Multi-attribute and Artificial Neural Network Analysis of Seismic Inferred Chimney-like Features in Marine Sediments: A Study from KG Basin, India

Abstract

In marine seismic sections, features like “chimneys”, characterized by low similarity chaotic zones or vertical disordered characteristics arise due to migration of gas. Numerous interpretation tools have been developed for the identification of chimney features from seismic data. Here, a study over a 2D seismic reflection profile, acquired by CSIR-National Geophysical Research Institute (NGRI), Hyderabad in the Krishna-Godavari (KG) basin in 2010 was carried out. The data is optimally conditioned or post-processed before extracting the attributes such as the similarity, dip variance, energy, and frequency washout, to capture the chimneys on seismic data. Finally, the attributes are combined and trained over interpreter’s acquaintances through a non-linear multi-layer perceptron (MLP) neural network to compute a hybrid attribute, defined as the chimney meta-attribute. The neural training results into an overall normalized root mean squared (RMS) error of 0.6 to 0.7 between the train and test data sets. Three chimneys of ∼1.5, ∼0.9, ∼1.3 km heights are observed over the seismic line. As chimneys represent gas migration paths, the computation of chimney attribute helps in understanding the genesis of gas-hydrates not only in the KG basin, studied here, but also in any other prospective basin.