Living in a fringe environment: three Late Roman settlements in the Kharga Oasis (Egypt's Western Desert)

Abstract This article presents a comparative analysis of three Late Roman sites located at the northern outskirts of the Kharga Oasis in Egypt's Western Desert: Umm al-Dabadib, Ayn al-Labakha, and the Gib/Sumayra Complex. These were part of the district of the Oasis Magna, which included the oases of Dakhla and Kharga. An analysis of their layout, including both shape and extent, is followed by an evaluation of their absolute and relative positions. These data are then compared to the administrative and historical contexts within which the three sites flourished. Both administrative and economic aspects are considered, as well as the presence of the army. The complex picture that emerges suggests that these three sites played several roles at the same time and were part of a large-scale strategic design that encompassed not only the Kharga Oasis but the entire Western Desert.

systematically excavated. 2 The characteristic that makes them especially interesting for a comparative study is the combined presence of both the built-up area, where the community lived, and the agricultural system; the latter includes both a system to retrieve water and cultivated fields. Settlements and agricultural systems represent two faces of the same coin: each was simultaneously the cause and effect of the other. This duality is mirrored by the structure of the LIFE project, a collaboration between the Politecnico di Milano, in charge of the archaeological and architectural study of Umm al-Dabadib, and the Università degli Studi di Napoli Federico II, in charge of the study of the agricultural system. 3 Unlike a plethora of other Roman sites that punctuate Egypt's Western Desert, these three sites survive in relatively good condition, albeit to different degrees of preservation: Umm al-Dabadib is virtually intact, Ayn al-Labakha coexists with limited but persistent modern activity, but substantial portions of the Gib/Sumayra Complexwell preserved until 20 years agoare being progressively removed by the encroachment of modern cultivation, especially in the area of Ayn Gib, around Qasr al-Gib (Fig. 3). This article represents an opportunity to describe and analyze them together, before the physical evidence for the parallels among them disappears. A methodological introduction is followed by a description and discussion first of the layout of these settlements and then of their position. Corinna Rossi et al. 744 The data collected on these two aspects are then placed within the broader context of the Kharga Oasis and wider Oasis Magna. Finally, all this information is combined to discuss the possible function(s) of the three sites. A conclusion summarizes the results.

The (im)possible study
The archaeological investigation of these sites is extremely difficult. To start with, they are located away from the inhabited part of the Kharga Oasis, in areas where there is no electricity; at Umm al-Dabadib, there is no water and no telephone signal. For this reason, working there has always involved a significant logistical and economic effort, and the amount of information retrieved so far has been strongly influenced by the environmental conditions. Fieldwork at these sites has also been impossible for a number of years: the Western Desert was closed to all foreign archaeological missions in 2016 amid security concerns, and the hope that the situation might improve was brought to a fresh halt by the 2020 pandemic.
The situation is made even more complicated by the fact that these sites consist of not only archaeological remains that are acknowledged as such and are therefore protected by the law (e.g., buildings and cemeteries) but also the elusive traces of irrigation systems and fields, which are not included in the protected areas. This latter situation stems from a combination of factors.
On the one hand, these remains are generally perceived as secondary by-products of a settlement and as minor traces of basic activities. Similar difficulties are encountered in the study of ancient desert routes and quarries. 4 On the other hand, it must be said that these faint remains are often nearly or completely invisible from ground level. This is certainly the case of the ancient fields at Umm al-Dabadib, which become visible to the naked eye only for a few minutes at sunset and only under specific conditions of light, but Living in a fringe environment 745 which are perfectly apparent on Google Earth. The same applies to most of the ancient and abandoned modern fields in the Kharga Oasis. The impressive extent of the mining area located in the desert to the west of Umm al-Dabadib can also only be appreciated on satellite images, given that the presence of a thick field of interlocking dunes makes it very difficult to explore the area from ground level. That said, the opposite is also true: some features invisible from the air are visible from the ground, either because the resolution of the available satellite images is insufficient or because the features themselves simply cannot be seen from above. In the case of mining areas, for instance, although the holes left by sieving operations can be clearly detected from above, even in the case of slightly blurred images, underground galleries are mostly invisible. Therefore, the actual extent of ancient occupation (and the mining areas are a perfect example of this) can only be determined by combining information retrieved from the air and from the ground.
Despite their relatively recent appearance, aerial images (satellite and remote sensing) are already contributing to an archive documenting the ongoing changes to the Earth's surface. 5 In the case of the Kharga Oasis, until 2007 the archaeological remains were barely visible on Google Earth. Then the situation improved significantly, and one of the best sets of images dates to the years 2010-2011. The images taken in the subsequent years document the changes that occurred (see Fig. 3) but are often less "legible," due to the absence of colors in the images or their excessive contrast.
Aerial views can play an extremely important role in identifying the remains of ancient cultivation and therefore contribute to their preservation, not only because they represent an efficient tool but also because they represent a way to circumvent at least some of the difficulties listed above relating to fieldwork activities. In this specific case, they offer the chance to adopt a different methodological approach, based not on the most logical actions but on the only possible ones.
What was once true for ancient fields now applies to other remains as well. As a matter of fact, currently the only way in which we can study the archaeological remains of the Kharga Oasis (as well as those of the entire Western Desert) is remotely. The LIFE research team therefore adapted the project strategy and focused on how to retrieve fresh information on Umm al-Dabadib from a distance. This goal was achieved not only by re-examining the data collected in the past but by reworking them, thereby implementing a sort of "remote" archaeological study of the site and its surrounding area. Combining data in a flexible way represented the key to success: three-dimensional models were used to perform geometrical and metrological analyses, yielding important information on the cultural background of builders and workmen; 6 photographs taken in the recent past were digitally processed to resolve doubts cast by lack of data in the three-dimensional survey or to correct mistakes made in the past; 7 and data retrieved from similar sites were used to test fresh theories. 8 Umm al-Dabadib, Ayn al-Labakha, and the area of Qasr al-Gib and Qasr al-Sumayra were all inhabited and exploited by the Early Roman period, at a scale that is difficult to define. 9 At the beginning of the 4th c. CE, they were selected to play a crucial role in a large-scale strategic program of control and defense of the Kharga Oasis. 10 Because this article focuses on the archaeological remains dating to this historical period, for simplicity, the names of the three sites will be here used to indicate the Late Roman remains. Early Roman and pre-Roman remains will instead be specifically identified as such.
Even if some important aspects of this large-scale, ambitious operation are bound to remain obscure until new fieldwork can be performed, a significant amount of information can still be retrieved by comparing these sites from a distance. This article therefore presents the results of a remote investigation into the layout, position, and function of these three Late Roman settlements. 6 Rossi and Fiorillo 2018;Rossi and Fiorillo 2020. 7 Fiorillo et al. 2020. 8 Rossi 2019. 9 Rossi and Ikram 2018, III.9. 10 Rossi 2013.
