Turbidite Probability Scenarios Generation Combining Generative Models and Geostatistical Techniques

The identification of characteristics of turbidite deposits is important for the oil and gas industry, as they represent possible oil reservoirs. The indirect measurement of turbidites through sonars can be a cause of several uncertainties in the reservoir data. This work aims to improve the data of turbidity reservoirs and quantify their uncertainties. For this, a probabilistic sandstone map of a turbiditical basin was used and a method based on geostatistics, kriging, and machine learning techniques, especially SinGAN, was developed to generate an ensemble of probabilistic sandstone maps. In order to analyze the performance of the solution, evaluation metrics such as continuous ranked probability score (CRPS) and receiver operator characteristic's area under the curve (ROC AUC) were used. The results showed that the method developed has high potential of producing trustworthy maps from a single original data and that it can be a solution to improve the measurement of probabilistic reservoirs.