Late Roman glass production at Tel Yavne revealed through geochemical analysis

Archaeologists working at a sprawling industrial complex near Tel Yavne in Israel have uncovered and chemically characterized one of the most significant primary glass production sites of the Late Roman world, and the results are reshaping what researchers thought they knew about how the ancient glass industry obtained and combined its raw materials. In a new study published in Archaeological and Anthropological Sciences, a team led by Yoav Ben Dor of the Geological Survey of Israel, together with researchers from the Israel Antiquities Authority and the University of Haifa, presents a detailed geochemical fingerprint of glass produced in furnaces at Yavne during the fourth and early fifth centuries CE, and compares it directly with the quartz-rich sandy deposits surrounding the site. Their findings suggest that while local sands could indeed have supplied the silica for Yavne’s furnaces, something unexpected was also going into the melt, possibly for the first time in the known history of natron glassmaking.

Primary glass production, the industrial-scale melting of raw materials into slabs of unworked glass that were later shipped to vessel-making workshops across the Mediterranean, was one of the great chemical enterprises of antiquity. During the first millennium CE, it was concentrated almost exclusively in the eastern Mediterranean, where workshops combined quartz-rich sand with natron, a sodium carbonate evaporite harvested almost entirely from Egypt’s Wadi Natrun, to produce the famous Roman glass that filled the empire’s cups, windows, and lamps. The industry operated according to highly consistent recipes, delivering glass of remarkably uniform quality over decades and even centuries, a feat that scholars regard as one of the earliest achievements of organized chemical and industrial engineering. Yet because furnaces were systematically dismantled after each production cycle and materials were consumed with great efficiency, clear archaeological evidence of primary production is extremely rare, and only a handful of sites have ever been identified along the eastern Mediterranean coast, in Israel and Egypt.

The Yavne discovery, made within a vast multi-period industrial landscape that also contained winepresses, pottery kilns, and tombs, is exceptional on several counts. In an area designated Area L, excavators revealed a structure they have dubbed the “Glass House,” measuring at least five by eight meters, which uniquely contained furnaces from both stages of glass production side by side: two large rectangular tank furnaces, each roughly two by four meters, used for melting raw glass, and four small circular furnaces for manufacturing vessels. It is the first in situ complex in the region where primary and secondary glass furnaces were found together within a single building. The furnace floors, built from hard carbonate material, preserved embedded glass lumps, and beneath them lay thick layers of dense black ash, testimony to the enormous quantities of fuel that fed the melt. Excavators believe the furnaces were deliberately dismantled after their working life, which is precisely why so few such installations survive anywhere in the ancient world.

The research team faced a rare opportunity. The site sits within about ten kilometers of five distinct quartz-rich sandy deposits, including modern beach sand, the semi-stabilized dunes of the Palmachim field, cemented aeolianite sandstone known locally as kurkar, buried fluvial sand beneath agricultural fields near the furnaces, and the clay-rich soils of the Soreq stream floodplain. All of these sediments ultimately derive from the Nile River, whose sands have been swept eastward along the Levantine coast by longshore drift since the Pliocene, so they share a common geological ancestry. That geographic accident allowed the scientists to do something few studies have attempted: a direct, side-by-side comparison of finished glass from a primary furnace with the actual raw materials potentially available to the glassmakers who worked there.

The analytical program was exhaustive. Twelve glass samples of various color shades were collected from the furnace tank floor, a fill south of the furnace, and an adjacent stone collapse, along with four slag specimens consisting of vitrified furnace material. Sixteen sediment samples were taken from the five deposit types along a west-to-east transect. Glass and sediment were digested in clean laboratories using multi-acid total dissolution and measured by inductively coupled plasma optical emission spectrometry and mass spectrometry, calibrated against certified reference materials, with procedural blanks held below one percent of measured values. Mineralogy was determined by X-ray diffraction, grain-size distributions by laser diffraction, and, in a methodological first for natron glass studies, stable carbon isotopes of organic carbon trapped within the glass matrix were measured by elemental analysis coupled to isotope-ratio mass spectrometry. Statistical treatment followed best practice for compositional data, with centered log-ratio transformation and z-score normalization feeding into principal component analyses of major elements, trace elements, and rare earth elements treated as separate datasets.