Living in a fringe environment 747

Shape
The ancient names of the three sites that are the object of this study are unknown. 11 The modern Arabic names of Ayn Umm al-Dabadib (in its extended form) and Ayn al-Labakha refer to the presence of water in the area (ayn means "spring"); the rest of their names have unclear origins. Both places are characterized by the presence of a Late Roman compact gridded settlement delineated by enclosure walls and surrounding a fortlike building. 12 They have been given modern names reflecting these characteristics: the "Fortified Settlement" of Umm al-Dabadib, endowed with a buttressed wall along its southern side, covered an L-shaped area (ca. 90 × 120 m), in the middle of which rises the Fort; the "Gridded Settlement" of Ayn al-Labakha surrounded the central Fort on at least three sides (covering an area of ca. 80 × 50 m), following a strict orthogonal pattern (Fig. 4). Both consisted of three-dimensional mosaics of interlocking domestic units separated by narrow vaulted corridors, which may well have originally looked like giant, geometrical anthills. 13 In both cases, the contemporary aqueducts started in the surrounding hills and discharged their waters into elongated cultivation systems located at some distance from the settlements, although within a radius of 1.5 km (Figs. 5 and 6).
The Gib/Sumayra Complex is a large installation that included the two forts of Qasr al-Gib and Qasr al-Sumayra, and a vast area locally known as Maghatta. This area includes At Umm al-Dabadib, a Coptic inscription engraved on a wall of the church might contain a clue to the ancient name of the site (Rossi and Ikram 2018, 242-43). 12 For the identification of these central buildings, see Rossi 2018a. 13 Rossi and Ikram 2018, 183-88, 225-36.
Corinna Rossi et al. 748 two settlements, respectively nicknamed Two Houses (because only two buildings are still standing) and Watermelon Settlement (because it is surrounded by modern watermelon fields), that are accompanied by a scatter of cemeteries. All of these settlements are linked by a large system of underground aqueducts (Fig. 7). The Gib/Sumayra Complex is organized differently from Umm al-Dabadib and Ayn al-Labakha: of the two fortlets, Qasr al-Gib rises on an isolated rock outcrop, whereas Qasr al-Sumayra lies on relatively low ground. The latter, ruined but still standing, was accompanied by a number of buildings that are now totally engulfed by sand; their layout is unclear, but they do not seem to have been organized in a compact gridded pattern within an enclosure wall. 14 In this area, in fact, the settlements are smaller and located exactly where the aqueducts discharged their waters and the cultivated areas lay. In this case, therefore, the relationship Living in a fringe environment 749 between the inhabited areas and the fields was more direct than at Umm al-Dabadib and Ayn al-Labakha, perhaps indicating a different social and administrative organization (see below).
The structure of Umm al-Dabadib and Ayn al-Labakha may be schematized as a nucleus (the gridded and enclosed settlement) from which the population departed to work the surrounding land (Fig. 8); the structure of the Gib/Sumayra Complex may instead be compared to a clusterthat is, a network of small units each made up of an inhabited nucleus and an adjacent patch of land (Fig. 9). The element that appears to be missing at the Gib/Sumayra Complex is the overall defensive character of the settlements, and yet the presence of the same architectural elements and characteristics of both the domestic and the military buildings clearly points to a common origin for the three sites, as if they were different conjugations of the same paradigm. 15 The cluster-type settlement seen at the Gib/Sumayra Complex is more difficult to spot than the nucleus type seen at Umm al-Dabadib and Ayn al-Labakha, because this type of installationless dense and spread over an area rich in wateris more easily obliterated or absorbed by the subsequent reuse of the same water sources. In the northern Kharga, cluster-like settlements dating to the Roman period can be seen in the area of Ayn al-Tarakwa and Ayn al-Dabashiya, as well as all around the central green core that nowadays hosts Kharga  Town (see below, Fig. 12). Bilayda and the surrounding scatter of agricultural installations represent a significant example, but many other sites that are currently classified simply as "Greco-Roman" might also be part of the picture. 16 The nucleus-type settlement, such as at Umm al-Dabadib and Ayn al-Labakha, instead generally leaves behind evident concentrations of archaeological traces, given that compact settlements tend to turn into ruined mounds that are unlikely to be easily covered by spontaneous vegetation or destroyed by modern land reclamation, as is the case with the Watermelon Settlement. The main risk for this type of settlement is represented by sabbakh-diggingthe dismantling of the ruins by local farmersas has happened at Qasr al-Nissima. 17 The "flattened" remains of Qasr al-Baramudy, located on higher ground and exposed to the strong sand-laden winds that batter the oasis, might be the combined result of human actions and natural erosion (Fig. 10). 18 Walled settlements can also be found at Ayn al-Tarakwa, 19 Nadura, and Qasr al-Ghwayta (Fig. 11), located respectively to the north, east, and south of Kharga Town,   16 Rossi and Ahmed 2019. 17 Ghica 2012, 217-21. 18 Rossi and Ikram 2018, 422-25. 19 Rossi and Ikram 2018, 339-48.
Living in a fringe environment 753 but with the significant difference that they developed within and around the temena of earlier stone temples. 20 A hybrid site might be Tulayb, where a 4th-c. CE fort incorporated an earlier mudbrick temple. 21 It is tempting to conclude that the convergence toward the same shape (walled settlements clustered around a central building) was due to the general condition of insecurity that characterized the 3rd c. CE, 22 which Diocletian tried to combat with a reorganization of the country, including the foundation of a number of fortresses meant to host legions along the Nile and alae and auxiliary forces in the Western Desert. 23 This explanation makes sense and is, broadly speaking, probably correct. It must be underlined, however, that although we have some evidence for early Roman remains, we know very little about the shape, layout, and extent of pre-Roman settlements in the Kharga Oasis. 24 Nadura dates to the Early Roman period (Klotz 2013). Qasr al-Ghwayta and Hibis were founded in the Saite period and subsequently decorated by Darius I (Arnold 1999, 77-79, 92). 21 Rossi and Ikram 2018, 131-48. 22 Starting from the second half of the 3rd c. CE, the extant sources mention raids of Libyan groups in the Fayyum (P. Princ. 29), in the Herakleopolite nome (BGU III 935) and in the southern Oxyrhynchite nome (P. Oxy. XXX 2681 and P. Oxy. XLVI 3292), as well as disorders in the district of the Oasis Magna caused by the Nobates, eventually pacified by Diocletian (Procop. Pers. 1.9). 23 Rossi 2013; Rossi 2018a, 449-51; Rossi forthcoming. 24 Snape 2014, 216.
Perhaps this type of enclosed, walled, anthill-like settlement had been adopted in the harsh desert climate of the Western Desert well before the Late Roman period, not necessarily (or at least not only) for security reasons. Testing this hypothesis will be possible only by comparing the results of thorough archaeological investigations into the pre-Roman levels of occupation of the oasesa major enterprise that is still far from being undertaken.

