A machine learning approach to Nemrut Dağ and Bingöl Obsidians: non-invasive and non-destructive geochemical analyses for provenance source of Tulūl al Baqarat tools (Iraq)

Abstract

Abstract A machine learning approach to Nemrut Dağ and Bingöl geological obsidians was performed to determine the provenance of source material of nine obsidian blades from the archaeological site of Tulūl al Baqarat (Iraq). Obsidians tools have been chemically analysed with a non-invasive and non-destructive approach by a low vacuum SEM-EDS microprobe for determining major and minor elements and by a bench-to-top micro-XRF for trace elements. To identify the supply volcanic complex which produced the blocks from which they were knapped, a geochemical identification of Turkish obsidian sources, from the volcano to the artefact samples, was carried out introducing the use of the machine learning approach as exhaustive discriminative approach with complementary well-known geochemical comparisons and Principal Component Analysis. Obsidian tools show a rather homogeneous comendite and peraluminous rhyolite composition with an high Zr amount which excludes most obsidian outcrops in Turkish and Armenian volcanic sites as original obsidian sources. However, only using a machine learning approach to major, minor and trace elements, the obsidian tools have resulted geochemically comparable to Nemrut Dağ mild alkaline rhyolitic obsidians from pre-caldera eruptions, and now outcropping inside the caldera and in Sicaksu exposure. This source provenance from Nemrut Dağ stratovolcano in South-eastern Turkey, located near the Turkish route of the Tigris River, supports the evidence of a network of trade and broad exchange since 4th Millennium from Turkey and the South Near East, presumably through the basins of the Tigris and Euphrates to the shores of the Persian Gulf.