Generating clusters for turbidite probability maps using machine learning methods.

Turbidite deposits are of great importance because of their correlation with the existence of oil reservoirs. Therefore, the spatial distribution of these deposits is frequently studied, considering the turbidite probability maps. The uncertainty of the original turbidite data, captured indirectly by sonars, is quantified by generation of possible scenarios of the probability map. This research aims, from the use of statistical characteristics and the creation of groups, through unsupervised machine learning algorithms, to identify the possible scenarios, generated by geostatistical methods, that are more similar to the original data considered in the study. Through the results, in addition to quantifying the uncertainties generating possible scenarios, it is verified that the methodology developed is capable of creating differentiated groups and identifying the group that has similar characteristics to the reference data, helping in the generation and identification of possible scenarios of turbiditic basins. The main novelty of this work is to combine geostatistical and machine learning methods, in addition to creating groups of images that have similar statistical characteristics. [ABSTRACT FROM AUTHOR]