Identifying the geochemical fingerprint of volcanic material in soils of distal areas using a machine-learning approach

• Three machine-learning algorithms were used to classify soil samples. • Background values were calculated for 48 elements in the different identified soil types. • The background values are highly variable depending on the type of soil. • Volcanic soils have natural background concentrations higher than the legal thresholds. • The algorithms enabled the identification of volcanic soil patterns in distal areas. In this study, the Campania region (Italy) was selected to test a novel approach for identifying the geochemical signature of volcanic material in distal soils using their chemical composition. The Campania soil database comprises analyses of 48 elements for 5553 samples. Previous studies allowed us to confidently label 1277 samples as volcanic soils and 353 as non-volcanic soils. These labeled samples were used to train three machine learning algorithms to classify 3903 uncertain samples. Three different soil types were effectively identified with 98 % accuracy: volcanic, non-volcanic, and mixed. Subsequently, regional geochemical background values for each element in the various identified soil types were determined using ProUCL software. The results show that volcanic soils have background values of some key macronutrients (K, Na) and potentially toxic elements (As, Be, Hg, Pb, U, Tl) up to 18 times higher than non-volcanic soils. On the contrary, non-volcanic soils show the geochemical signature of materials of carbonate and clay origin, with enrichments of Ca, Mg, Co, Mn, Ni up to 4 times higher than volcanic soils. All these findings are fundamentally important for accurately establishing local reference background concentration values, which are crucial for promoting sustainable soil management practices. Moreover, the geochemical information generated by this study also yielded valuable insights into the geographic distribution of pyroclastic fallout from ancient eruptions, which is essential for understanding the historical dynamics of volcanic activity in the region.