Subsurface porosity estimation: a case study from the Krishna Godavari offshore basin, eastern Indian margin, India

Abstract Porosity is one of the fundamental petrophysical parameters essential for appraisal of hydrocarbon reserve and planning production operations. Several interpretation techniques have evolved over time that allows geologists to meticulously estimate reservoir properties by integrating seismic and borehole data. The present study investigates the use of Multi-Attribute Transform Method (MATM) and Artificial Neural Networks (ANN) for the delineation of lateral variation of porosity along a 2D seismic reflection dataset in the gas hydrate inferred area of Krishna–Godavari basin, eastern Indian margin. The best set of attributes that can predict the porosity are derived using multi-attribute regression analysis. These attributes are applied in MATM and are also used to train the neural networks. This is achieved using different networks of neural computation like Probabilistic Neural Network (PNN), Multi-Layer Feedforward Network (MLFN) and Radial Basis Function (RBFN) with an aim to decipher an optimized approach best suits to predict porosity of the study area. The correlation between observed and predicted porosity logs are tested both by cascading and non-cascading the networks with the MATM. The predictive power is found improved as we progress from MATM to neural network with the trend of MATM. Finally, it is observed that PNN cascaded with the results of MATM shows the least prediction error than the other networks