Consistency and prior falsification of training data in seismic deep learning: Application to offshore deltaic reservoir characterization

TL;DR

This paper investigates the consistency of synthetic seismic data used in deep learning for reservoir characterization, emphasizing prior falsification, regularization strategies, and validation with real seismic data.

Contribution

It introduces a prior falsification approach to ensure synthetic data aligns with real seismic data and proposes regularization methods to prevent CNN overfitting on synthetic datasets.

Findings

Regularization improves CNN robustness on real seismic data

Prior falsification enhances synthetic data reliability

Proposed strategies reduce overfitting to synthetic data

Abstract

Deep learning applications of seismic reservoir characterization often require generation of synthetic data to augment available sparse labeled data. An approach for generating synthetic training data consists of specifying probability distributions modeling prior geologic uncertainty on reservoir properties and forward modeling the seismic data. A prior falsification approach is critical to establish the consistency of the synthetic training data distribution with real seismic data. With the help of a real case study of facies classification with convolutional neural networks (CNNs) from an offshore deltaic reservoir, we highlight several practical nuances associated with training deep learning models on synthetic seismic data. We highlight the issue of overfitting of CNNs to the synthetic training data distribution and propose regularization strategies to address it. We demonstrate the efficacy of our proposed strategies by training the CNN on synthetic data and making robust predictions with real 3D partial stack seismic data.