Extent
Plotting on Google Earth the extent of the Late Roman-period archaeological remains of Umm al-Dabadib, Ayn al-Labakha, and the Gib/Sumayra Complex immediately conveys the scale of the Late Roman intervention in this region (Fig. 12). 25 When the Roman authorities decided to build a chain of new settlements at the very beginning of the 4th c. CE, the presence of earlier settlements must have played an important role in the decision-making process because their existence testified to the presence of reliable water sources. Umm al-Dabadib, Ayn al-Labakha, and the Gib/Sumayra Complex all yielded evidence of earlier occupation, even if its exact scale and extent are difficult to establish with precision. At Umm al-Dabadib, the earlier Northern Settlement (dating to the 3rd c. CE and perhaps extending back to the 2nd c. CE) consisted of scattered houses that covered an elongated area of about 450 × 180 m. 26 At Ayn al-Labakha, a pre-4th c. CE settlement (probably contemporary with the Pyris temple and the necropolis to the west) might be located immediately to the north of the Gridded Settlement, over an area of about 250 × 100 m that is nowadays totally engulfed by soft sand ( Fig. 13; see also . 27 Scattered clues indicate that the area of the Gib/Sumayra Complex was inhabited before the Late Roman period, but nothing can be said of the type and extent of the earlier settlements without extensive excavations. At Umm al-Dabadib and Ayn al-Labakha, the extent of the new 4th-c. CE inhabited areas was smaller than that of the earlier inhabited areas, but the settlements were far denser in terms of construction. Although the surface covered by the settlements shrank, the total extent of the field of action of all three settlements increased enormously due to the installation of their agricultural systems (see Figs. 5-7, 12, and 13). Overall, including water catchment and cultivated areas, Umm al-Dabadib covered an area of 6 × 2 km; the mining area in the west covers a C-shaped area with a maximum extent of 3 × 4 km. The remains of Ayn al-Labakha spread over an area of 4 × 1.5 km, whereas the Gib/Sumayra Complex had an extent of 10 × 4 km.
Estimating the number of inhabitants of the new 4th-c. CE settlement at Umm al-Dabadib is extremely difficult. Although its general layout is clear, its internal organization is not. A number of domestic units can be clearly singled out, but the rest of the constructions remain to be identified. Moreover, the total height and number of floors of several buildings are still unclear. An interesting aspect of this settlement is the apparent 25 The attribution of the subterranean aqueducts to the 4th c. CE is discussed in Rossi 2018b, 517-21. 26 The presence of a pre-Roman settlement cannot be ruled out, but the lack of substantial archaeological traces makes any estimate of its possible size currently impossible. 27 Although traces of parallel and orthogonal mudbrick walls suggest an orderly design, it is impossible to tell whether this settlement consisted of scattered, independent houses like the Northern Settlement at Umm al-Dabadib or a dense concentration of buildings. absence of streets: the built-up area consisted of a solid mass of interlocking buildings served by narrow vaulted passages. The 4th-c. CE settlement of Ayn al-Labakha is only partly preserved, but it shows identical characteristics.
The total area of the settlement of Umm al-Dabadib is about 8,100 m 2 (less than 1 ha). In general, Roman settlements are known to have a density of inhabitants ranging between 100 and 400 per ha. 28 Given that here there are no large streets and squares, and judging from what is visible, it may be suggested that about 75% of this settlement consisted of domestic units and the rest consisted of passages, and storerooms and warehouses. This results in an inhabited area of 6,075 m 2 , which may be divided by the average surface of the various domestic units (75 m 2 ) for a total of about 80 households. A low estimate of six family members per household would imply 480 inhabitants; a figure of eight would produce 640 inhabitants; and with 10 family members, the total figure would reach 800. Considering the density of the internal organization of this specific settlement, a higher figure is probably more likely than a lower one.
The earlier settlement at Umm al-Dabadib consisted of at least 36 buildings, many of which can be clearly identified as houses; a few more are likely to be buried under the sand. A figure of eight members per household would give a total of 320 inhabitants, whereas 10 would produce a total of 400both definitely lower than the figures for the later settlement. Given that this settlement does not contain evidence of 4th-c. CE occupation, it might be inferred that it was abandoned, or at least used much less, after the new settlement was built. 29 It is possible that the population moved from the old to the new settlement; the significantly higher number of inhabitants of the newer settlement (from It is tempting to interpret Ayn al-Labakha in the same way, but unfortunately, the remains of the pre-4th-c. CE settlement are too engulfed in a thick layer of soft sand and debris to provide any conclusive evidence.
The situation at the Gib/Sumayra Complex is uncertain: apart from the two forts and the two standing buildings at Two Houses, it is difficult to assess the amount of new building work performed in the Late Roman period. The ceramic evidence and some architectural remains indicate a substantial occupation dating to the 3rd and 4th c. CE, but how much of this scattered settlement was really built at that time is currently impossible to tell. At any rate, the fact that all of these settlements were connected by the same system of subterranean aqueducts dating to the 4th c. CE implies that all were active in the same period. Even if only Umm al-Dabadib appears to provide substantial evidence pointing in this direction, the addition of new settlers to the original inhabitants of the three sites at the beginning of the 4th c. CE is a likely scenario. This will be further considered below, in the discussion of the function of these sites.

Relative position
Umm al-Dabadib, Ayn al-Labakha, and the Gib/Sumayra Complex are located in the northern portion of the oasis, at a distanceas the crow fliesrespectively of 30, 25, and 32 km from the ancient capital Hibis. An average distance of 7-10 km separated the various sites that punctuated the northern area (Fig. 14).
Substantial human activities depend heavily on the presence of water, and in these desert areas, the places where water is easily available are basically always the same. For this reason, the modern land-reclaiming process is quickly taking over the ancient sites. The areas of Tulayb and Ayn Gib have already been encroached on: the main buildings are still standing, but the rest is rapidly being destroyed. The surroundings of Ayn al-Dabashiya and Ayn al-Tarakwa, which were clearly heavily exploited in the pastto judge from the remains of dozens of wells and a vast spread of fieldslie dangerously close to the expanding modern field systems. The low-intensity agricultural exploitation of a patch of land at Ayn al-Labakha might turn into a more substantial and problematic presence in the future. Only al-Dayr to the east and Umm al-Dabadib to the west remain, for the moment, untouched. 30 The actual traveling distances between the ancient sites are longer than the totals given in Fig. 14, which are measured as the crow flies, because the actual paths between them are not straight. Moreover, expressing distances through the desert in kilometers often makes little sense because the terrain plays a crucial role in determining the necessary effort. It is safer to speak about time, and yet, even this can be difficult to estimate: it depends on whether travelers walked or rode and what their load was. Finally, short trips were less likely to be affected by bad weather, whereas long journeys could be significantly slowed down or even halted by sandstorms and other events. In this instance, a speed of 4 km per 30 A few years ago, the request to install a field of wind turbines at Umm al-Dabadib was rejected by the Egyptian Environmental Affairs Agency of the Ministry of the Environment.
Living in a fringe environment 757 hour has been used to calculate the lowest estimate; the upper limit of the range has been suggested instead on the basis of the conditions of the terrain. With these criteria in mind, it may be estimated that distances of 8-10 km could be covered on foot in 2-3 hours, depending on the load, whereas the slightly longer route between Ayn al-Labakha and Umm al-Dabadib probably took 5-6 hours. In general, all these sites were equally distributed in the territory and traveling among them must have been relatively simple.
