“…Coblentz, 1906) and its applications in geology are well established (Lyon & Burns, 1963). More recently, near-infrared remote sensing has been applied in soil sciences for multiple purposes, including: mapping proportions of sand, silt, and clay content (Saleh, Belal & Arafat, 2013;Silva et al, 2016); determination of soil salinity (Feyziyev et al, 2016); monitoring soil moisture and capacity of water absorption (Ben-Dor et al, 2002;Whiting, Li & Ustin, 2004); estimate organic carbon content (Hu, Chau & Si, 2015;Viscarra Rossel & Hicks, 2015); differentiate types of clay minerals (Surech, Sreenivas & Sivasamy, 2014); and assessing soil contamination and detection of heavy metals (Mohamed et al, 2016(Mohamed et al, , 2018). Yet, until now, machine learning approaches in geospatial paleontology tended to achieve good results but as "black-boxes" that did not reveal how exactly they were interpreting a "fossiliferous signal", and thus the relative importance of specific spectral bands for remote fossil site detection has not previously been demonstrated.…”