"It's written in the fine print" - A machine learning based, holistic approach to sedimentary provenance of fine-grained deposits

• Applicant: Dr. Nils Keno Lünsdorf
• Subject Area: Geology
• Term: since 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 528496994

Sedimentary Provenance Analysis (SPA) provides important concepts and increasingly sophisticated tools to find and investigate sediment sources and constrain the relevant processes in both modern and ancient sediment routing systems. Most methods applied and many case studies focus on sand-sized sediment, which is in strong conflict with the fact that the majority of clastic sediments are finer grained. To solve this conflict, our approach aims at pushing the limits of the grain-size range suitable for single-grain SPA towards fine grades. This will be achieved by a workflow that combines optical microscopy, Raman spectroscopy as well as several electron and laser beam-based techniques with machine learning methods to infer an integrated petrographic, mineralogical, geochemical and geochronological dataset for each grain. Fundamental to this approach is the streamlining of all steps from sample preparation to analysis with emphasis on sample-throughput which requires a high degree of automation with respect to analysis. Automation is drastically increased by exploiting the object detection capabilities of convoluted neural networks (CNN) to identify mineral grains in optical microscopic images and to collect spatial information (e.g. position, size, shape) of hundreds of thousands of grains within hours. This spatial information is used to further automate analysis by Raman spectroscopy, electron mircoprobe (EMP) and laser ablation inductively coupled plasma mass spectroscopy (LA-ICP-MS). Applying the proposed procedure will allow to investigate fine-grained sedimentary deposits at so far unprecedented spatial, temporal and/or compositional resolution. For instance hemi-pelagic to pelagic mudstone successions form extensive records of Earth history, but are usually under-investigated in terms of sediment provenance, especially single-grain SPA. Terrestrial fine-grained deposits like loess-paleosol-sequences carry relevant paleoclimatic information (dust concentration, transport direction, source, etc.), often at very high temporal resolution. Being able to access such deposits with mentioned resolution will pave new pathways in sediment research. To demonstrate the feasibility of our approach we devised three case studies that investigate (1) the contemporaneous provenance variability and paleoclimatic implications of loess-paleosol-sequences across a SW-NE transect through central Germany (2) the preservation potential of fine-grained unstable heavy minerals in Triassic mudrocks affected by significant burial and (3) the timing of diagenesis and/or low-grade metamorphism by documenting the transfer of µm-scaled outgrowths of zircon and xenotime as well as micro-monazite into the fine-silt fraction of modern sediments.

• DFG Programme: Research Grants