Barcoding Magmas: Applying Machine Learning In Zircon Geochemistry To Increase Provenance Accuracy

Sedimentary provenance allows researchers to reconstruct and interpret the history of a sediment from its parent rock (source) to its deposition (sink). The "sources" are eroding mountains and the "sinks" are sedimentary basins. Sedimentary provenance studies can allow us to reconstruct past environments, climate and tectonics. It allows us to assess the suitability of sedimentary rocks for resource exploration (hydrocarbons, geothermal, aquifers) or carbon capture/storage (CCS).
A key feature of mountain building events (orogenesis) is the emplacement of large bodies of molten rock (magma) which cools and crystallises as igneous rocks, in particular granite. Granites are primarily composed of quartz, feldspar, amphibole and/or mica. In addition to these, a common mineral, which grows within the magma during emplacement, is zircon.
Zircon is a remarkable material. It's easily datable using Uranium-Lead (U-Pb) geochronology, yielding the age of the igneous body in which it formed. It is physically and chemically robust, withstanding alteration during erosion, transport and later diagenesis (physical and chemical changes in sediments after their deposition). Traditionally, provenance studies are conducted by collecting large suites of zircons from sediments, dating them (U-Pb geochronology) and matching the acquired ages to previously dated geological regions. This gives a very broad picture of the source of the sediment but lacks accuracy. For example, zircons from the Chinese Loess Plateau are sourced from granites in the ~2.5 million km2 Tibetan Plateau. Closer to home, the Carboniferous age sedimentary basin of North West Ireland sources material from Caledonian Granites in Co. Donegal. However, these granites are part of a suite of coeval granites found throughout the ~92,000 km2 Caledonian mountain belt in Britain and Ireland. As a result, links between these sediments and the Donegal granites remains speculative, and existing models of sediment transport may be highly inaccurate.
During crystallisation in the magma chamber, zircons incorporate trace elements which geochemically record the changing conditions in the magma (much like tree rings record the passing of the seasons). Each individual zircon only records part of the geochemical history of that magma chamber from when it was crystalizing. Therefore, to reconstruct a more complete geochemical history of the magma many zircons must be analysed. We propose to use machine learning and pattern matching algorithms to analyse the growth patterns and trace element distribution within 10,000 individual zircons from the five granite bodies (2000 per body) making up the Donegal Batholith. This analysis will create a geochemical barcode, representing a continuous timeline of conditions within their corresponding magma chambers. Because magma bodies are natural and dynamic systems, no two bodies will have an identical geochemical signature, so the generated barcode will be unique to each.
Zircons from sediment in the Carboniferous sandstones of NW Ireland believed to derive from the orogenic belt containing our magmatic systems will be geochemically analysed using the same techniques. Each of these zircons will contain a partial barcode of its parent magmatic body. Using bespoke pattern matching algorithms, these partial barcodes will be compared to new barcodes of the nearby granite bodies to ascertain their exact source & reconstruct the regional sediment transport pathways during the Carboniferous. In essence, this will be akin to matching a partial DNA sequence from the zircons in sediment to the complete DNA genomes of the nearby igneous units.
This research brings together a UK-based team of international scientists, Dr Dempsey & Dr Bird (University of Hull) and, Dr Einsle & Dr Neill (University of Glasgow). Our team combines a wealth of experience of Caledonian tectonics & magmatism, zircon geochemistry & provenance, exceptional observational & analytical skills.