Reprocessing and interpretation of legacy seismic data using machine learning from the Granada Basin, Spain

Abstract

The Granada Basin (Spain) is a Neogene sedimentary depression with irregular geomorphology and deep depocenters. It is located in the most seismically hazardous part of the Iberian Peninsula with an historically experienced extremely destructive earthquakes, followed by periods of low to moderate seismicity. In 1980s the Chevron Oil Company collected a set of 30 deep seismic reflection sections in this Basin of which only the results on paper are kept. Due to the fact that many of these seismic profiles are currently located in urban areas and the economic cost of carrying out a similar exploration, it was decided to recover these old data and apply a post-stack treatment to improve their quality. The purpose of this study is to show the applied reprocessing flow and, with the new sections, to present a spatial model of the basin. The first stage of recovery and enhacement of seismic sections has consisted in three phases: first, high-resolution scanning of paper copies to TIFF images followed by the transformation of TIFF images to SEG-Y format; second, poststack processing workflow to increasing resolution and lateral coherence of these seismic lines; and third, it has been used a machine learning algorithm, among others, increasing the spatial resolution, signal-to-noise ratio, and coherence of the seismic signals. In addition, basement horizons, as well as three sedimentary sequences, were identified in all seismic sections and interpolated to create a three-dimensional basement model composed by normal faults, horst and grabens related to the seismotectonic behavior of the basin. As an overall assessment, this work is an example of the usefulness of ‘recycling’ legacy seismic data, which nowadays are usually in archived boxes, but at the time required a great economic and acquisition effort.