Skip to main content

Research & Development
Industrial AI blog

Robust subsurface modeling for high-precision oil and gas exploration

8 May 2020

Masaki Hamamoto
Research & Development Group, Hitachi, Ltd.


We developed a subsurface modeling technique that works with poor-condition data through a collaborative research with Universiti Teknologi PETRONAS (UTP). Our technique contributes to accurate seismic imaging for oil and gas exploration.


Today, fossil fuels are still a main source of energy for society. To ensure a stable supply while minimizing related costs such as exploration and environmental costs, technologies such as seismic imaging technology that visualizes subsurface structure and leads to an accurate estimation of oil and gas reserves are essential in oil and gas exploration. This not only contributes to reducing drilling costs but also minimizes environmental cost caused by the drillings. With the recent rapid improvement of computer performance, a modeling technique named full waveform inversion (FWI) that can build high-resolution velocity model of the subsurface has started to become practical [1][2]. Since the high-resolution velocity model is essential for highly accurate seismic imaging, recently FWI has also become increasingly important.
To build an accurate velocity model by using a typical FWI, the availability of low-frequency components of seismic signal is the key to success [3]. However, the low-frequency components are not available in many cases because of the poor signal-to-noise ratio of recorded seismic data especially at low frequencies. Therefore, we started tackling this problem through the collaborative research with UTP.

Approach to the challenge

Why is low frequency necessary? That was our first question when we started the research activity. FWI estimates the velocity model by iteratively updating the model so that the misfit between observed and computed seismic data becomes smaller. Here, to mitigate the risk to be trapped in a local minimum, the update starts with building a low-resolution model by applying low frequencies and a high-resolution model can be obtained by increasingly applying higher frequencies to the low-resolution model. Thus, the low frequency plays a crucial role in FWI.
We therefore took an approach that randomly producing many various low-resolution models, applying FWI to them without using low frequency and finding the best model among the results. More specifically, as shown in Fig. 1, low-resolution models are obtained by a genetic algorithm (denoted as Coarse-grid GA) with high-resolution models and the high-resolution models are obtained by a typical FWI (denoted as Fine-grid LS) with the low-resolution models. This loop incrementally refines the estimated model.

Figure 1:Overview of the developed technique.


We evaluated our technique with the Marmousi model [4] and the results are shown in Fig. 2. Compared with the estimated model (c), the estimated model (d) obtained by applying our technique better fits to the true model (a).

Figure 2:(a) The true model, (b) a sample of initial models, (c) the estimated model by a conventional FWI, and (d) the estimated model by our technique.


Through the collaborative research with UTP, we developed an FWI technique that works under the condition that low-frequency components are not available. Our research work contributes to improve the accuracy of seismic imaging and this may lead to raising not only the economic value but also the environmental value in our society by reducing the number of drillings in oil and gas exploration.

For more details, we encourage you to read our paper, “Full Waveform Inversion based on Genetic Local Search Algorithm with Hybrid-Grid Scheme”.


A. Tarantola, “Inversion of seismic reflection data in the acoustic approximation,” Geophysics, Vol. 49, pp. 1259-1266, 1984.
P. Mora, “Nonlinear two-dimensional elastic inversion of multioffset seismic data,” Geophysics, Vol. 52, pp. 1211-1228, 1987.
J. Virieux and S. Operto, “An overview of full-waveform inversion in exploration geophysics,” Geophysics, Vol. 74, WCC1-WCC26, 2009.
R. Versteeg, “The Marmousi experience: Velocity model determination on a synthetic complex data set,” The Leading Edge, Vol. 13, pp. 927-936, 1994.