A short distance separates Umm al-Dabadib from a large mining area, located to the southwest, beyond a chain of dunes (Fig. 15). This area has not been systematically investigated. The presence of at least four rather basic settlements, accompanied by ceramics of Roman date and surrounded by shallow diggings, was first noted in 2004. 31 The site was subsequently re-examined and identified as a source of alum. 32 The true extent of the ancient activities is only visible from the satellite. Here, alum was evidently retrieved by sieving the terrain right under the desert surfacea technique that left behind thousands of holes, now filled by sand, that punctuate the desert surface at intervals of 10-20 m. More must currently be buried under the sand dunes (Fig. 16). Three of four small clusters of Corinna Rossi et al. 758 shelters are located along the eastern edge of the mining area in the direction of Umm al-Dabadib ( Fig. 15 and detail in Fig. 17). Desert tracks connecting the two areas are clearly visible from the satellite; some parts have been enhanced by modern tire marks because they represent the more logical way to travel in an east-west direction. At least one large well can be seen in the eastern portion of the area. The presence of others cannot be ruled out, but no traces of cultivation can be seen. Water was probably retrieved in this area and distributed in the jars and kegs noted there in 2004, but food and other supplies are likely to have been brought there regularly from Umm al-Dabadib. The walking distance was about 2.5 km to the southernmost edge of the Western Cultivation, and a total of 4.5 km to the Fortified Settlement. The first leg (corresponding to the distance between water sources) could easily be covered in one hour, regardless of the load.
Another mining area is located at a short distance from the Gridded Settlement at Ayn al-Labakha, precisely 1.2 km to the southwest (see Figs. 6 and 14). Noted for the first time by Beadnell,33 it consists of a small cluster of crude shelters and an elongated area of underground   galleries, corresponding to the upper edge of the escarpment bordering the site. 34 This mining area, ca. 750 × 160 (maximum) m, appears to be smaller than the one lying to the west of Umm al-Dabadib, but it consists of underground galleries and is therefore far more concentrated. The low resolution of the satellite images of this area prevents any detailed analysis of the remains from a distance. No mining area is directly associated with the Gib/Sumayra Complex. 35

Absolute position
Moving to a regional scale, the position of Umm al-Dabadib, Ayn al-Labakha, and the Gib/Sumayra Complex took advantage of the presence ofand ensured better control over the caravan routes that crossed the oasis by enlarging the portion of desert under direct administrative control.
The north-south caravan route, nowadays called Darb al-Arba'in, cut across a merciless and waterless desert but allowed travelers to head straight north from Wadi Halfa, in modern Sudan, to Middle Egypt, avoiding the long eastward curve that the Nile Valley makes. In Kharga, it met an equally important east-west route, linking this oasis to the Valley and to the other half of the Oasis Magnathe nearby Dakhla Oasis. 36 Even if these major thoroughfares fanned out into a network of paths, the point where the major north-south and east-west paths intersected was located in the northern portion of Kharga, halfway between Tulayb and Ayn al-Labakha. 37 From this crossroads, travelers could head east via al-Dayr to Upper Egypt, north to Middle Egypt, west across the chain of oases that punctuate the Western Desert to Lower Egypt and the Libyan coast, south to Nubia, and southwest to the Gilf al-Kabir and its then-green valleys, from where it was possible to reach the massif of Uwainat. Nowadays, this area is completely dry, but the situation was different in the past. In 1923, there were still at least four active springs at Uwainat. 38 The routes through Kharga, therefore, could allow travelers to bypass the Nile Valley, a well-known escamotage used several times in antiquity 39 as well as in more recent times: the daring Operation Salam, carried out by László Almásy in 1942 on behalf of German military intelligence, took full advantage of the possibility of travel along this alternative network of desert routes. 40 These long-range travels obviously required specific arrangements, and the caravans that embarked on long journeys needed to complete their preparations at the last water stations before they headed into the barren desert. The northernmost area of the Gib/Sumayra Complex corresponded to the last watering point for the caravans leaving Kharga for Middle Egypt before they embarked on a 160 km-long journey in a totally barren land, and it was the first green place to be encountered by incoming travelers. 41 This journey could take five days for fit travelers maintaining an average of 4 km per hour and walking for eight hours a day. 34 Ikram et al. 2020, 322-23. 35 On the suggestion that an alum deposit existed in the area, see Shortland et al. 2006, 154-55, andIkram et al. 2020, 322-23. 36 Rossi and Ikram 2013. 37 Rossi and Ikram 2018, 129. 38 Hassanein Bey 2006, 199-200. 39 Rossi and Ikram 2018, 20-23. 40 Gross et al. 2013. 41 Rossi 2018b, 551-53.
Living in a fringe environment 761 Things could be tragically different, however. In 1817, Caillaud witnessed the arrival in Asyut of a caravan of 16,000 individuals, including 6,000 slaves (men, women, and children), who "had been two months travelling in the deserts, in the intense heat of the year; meager, exhausted and the aspect of death on their countenance, the spectacle strongly excited compassion." 42 On their departure from Kharga, this caravan had probably loaded their last water at Ayn Ghazal. 43 In the Late Roman period, this operation would take place ca. 8 km further north, at the foot of Qasr al-Gib (see Figs. 2 and 14).
In antiquity, Umm al-Dabadib was en route to Dakhla along the track now called Darb Ayn Amur, 44 which exploited the presence of the tiny water station bearing this name, located halfway up the escarpment along the western border of the Kharga Depression (Fig. 2). 45 The distance between Umm al-Dabadib and Darb Ayn Amur was about 45 km as the crow flies, across a rather difficult and inhospitable terrain. The journey required at least a two-day march, depending on the load (see Fig. 14). Unlike the short-distance movements among sites and the long-range journeys to Middle Egypt or to the Wadi Halfa region, traveling to the other half of the Oasis Magna could therefore be classified as an intermediate, medium-range journey that required minimal assistance and support.
In conclusion, in northern Kharga, a scatter of Late Roman settlements of various sizes created a web-like network that was based onand at the same time supportedshort-, medium-, and long-range journeys. This system therefore played an important strategic role at the local and regional level, but it was also part of a larger, transregional system of military control of key communication routes.

Context
The Oasis Magna: a pluricentered district Over the years, the desert surface of the Kharga Oasis has been thoroughly explored by archaeologists, travelers, and locals, but only a few sites have been subject to significant and systematic archaeological excavation. Fresh textual sources that may offer new insights into the administration and management of this portion of the Oasis Magna can probably be retrieved only by accessing archaeological layers currently buried under the sand. Until this happens, we can only attempt to place the visible archaeological remains within the broader picture provided by the information collected so far in this area.