The chemistry told a nuanced story. The Yavne glass displays the overall hallmarks of Levantine natron glass, with sodium oxide content between roughly 11.6 and 16.3 percent, silica estimated between 68.5 and 76.1 percent, and chondrite-normalized rare earth element patterns nearly indistinguishable from the surrounding sands, confirming a shared Nilotic provenance. Yet the Yavne assemblage forms its own distinct compositional group, which the authors propose to call the Levantine-Yavne type. Its characteristic fingerprint, approximately 13.5 percent sodium oxide, 9.5 percent calcium oxide, 3 percent alumina, 0.1 percent titanium, and 0.5 percent iron, resembles the glass of the well-known fourth-century workshop at Jalame but is measurably different in its sodium, titanium, iron, and zirconium contents, and it plots clearly apart from Egyptian glass groups in ratio diagrams of calcium oxide versus alumina and titanium versus alumina. The soda levels are also lower than in earlier Roman glasses, hinting that Yavne stands at the beginning of a compositional trend that developed in later centuries.

The central puzzle emerged when the glass was compared with the candidate sands. The sodium and silica contents of the glass could, in principle, be achieved by melting natron with any of the local deposits, and the best overall chemical match is the orange-tinted dune sand of the Palmachim field several kilometers to the west, which agrees with the glass in its magnesia, lime, and alkali contents. But the Yavne glass is anomalously enriched in aluminum, potassium, and, to a lesser degree, phosphorus and rare earths, and no studied sand deposit, and not even the famously alumina-rich sand of the Belus River further north, can account for that signature on its own. Principal component analysis of trace elements reveals a striking triangular geometry, with the sandy deposits, the glass, and the Soreq floodplain soils occupying the three apexes, and the furnace slags plotting between them. The authors weigh two explanations. Either the glassmakers deliberately added a previously unreported ingredient, most plausibly the clay-rich alluvial soils of the Soreq floodplain, which contain quartz silt and abundant clays rich in aluminum and potassium, or the excess aluminum and potassium reflect contamination of the analyzed glass chunks by furnace material. A third possibility is that an as-yet unsampled sand source was used. Whichever proves correct, the study is the first to raise the possibility that natron glass, long believed to be a simple two-ingredient recipe of quartz sand and natron, may sometimes have contained a third component.

The stable isotope work added an entirely new dimension to the investigation. Organic carbon trapped in the glass ranges from 0.02 to 0.07 percent by weight, with delta carbon-13 values between minus 26.7 and minus 22.6 per mil relative to the Vienna Pee Dee Belemnite standard, implying that somewhere between about 70 and 100 percent of the fuel carbon derived from C3-type vegetation, the trees and shrubs whose photosynthetic pathway yields characteristic isotopically light carbon. Slag samples, richer in organic carbon at 0.38 to 0.56 percent, show even lighter and tightly clustered values around minus 28 per mil, indicating an essentially pure C3 fuel source. The researchers suggest that debris from Yavne’s extensive wine and olive presses, abundant across the industrial complex, may have fed the furnaces. Notably, the mean carbon isotope value of the glass, about minus 24.7 per mil, differs significantly from that of the slag, about minus 28 per mil, hinting at a more complex carbon incorporation process during melting that the authors say deserves further study.

The isotopic data even offered a crude thermometer. Because calcite decomposes to quicklime above about 700 degrees Celsius, and re-carbonated lime inherits strong kinetic isotope fractionation that grows more negative with firing temperature, the carbon isotope values of carbonate in the slag can serve as an “isotopic thermometer.” Extrapolating an established calibration from fired pottery clays, the most negative slag value of minus 23 per mil corresponds to an equivalent temperature of roughly 550 to 800 degrees Celsius. Since natron glass melts require higher temperatures, the authors interpret these figures as minimum estimates and conclude that the sampled slag likely came from cooler, peripheral parts of the furnace rather than the hottest melting zones.

Beyond its technical findings, the study carries broader implications for how archaeologists reconstruct ancient supply chains. What was once imagined as a single sand quarry behind each compositional glass group is now understood as a mosaic of geographically variable deposits and production sites, and the Yavne results demonstrate that meaningful answers require sampling not just the glass and the obvious beach sands but also dunes, buried sediments, kurkar, and soils. The southernmost known primary production site in Israel, Yavne now defines a new compositional group in the Levantine glass landscape and, perhaps more provocatively, opens the possibility that Roman glassmakers were more experimental chemists than their reputation for rigid standardization suggests. As the authors note, only additional sampling of sedimentary deposits around other primary production sites will reveal whether the Yavne recipe was a local innovation or a lost chapter of a much older tradition.