Generative Geostatistical Modeling from Incomplete Well and Imaged Seismic Observations with Diffusion Models

Huseyin Tuna Erdinc,Rafael Orozco,Felix J. Herrmann

2024-05-17

Abstract:In this study, we introduce a novel approach to synthesizing subsurface velocity models using diffusion generative models. Conventional methods rely on extensive, high-quality datasets, which are often inaccessible in subsurface applications. Our method leverages incomplete well and seismic observations to produce high-fidelity velocity samples without requiring fully sampled training datasets. The results demonstrate that our generative model accurately captures long-range structures, aligns with ground-truth velocity models, achieves high Structural Similarity Index (SSIM) scores, and provides meaningful uncertainty estimations. This approach facilitates realistic subsurface velocity synthesis, offering valuable inputs for full-waveform inversion and enhancing seismic-based subsurface modeling.

Geophysics,Artificial Intelligence,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

This paper attempts to address the problem of how to synthesize high-fidelity subsurface velocity models using incomplete well data and seismic observation data in subsurface velocity model generation. Traditional methods usually require large, high-quality datasets, which are often difficult to obtain in practical applications. Therefore, this paper proposes a method based on a diffusion generative model that can generate samples highly similar to real subsurface velocity models with only a small amount of well data and seismic images, and can provide meaningful uncertainty estimates.

Specifically, the main objectives of this study include:
1. **Addressing the issue of data insufficiency**: By introducing additional noise and subsampling techniques, the model can be trained under conditions of incomplete data, thus avoiding reliance on fully sampled data.
2. **Generating high-fidelity samples**: The subsurface velocity models generated by the diffusion generative model can accurately capture long-range structures, are highly consistent with real models, and perform excellently in terms of structural similarity index (SSIM).
3. **Providing uncertainty estimates**: The generated models not only provide high-fidelity velocity samples but also reflect the uncertainties arising from incomplete data and the imaging process.

These achievements are of great significance for full waveform inversion and subsurface modeling based on seismic data, providing valuable input for subsequent geological exploration and resource assessment.