Unsupervised clustering for prospectivity mapping of lithium pegmatites in Leinster, Ireland: A transferable workflow

This study aims to assess the transferability of unsupervised clustering with subsequent data-driven pegmatite prospectivity assignment to the under-explored south Leinster region, Ireland, where consistent genetic or targeting models for spodumene-bearing pegmatites are limited by very poor (< 1 %) outcrop. We apply self-organizing maps to simple, reproducible, and machine-readable features derived from airborne magnetic and radiometric data that capture lithological variability and tectonic stress pattern, deliberately excluding data sets potentially affected by displacement, anthropogenic activity or glacial overprint. The methodology segments the survey area by unsupervised clustering without using known pegmatite occurrences as training input. Probabilistic prospectivity maps are generated by aggregating multiple clustering realizations with known pegmatite occurrences. The results show that the approach reliably separates pegmatite-bearing from pegmatite-free boreholes and delineates a focused set of prospective zones, mainly along the eastern margin of the Leinster Granite, within the East Carlow Deformation Zone and along other major faults. Treating barren pegmatites separately further improves discrimination of targets. We conclude that airborne magnetic and radiometric data contain exploitable information on pegmatite emplacement, as they capture lithological variability and structural patterns in the near-surface ground, and that our approach provides a robust framework for early-stage exploration in data-poor regions.