From 48 CE to the third quarter of the 4th c. CE, the Oasis Magna combined the oases of Kharga and Dakhla into a single, strictly interconnected administrative unit. It included three poleis -Hibis in Kharga, and Mothis and Trimithis in Dakhlaeach with its own civic institutions. Hibis and Mothis were the chief cities of the Hibite and the Mothite nomes respectively (see Fig. 2). 46 The picture that emerges from the documentary sources is that in the Late Roman Period, a scatter of rural settlements punctuated the entire extent of the oases. Among 42 Beadnell 1909, 34. 43 See Beadnell 1909, 28. 44 Ikram 2019. 45 Rossi and Ikram 2010, 239-41. 46 Bagnall and Tallet 2019. 762 them, there were full-scale villages, such as the komē of Kellis (modern Ismant al-Kharab, in Dakhla) and Kysis (modern Dush, in Kharga), 47 and minor inhabited centers closely associated with fields and vineyard-orchard properties (chōria, sing. chōrion). 48 In the vicinity of Kellis, large estates called epoikia (sing. epoikion, "farmstead") are also attestedsuch as Thio, 49 Pmoun Tametra, 50 and others unfortunately unnamed 51that functioned as centers of agricultural properties. 52 Even if their origins and evolution were different, 53 epoikia shared some typical features of self-sufficiency, such as the presence of storehouses, weaving workshops, presses for oil and wine, mills, associated water sources, and a variety of plantings of trees and farmland. 54 The rural settlement of Pmoun Berri (modern Ayn al-Gadida), 5 km northwest of Kellis, has been tentatively identified as an epoikion (see Fig. 2). 55 In general, the rural landscape of the Oasis Magna comprised large estates operated by a staff of managers 56 and tenant farmers (geōrgoi) who cultivated the land and paid their rent to the estate storerooms in kind. This included field crops, cotton, and high-value fruit tree crops, such as olives and grapes, also in the processed form of olive oil and wine. 57 Wealthy oasite landowners were involved in the city councils as well as in the administration of the district. 58 The economic interests of landlords, aided by local managers, are likely to have encompassed both the Kharga and Dakhla Oases. 59 An example is Faustianus, landlord of the estate for which the Kellis Agricultural Account Books were kept in the 360s CE. 60 He resided in Hibis, owned landholdings in the areas of Kellis in Dakhla as well as Hibis, 61 and likely had commercial interests at Trimithis. 62 Umm al-Dabadib, Ayn al-Labakha, and the Gib/Sumayra Complex must have played a role within this framework. Unfortunately, so far no textual evidence has come to light that 47 In the documentary sources, the komē Madiophris (P. Kellis I 2, 301 CE) and that of Chosis (SB III 7205, 290/292 CE) are also mentioned; they were probably in Kharga Oasis. 48 On the meaning of the word chōrion in papyri, see Bagnall 1999, 332 Rathbone 1991, 31. 53 4th-c. CE epoikia were established either by taking over preexisting villages (e.g., the village of Pagkuleōs, attested as an epoikion after 360 CE, P. Merton 36) or plots of land no longer (or not yet) fully exploited (e.g., P. Flor. II 150, 268 CE). 54 Lewuillon-Blume 1979, 184. 55 Aravecchia 2018, 296-303. 56 The estate managers are called pronoētai (sing. pronoētēs) in the Kellis Agricultural Account Book, a term attested also in O. Waqfa 37 from Kharga, as well as phrontistēs (O. Trim. I 4), indicating, literally, "a person who takes care" (Rathbone 1991, 71-82). See also Bagnall 1997, 27;Bagnall and Ruffini 2012, 40. 57 Ast 2019, 110;Bagnall 1997, 25-61, 73-76. 58 See, for instance, Ast andBagnall 2015, 110, andBagnall 2017, 215-17  Living in a fringe environment 763 might explain their status within the Oasis Magna. Concerning the Kharga Oasis, it is known that at the beginning of the 4th c. CE, three kōmai were associated with the polis of Hibis: Kysis 63 (modern Dush; see Fig. 2), located in the south, and Madiophris and Chosis, 64 possibly located in the north. There are currently no clues to associate them with specific archaeological remains; moreover, there might have been several other kōmai that are currently unknown to us.
The constellation of sites that constitute the Gib/Sumayra Complex certainly includes one substantial settlementthe Watermelon Settlement (endowed with a stone temple)plus at least four (smaller?) agricultural installations: Sumayra (around the fortlet), Two Houses, Ayn Ghazal (the extent of which in antiquity is difficult to establish), and Umm al-Qusur (see Fig. 12). It is tempting to see these sites as the remains of either a kōmē and its associated epoikia or as a scatter of epoikia and geōrgia. However, the presence of the forts of Qasr al-Gib and Qasr al-Sumayra and of the last water sources before the caravan route left the oasis clearly indicates that the function of this complex was not merely agricultural. This is even more evident in the case of Umm al-Dabadib and Ayn al-Labakha, isolated settlements devoid of satellite installations. The presence of alum mines nearby and the defensive character of their buildings suggest that these two settlements might have played more than one role and might therefore have been part of a complex network that ranged well beyond the local area. How this reflected on their administrative status, however, can only be revealed by future excavations and the retrieval of textual sources.

Agricultural exploitation
At the heart of agricultural exploitation lay systems to retrieve water, consisting of wells and subterranean aqueducts. Rainfall in Kharga is extremely rare. Years might pass between significant downpours, but when these do take place, they form temporary lakes that last a few months. The water slowly disappears and is replaced by patches of spontaneous vegetation that last several years. 65 No archaeological evidence has so far been retrieved regarding systems that could gather or store rainfall in this area in antiquity, but the rare and unpredictable occurrence of these downpours would probably make them unreliable for agricultural purposes.
Digging wells was the easiest and most widespread way to retrieve water. The importance of wells is clearly shown from a document dated to 246/249 CE addressed to the Rationalis Aegypti that concerns the hydreumata (sing. hydreuma) near the polis of Hibis. 66 It was compiled by the amphodarch of Hibis, who made a systematic survey of the area and listed the water sources near the polis, also recording hydrographic details. The name of each hydreuma (literally, "something that collects water"probably corresponding to a pool of artesian water) 67 is accompanied by a genitive eponym; and for each of them, the amphodarch numbered the artesian borings that fed the reservoirs (Gr. pēgai, sing. pēgē). He also noted whether the pool was stagnant, whether the pēgē did 63 P. Grenf. II 72 (308 CE) and P. Grenf. II 74 (302 CE). 64 For Madiophris, see P. Kellis I 2 (301 CE); for Chosis, see SB III 7205 (290/292 CE). 65 Rossi and Ikram 2018, 7. 66 SB XIV 11938. 67 Bonneau 1993, 61. not generate a pool on the surface, or whether two or more hydreumata were connected (Gr. sunepirrein). The eponym is sometimes preceded by the word pmoun, the transliteration for the Egyptian expression "the water of." Less frequently, the word hydreuma was followed by phrear ("well," somewhat different from the artesian one), or by the Egyptian term tchon, the meaning of which remains uncertain, although it is certainly related to water. 68 The references to a previous survey suggest that the conditions of water sources were periodically monitored. 69 Only a quarter of the 86 hydreumata listed in this document were associated with a plot of land (Gr. edaphē), whereas the others were en apeirō ("in the open")that is, they did not lie within cultivated plots. It can be inferred that the productivity of the area had the potential to be increased, possibly under the supervision (or at least with the consent) of the Rationalis Aegypti. It is therefore possible that this survey was performed to assess the situation and organize a better exploitation of the local water sources. 70 According to Olympiodorus of Thebes, in the oases, wells were dug by geōrgoi, who were allowed to use them to irrigate their own fields. 71 The use of the wells was time based, and the local tenants had to pay a rent ( phoros) for the use of water, which was calculated in days and hours. 72 The cost of digging a well and providing the necessary system for control and distribution is still unknown, but an account of the rent paid for the use of water at Trimithis allows the annual revenue for a fully exploited well to be calculated at around two-thirds of a pound of gold per year. 73 Even if there must have been expenses relating to maintenance, 74 the capital value was so high that wells might represent profitable investments for the private (and wealthy) owners. Wells not only made agricultural activities possible but also represented a defining element of the regional topography: 75 the documentary sources for Kellis, Trimithis, Hibis, and Kysis include several kōmai 76 and epoikia 77 that were probably named after the eponymous wells around which they developed.
