Phosphate Mineral Deposits Characterization Using Multivariate Data and SOM-based Processing

Wednesday, 17 December 2014
Lucas P Moreira1, Luciano S da Cunha2, Michael J Friedel3, Jose E Campos2 and Fabio C de Mendonça4, (1)Catholic University of Brasilia, Civil Engineering Department, Brasilia, Brazil, (2)UNB University of Brasilia, Geosciences Institute, Asa Norte, Brazil, (3)GNS Science, Lower Hutt, New Zealand, (4)DuSolo Fertilizers, Brasilia, Brazil
Phosphate deposits provide an important primary nutrient for fertilizer and agricultural industries worldwide in addition to feedstock for phosphate chemical processing plants. Phosphate mineral formation as well as its concentration may vary in tropical areas, due to strong weathering processes. Phosphate mineralization at Bonfim Hill, Central North Brazil, is stratiform, lens-shaped with deposition controlled by paleochannels and erosion controlling structures. Identification and characterization of phosphate mineral deposit at Bonfim Hill were performed analyzing geochemistry, electrical resistivity, x-ray fluorescence, mineral types and lithotypes using Kohonen's unsupervised neural network, the so-called self organizing maps (SOM). SOM-based data analysis enables and facilitates thorough investigations of multivariate data systems and provide additional statistical information compared to traditional methods.

The geochemical and geophysical data set was also used to train and validate a SOM-based classification system to detect phosphate mineral deposit, achieving 78% of correct classification.