, (b) is part of an ancient Itinerary (chiefly the Antonine Itinerary and Tabula Peutingeriana), (c) it shares (a large part of) its course with a historically known major road indicated on 19th/early 20th century maps. The remaining roads were classified as a ‘Secondary Road’. This two-tier hierarchy is satisfactory for most of the study area, although in several regions with detailed road data an additional third tier could have been used (e.g., the region of Sicyon in Greece, Sagalassos in Turkey, Durocortorum–Reims in France, etc.), but due to the limited extent of these regions this was not implemented.

Certainty categories

The segment certainty specifies the spatial accuracy and confidence in the digitised location of the road segment. Three values are defined: ‘Certain’, ‘Conjectured’, and ‘Hypothetical’. ‘Certain’ designates well-documented segments in our sources that were digitised with high spatial accuracy (less than 50 m deviation in the mountainous terrain, less than 200 m in the plains). Most roads fall into the ‘Conjectured’ category, i.e., identified road segments with lower spatial accuracy due to lower level of documentation in our sources. ‘Hypothetical’ is reserved for identified but not located roads, or identified roads where the physical infrastructure of the roads was less fixed or where multiple parallel tracks might have existed (e.g., desert areas, flood plains). In addition, ‘Hypothetical’ is used for roads which are speculated to have existed in antiquity but insufficient evidence exists to classify them as either ‘Certain’ or ‘Conjectured’.

Regional overviews

The following section provides an overview summarised by region of the main sources used in the processes of identifying and locating roads and in specific cases methods used in the process of digitising roads.

North Africa

For Morocco, the main source was the Barrington Atlas31, supplemented by a few other studies32,33. Except for a road intersection at the site of ancient Volubilis (Fertassa) which is visible on the satellite imagery, all roads are only conjectured, following existing roads. The primary topographic map used for digitisation was a 1:200,000 map from the Army Map Service34. In comparison to previous datasets, this work resulted in improving the spatial accuracy of the road data.

Algeria, Libya, and western Libya (Tripolitania) is covered by a map by Pierre Salama6. This source was supplemented by a study on the region of Sitifis (Sétif)35. Data on ancient sites from Pleiades and Vici were corroborated with the Atlas Archéologique d’Algérie36. Milestone data in LIRE were cross-checked against the Corpus Inscriptionum Latinarum37, but their interpretation and the assignment of construction dates to roads was found to be beyond the scope of the present research and only a few roads were unambiguously dated. The data for Tunisia is more detailed, with up-to-date research on several roads (Carthage–Theveste, Capsa–Tacape, etc.)38. Further roads could be digitised using an atlas of Roman land divisions (centuriation), which marks several Roman roads on modern 1:50,000 topographic maps28. A number of these roads could have been digitised with high accuracy as it was possible to visually identify them on modern satellite imagery. The majority of roads tend to follow a variety of modern roads, field roads, and tracks marked on topographic maps and corroborated on modern satellite imagery. Primary topographic maps used for Algeria were 1:50,000, and 1:200,000 Army Map Service maps, and for Tunisia, 1:50,000, 1:100,000, and 1:200,000 Army Map Service maps39,40,41,42,43. In comparison to previous datasets, this work resulted in improving spatial accuracy across the region, and increased coverage in northern Tunisia, the desert region, and Algeria.

The primary source for Libya was Mattingly44 and two sheets of the Tabula Imperii Romani45,46, and the outlines of the main Roman roads and their milestones are provided in Inscriptions of Roman Tripolitania47 and Inscriptions of Roman Cyrenaica48. A study of roads in the hinterland of Lepcis Magna (el-Khoms)49 and Cyrene50 provided detailed data that could be corroborated with modern satellite imagery and topographic maps, resulting in high accuracy of digitisation for the two cities. The road network for the Fezzan region was reconstructed based on the Archaeology of Fazzan51. Roads around Bani Walid, Misrata District, mostly follow wadi beds, along which settlements and water sources are concentrated, occasionally crossing rocky plateaus where indicated on modern topographic maps. However, their exact course, as with other desert roads, is hypothetical. Coastal roads in both Libya and Tunisia avoided sabkhas – seasonally inundated coastal salt pans not suitable for travel. Principal topographic maps used for coastal Tripolitania and Cyrenaica were 1:100,000 maps produced by the Army Map Service52,53. The desert area was covered by 1:250,000 maps by the Army Map Service54. In comparison to previous datasets, this work resulted in increased spatial accuracy and increased coverage in Tripolitania and the city of Cyrene.

In Egypt, many desert routes were identified on both modern topographic maps and satellite imagery, following the approach to the identification of caravan routes outlined by Bubenzer and Bolten55. A specific problem is presented by the Nile Valley and its delta, which traditionally relied primarily on water transport. Both the Barrington Atlas56,57 and the Tabula Imperii Romani58 suggest a land route along the Nile and the delta, based on ancient sources (the Tabula Peutingeriana and the Antonine Itinerary); therefore, these roads were included. However, due to the regular flooding of the Nile and numerous waterworks in the valley (irrigation channels, dykes), the location of these roads is very uncertain and they were likely located closer to the edges of the Nile Valley (such as modern roads). The main reference work used for the Western Desert and Sinai was Paprocki59. Primary topographic maps used for Egypt were 1:250,000 maps by the Army Map Service54. In comparison to previous datasets, this work resulted in increased spatial accuracy and increased coverage in the Eastern Desert, Western Desert and the Nile Valley.

The road system in the Eastern Desert included in our dataset was developed by the ERC Desert Networks project (https://desertnetworks.huma-num.fr/). Our knowledge about these routes came from the Tabula Imperii Romani (sheet 36 Coptos)58, supplemented by regional surveys60,61 and archaeological work on the stations of the Coptos–Myos Hormos road62 and the Coptos–Berenike road63. However, the roads themselves and their exact routes were not well known as there are, e.g., no milestones. To overcome this lack of data, the Desert Networks project has adopted an original, semi-empirical approach, combining 1:50,000 scale maps from the Egyptian Survey Authorities, a DEM with a spatial resolution of 11 m, a corpus of 288 archaeological sites with precise locations, and 60 reconstructed itineraries by modern travellers (18th–19th centuries) who travelled through the desert in conditions comparable to Roman ones (environment, transport means and logistics)64. These data were used to calibrate various factors involved in camel caravan travel and to validate the least-cost routes calculated between sites. The modelled network was then used to generate routes, taking into account transport infrastructure, navigation conditions, difficulties of the terrain and the topographical constraints specific to camels. The Roman roads were reconstructed by calculating the least-cost paths65,66 between the known Roman road stations according to ancient sources (the Tabula Peutingeriana and the Antonine Itinerary, and ostraca found at the stations67). The occupation of most of them is well-dated thanks to recent excavations. It was thus possible to refine the route and propose an itinerary for each 50-year period between 1 CE and 300 CE, allowing us to consider the evolution of the network over time. The ERC Desert Networks project data were adapted to our data structure (removal of multiple overlapping segments, merging of segments diverging less than 200 m from each other, adjustment of start and end point vertices to archaeological sites).

The Near East

The principal source for the southern Levant (Israel, Jordan, Palestinian Territories, southern Syria) is the