The second system used to retrieve water in both Kharga and Dakhla was the draining tunnel or qanāt (plur. qanawāt), which collected water circulating in the fissures of the rocks at the junction of different geological layers and brought it by gravity to the areas to be cultivated. This system was already being implemented in Kharga during the Persian occupation of Egypt in the 5th c. BCE. The area of Kysis, in the south of the oasis, shows evidence of repeated reuse of and modifications to the original Persian system of qanawāt down to the Late Roman period. 78 The northern qanawātthat is, those of 68 Parsons 1971, 173-75;Reddé 2004, 192 Bagnall and Ruffini 2012, 40-41. 74 For instance, the cost of the work of hydrokathartēs (literally, "well cleaner") related to the maintenance of wells and probably of the water-distribution system too (O. Kellis 133; O. Trim. I 53; P. Kellis IV 96, ll 17, 345, 1134, 1402, 1748. 75 Wagner 1987, 279-83;Bagnall and Ruffini 2012, 33- Unlike the Nile Valley, where the agricultural rhythm was dictated by the annual inundation, in the oases, the combination of wells and qanawāt allowed continuous irrigation of the land and therefore agricultural exploitation spread over the entire year. The fields, fertilized by guano collected from the pigeon towers that accompanied the cultivated areas, 80 produced wheat, barley, flax, and pulses in winter, and sorghum and pearl millet in the summer. Cotton became one of the most important products in the Roman period and is likely to have played a significant role in the local economy. 81 Keeping centers of production and settlements connected with one another required considerable local movement and transport, 82 accomplished by the hard work of armies of donkeys and their drivers. 83 Agricultural commodities were sent from farms to central storehouses where they were recorded and redistributed within the estate (to supply the local needs of humans and animals) and beyond to the Nile Valley (e.g., to Hermopolis). 84 Cotton and highvalue tree crops (olives, dates, figs, and jujubes) that were not consumed locally constituted an effective surplus that could be exported out of the oasis. 85 Finally, a significant part of local produce, perhaps alongside cash payments, must have been requisitioned through the annona militaris for the needs of the army stationed in the oasis (see below).
In this context, Umm al-Dabadib, Ayn al-Labakha, and the Gib/Sumayra Complex probably belonged to a large-scale program of coordinated exploitation of the agricultural potential of the oasis. The visible remains of their fields range between 10 and 50 ha; however, these total figures do not indicate single, large patches of cultivated land. Rather, they correspond to the sum of smaller, separate patches of cultivated fields fed by different water sources. At Umm al-Dabadib, there were three cultivated areas, corresponding respectively to 30, 15, and 5 ha (for a total of 50; Fig. 5); at Ayn al-Labakha, there were two areas of 10 and 22 ha (for a total of 32; Fig. 6); the Watermelon Settlement had two areas, of 10 and 15 ha, respectively (for a total of 35); Sumayra and Two Houses each had one area, of 20 and 22 ha, respectively (Fig. 7). Pairing this with the archaeological evidence retrieved from Ayn al-Tarakwa and Ayn al-Dabashiya, and around Hibis, 86 it may be concluded that the Late Roman agricultural system consisted of a scatter of relatively large cultivated patches, each fed by its own water source, an arrangement mirroring the overall impression conveyed by the list of hydreumata drawn by the amphodarch of Hibis. The success of this spread-out system resided in the possibility of counting on different, independent sources of revenues, but it depended heavily on keeping them connected and controlled. 79 Rossi 2018b, 520-21. 80 Ikram and Warner 2012.  85 Bagnall 2015b, 169. 86 Rossi and Ahmed 2019. The largest remains of ancient cultivations in northern Kharga are located around Ayn al-Tarakwa and Ayn al-Dabashiya. It is likely that not all wells were active at the same time and that, in ancient times, the area hosted some sort of "wandering cultivation" that progressively exploited a succession of water sources (Rossi 2018b, 509).
Considering the distance to be covered across the desert that separated the Oasis Magna from the Nile Valley, the amount of perishable agricultural produce that departed from the oasis must have been relatively limited. The substantial investment that the Roman authorities poured into the agricultural exploitation of the Kharga Oasis, therefore, is likely to have been meant to give a boost to the local communities. This suggestion appears to be even more likely if paired with the arrival of new settlers, perhaps relocated from the Nile Valley. The increase in agricultural production might therefore have been instrumental in the implementation of large-scale exploitation of the region, based not only on agricultural products but also on other commercial and strategic assets.

The exploitation of mineral wealth
It is generally assumed that the ancient mining areas that can be seen in the Kharga Oasis were devoted to the retrieval of alum. No physicochemical analyses have been carried out so far on samples retrieved there, making precise identification of the mined material impossible at present.
In ancient times, alum (Gr. styptēria; Lat. alumen) referred to a large group of minerals with astringent characteristics (LSJ, s.v. styptēria), including potassium aluminum and impure mixtures of sulphates of iron and aluminum, which occurred naturally. 87 As a result of different provenance, color, and specific features, alum and alunites were distinguished into different genera and species. 88 Among them were alumen liquidum, 89 which refers to a deliquescent variety of alum (Gr. styptēria hygra, "wet, moist, fluid"); 90 alumen strongylen, which is round and compact, and may be associated with the mineralogical class of the botryoidal alum (Gr. stroggylē styptēria); 91 and alumen concretum, which is characterized by a scissile or fibrous form (Gr. schiston, "being split, divided"). 92 A superficial analysis of the desert surface in the mining area west of Umm al-Dabadib suggests that the last type might have been quarried there.
Living in a fringe environment profitable business for all actors involved. 96 The procurator ad Mercurium (Gr. Epitropos Hermou) 97 in Alexandria was in charge of the general supervision of the alum monopoly, and an array of local functionaries were involved in the control of its production and transport, andfrom the mid-3rd c. CEalso its sale and distribution. 98 Work contractors (Gr. mistōthai) 99 controlled the lease of the mining areas by the state, whereas private contractors (Gr. telōnai, literally "tax buyer") 100 and liturgical officials (Gr. epitērētai) supervised the transports of alum to the Nile Valley. 101 The beasts of burdendonkeys and camelswere probably provided by private owners who were involved in the alum transport (similarly to the ktēnotrophoi and nauklēroi, respectively involved in the transport of the Egyptian state grain by land and by sea). 102 The main Egyptian alum-mining areas lie in the two districts of the Oasis Parva and Oasis Magna, from where the alum was transported to the Nile Valley across the desert. A network of functionaries dedicated to the alum monopoly probably existed only in the nomes directly associated with the mining of alum and related transport activities. 103 Concerning the Oasis Parva, mistōthai, telōnai, and epitērētai associated with local alum mining are documented in the Oxyrhynchite 104 and Arsinoite nomes. 105 Concerning the Oasis Magna, the textual sources are limited to a 4th-c. CE tax receipt for alum found at Kellis. 106 The archaeological evidence is more abundant, and it includes the two mining areas of Umm al-Dabadib (at least 650 ha) and Ayn al-Labakha (mainly underground) in Kharga, described above, as well Gabal Tarif (over 3,000 ha), and a smaller site near al-Qasr in Dakhla. 107 A network of functionaries, similar to the one attested for the Oasis Parva, might have managed the alum of the Oasis Magna in the two Nile Valley nomes of Lykopolite and Panopolite. 108 The labor force in the Roman mining districts in Spain, Dacia, and Egypt's Eastern Desert included not only slaves but also a substantial portion of free workers engaged both in extraction and in operative organization; 109 a similar workforce might also have been employed in the mining areas of the Oasis Parva and Oasis Magna. It is possible that the site had been exploited since the New Kingdom well into the Roman era (Hope et al. 2009;Ikram et al. 2020, 325). 108 Kruse 2007, 543. 109 Edmondson 1989, 95;Mrozek 1989;Andreau 1990, 89-92; for Egypt's Eastern Desert, see Cuvigny 1996;Hirt 2010, 206-8. Domitius, 110 alum mining required a year-round labor supply and consequently the accommodation of workers and their supervisors in the vicinity of the mines.
The 4th-c. CE settlements of Umm al-Dabadib and Ayn al-Labakha seem to suit such needs: located near the mining areas and self-sufficient in terms of basic requirements, both settlements could support the workforce dedicated to the mining activities and act as administrative centers to organize and manage the ensuing trade. At Ayn al-Labakha, the mining area was very close to the main settlement, and water could easily be transported daily from the nearby sources. At Umm al-Dabadib, however, it was distant and larger, and it was therefore endowed with temporary shelters and at least one major well.
Concerning the subsequent transfer of the material away from the mining area toward the Nile Valley, the shortest and most logical way to send the alum from Kharga to the core of the Roman Empire was to travel straight east, across Ayn al-Labakha, then head north and leave the oasis from Qasr al-Gib. Alternatively, to serve Upper Egypt and other commercial routes heading east across the Red Sea or south along the Nile, caravans could head straight east and load the last water at al-Dayr before heading to Upper Egypt. 111 It is therefore likely that at both Umm al-Dabadib and Ayn al-Labakha, a number of mistōthai operated in connection with a network of telōnai and epitērētai, and that the two sites were the seat of officers of the Roman Empire who kept in contact with other sites belonging to the same network, probably in strict connection and coordination with the army.

The presence of the army
The increasing presence of the army in northern Kharga is attested by a large number of archaeological remains but, until now, by only a small number of textual sources. In a first phase, corresponding to the last years of the 3rd c. CE, the main oases of the Western Desert were endowed with a chain of fortresses meant to host alae and other auxiliary forces. These were the nearly-identical enclosures at al-Dayr in Kharga, al-Qasr in Dakhla, and Qarat al-Tub in Baharyia. Unlike the other oases, Kharga was also endowed, immediately afterward, with a scatter of fortlets and fortified-looking settlements. Umm al-Dabadib, Ayn al-Labakha, and the Gib/Sumayra Complex belong to this second phase, together with Tulayb, possibly Qasr al-Nissima, Qasr al-Baramudy, and the reuse of Dush in the south. 112 The most important textual source on the distribution of the Roman army in this period, the Notitia Dignitatum, contains only scant information on this area. Compiled toward the end of the 4th c. CE but also including information dating from a slightly earlier period, 113 it lists the garrisons distributed in the western oases. The Ala prima Abasgorum, said to be 110 P. Oxy. XVII 2116. 111 If alum was quarried also in the western half of the Dakhla Oasis, it is possible that part of it reached Kharga via Ayn Amur and, therefore, via Umm al-Dabadib; a longer but waterless route circumnavigated Gabal Tartur along its southern edge. Alternatively, Middle Egypt could be reached by embarking on a long march straight across the desert along the route now called Darb al-Tawil (literally, the "Long Track"), which probably exploited the water source of Abu Garara, an elusive site located out in the desert to the north of Dakhla (Fig. 2), and which still awaits a thorough archaeological investigation. 112 Rossi 2018a, 450-51. 113 Southern and Dixon 1996, 1.
Living in a fringe environment 769 stationed at Hibeos -Oaseos maioris, served sub dispositione viri spectabilis ducis Thebaidos (under the command of the dux of the Thebaid); 114 the same garrison is mentioned again a few lines below among those quae de minore laterculo emittuntur ("which are assigned from the lesser register"that is, from the list of the minor military officers), together with the Ala prima Quadorum, Oasi minore -Trimtheos. 115 The Oasis Parva fell instead under the command of the Comes limitis Aegypti and was guarded by the Ala secunda Armeniorum -Oasi minore 116 and the Ala secunda Assyrorum -Sosteos 117that is, Psobthis, modern Qarat al-Tub. 118 The Ala prima Abasgorum is probably also mentioned in an earlier source dating to 309 CE. 119 A slightly later list of soldiers indicates the presence around Hibis of an unnamed cohors. 120 Seemingly, the polis of Mothis in Dakhla hosted, at some point, the cohors scutata civium romanorum Mutheos, mentioned in the Notitia Dignitatum. 121 Detachments from garrisons stationed along the Valley appear to have also spent some time in the Oasis Magna. 122 Estimating the total number of soldiers who were present in the area throughout the 4th c. CE would be extremely difficult for a number of reasons, ranging from the lack of sources to the lack of precise information on the size of the garrisons, a problem that is typically encountered in the study of the Late Roman army. 123 At any rate, the need to collect the annona militaris for all these troops is attested both at Dakhla 124 and at Kharga. 125 The most obvious explanation for the presence of armed soldiers in the area would be an external threatperhaps the recurrent attacks carried out by the nomadic populations that lived along the edges of the Empire. After the isolated disorders in the district of the Oasis Magna at the end of the 3rd c. CE, caused by the Nobates and successfully suppressed by Diocletian, 126 turmoil is again documented from the 5th c. CE, this time due to a group of Libyan origin called the Mazices. The disorders affected the entire Western Desert 127 and, in particular, the Oasis Magna. They started in the time of the exile of the deposed Archbishop of Constantinople, Nestorius, after 435 CE, 128 and continued at least until the first half of the 6th c. CE, as may be inferred from the episode recorded by Moschus during his visit to the oasis. 129 Ast and Bagnall 2015, 2;Reddé 2004, 199-205. 123 Southern and Dixon 1996, 16-20. 124 Ast and Bagnall 2015. 125 Wagner 1987, 82, 378. 126 Procop. Pers. Corinna Rossi et al. 770 Egyptian oases is documented from the last decades of the 4th c. CE onward, but initially they did not seem to be considered a threat. 130 This changed in the first decade of the 5th c. CE, when the Mazices started to be perceived as barbaric marauders who sacked and raided villages throughout the entire Western Desert; 131 their violence and cruelty was also acknowledged by other nomadic groups. 132 It is possible that this sudden shift of perception was due to the fact that a group of Mazices established themselves in the proximity of an Egyptian oasis, 133 possibly Bahariya (given that the distance between Oxyrhynchus and the oasis ubi genus est Mazicorum ("where the people of Mazices are located") was said to correspond to a fourday journey). 134 From this convenient position, the Mazices could easily move across the Western Desert and disrupt communications and trade that ran along the network of desert tracks. This disturbance, particularly if prolonged, might have caused a slow but constant decline of many oasite settlements, especially those that heavily depended on overland transport, such as for the alum trade.
Umm al-Dabadib, Ayn al-Labakha, and most of the Gib/Sumayra Complex do not seem to offer evidence indicating a full-scale occupation during the 5th c. CE. It must be stressed, though, that the study of the ceramic material of the oases (of Kharga in particular) still needs to be refined, because not many sites have been thoroughly excavated. For the moment, according to our current knowledge, it may be suggested that all the peripheral sites of Kharga were abandoned around the turn of the century. A dramatic lowering of the water table in the entire periphery of the oasis's depression is attested in the same period. Whether this was caused by the extensive agricultural exploitation that had been taking place in the previous decades or whether it depended on other factors remains unclear. It is difficult to establish whether the disturbance by the Mazices and the lack of water resulted in a "perfect storm" that sealed the fate of these settlements or whether the Mazices dared to settle nearer to the oases because they felt emboldened by the abandonment of the fortified settlements, which was itself caused by the lack of water. 135 Despite the decline and abandonment of many settlements and the presence of the Mazices, the Egyptian oases remained part of the provincial territory of Late Roman and Byzantine Egypt. In fact, until the first third of the 6th c. CE, the Oasis Magna continued to be officially mentioned as a place of banishment. 136  large mining area west of Umm al-Dabadib, located in a harsh environment and in an isolated position were it not for Umm al-Dabadib itself: the contemporaneity of settlement and mining area suggests that the exploitation of the latter might have been one of the reasons for the expansion of the former.
The pattern at Ayn al-Labakha is the same as at Umm al-Dabadib: a Late Roman settlement of fortified appearance, including a central fort, and built near a large mining area exploited in the same period. Considering that alum was subject to state monopoly, the coordinated construction of Umm al-Dabadib and Ayn al-Labakha near two important mining areas reinforces the hypothesis that these two sites belonged to a larger strategic program, set up by the Roman authorities, that appears to have encompassed the entire northern portion of the Kharga Oasis. 142 Unlike most of the agricultural produce, alum was expected to be mainly traded away from the area. Once more, this reinforces the need for an organized network of communications. In a desert environment like the Kharga Oasis, this translated into the construction of a series of stations and checkpoints rather than actual paved roads. The stretch of desert between the northern and eastern escarpment of the Kharga Depression and the Nile Valley was hopelessly waterless, so not much could be done there. What was instead done in the oasis was to extend the human occupation of the area as much as possible, thanks to a network of sites provided with independent water sources and local, basic agricultural produce. All the local movement among settlements and areas under exploitation took place along this network of local routes. At its northern edge, the Gib/Sumayra Complex took on the burden of preparing the large caravans for the ascent of the plateau and the difficult journey toward Middle Egypt, and of welcoming those who managed to arrive from there. Most of the alum quarried at Gabal Tarif, Umm al-Dabadib, and Ayn al-Labakha must have departed northward from there, under the close supervision of the Roman authorities.
Connectivity appears to have been the essence of the entire operation, which was meant to control the movements of people and goods. The amount of short-and medium-range movement of people and beasts of burden that took place on a daily basis in the Late Roman period along the desert track connecting these sites must have been significant; the cemetery for donkeys found at Umm al-Dabadib represents a reminder of the hard work performed by these animals during that historical period. 143 In terms of large-scale movements, all three sites represented important rest stations along medium to long desert routes, thereby providing at the same time support for and control over travelers.
The military component of the strategy of control of the Oasis Magna appears to have changed from being predominant at the end of the 3rd c., when each oasis of the Western Desert was endowed with a similar Late Period legionary fortress, 144 to being part of a more complex scenario that took shape in the first quarter of the 4th c. CE, with the construction of the fortified settlements in Kharga. Trade was an important component of this picture, but it alone would not explain the complexity of the operation that was staged attribute all the mining activities to a specific historical period. Future investigations on the ceramic material scattered in the area might shed further light on this issue. 142 Rossi and Ikram 2018, 551-59 143 Rossi and Ikram 2018, 258-59. 144 Rossi 2018a.
Living in a fringe environment 773 there: alum was quarried also in Dakhla, which was equally exposed to the threat of nomadic incursions, and yet, no settlements of fortified appearance were built in that part of the Oasis Magna.
The main difference between the eastern and the western part of the Oasis Magnarespectively, Kharga and Dakhlais the fact that Kharga acted as a major desert crossroads (see Figs. 1 and 2). Controlling this crossroads meant controlling access to various parts of the Nile Valley and movements along the border of the country. This might be one of the reasons why, perhaps as early as the Ptolemaic period, the capital of the Oasis Magna was transferred from the larger and wealthier Mothis in Dakhla to the smaller but evidently strategically more important Hibis in Kharga. 145 Thus, when Diocletian reestablished the southern border of the empire, Kharga was directly involved in the reorganization of the frontier thanks to the possibility of controlling the major desert thoroughfares, along which lucrative trade could be carried out for the empire's benefit.
Umm al-Dabadib, Ayn al-Labakha, and the Gib/Sumayra Complex are likely to have been an integral part of this design: they had a fortified appearance and probably hosted small contingents of soldiers, located either near mining areas or at the exit/entry point of the oasis. Not only were these settlements self-sufficient in terms of water and agricultural production but they might even have contributed to the export of some specific crops. They therefore played a coordinated role within the northern portion of the Kharga Oasis and are also likely to have been part of a larger strategy of control encompassing not only the eastern part of the Oasis Magna but the entire Western Desert.
Further studies will be necessary to reconstruct the large-scale mosaic of the Late Roman strategy of control of the empire's desert frontier. The Kharga Oasis may represent a crucial tile, because it contains archaeological remains that can still be interrogated to retrieve specific answers. In particular, if the interpretation that Umm al-Dabadib, Ayn al-Labakha, and the Gib/Sumayra Complex belonged to a network encompassing the entire Western Desert is correct, at least some characteristics of these sites should be present in other oases.
The degree of analogy among sites might not be confined to their function but may also include the adoption of the same architectural solutions. The similarity in dimensions and design of the three late-3rd-c.-CE legionary fortresses of Kharga, Dakhla, and Baharyia is straightforward given that the three buildings played the same role. More intriguing is the striking architectural similarity between the Fort of Umm al-Dabadib and granary C65 at Karanis. The latter was dismantled during the archaeological excavations carried out by the University of Michigan, but its drawings, pictures, and samples are being reexamined together with other material from the same site. 146 The date originally attributed to this building is one century earlier than the Fort of Umm al-Dabadib, but the impressive similarity between the two buildings cannot be ignored and definitely needs further investigation.
Because the level of preservation of the Late Roman remains in the rest of the Western Desert is uneven, many sites have never been properly investigated and some no longer exist. The well-preserved remains of Umm al-Dabadib, Ayn al-Labakha, and the 145 Cf. Bagnall and Tallet 2019. 146 Burton 2020 and pers. comm. Gib/Sumayra Complex may serve as sets of reference points to be used to recognize further elements of this complex picture.