Real‐time deep‐learning inversion of seismic full waveform data for CO₂ saturation and uncertainty in geological carbon storage monitoring

Abstract

Deep learning inversion has recently drawn attention in geological carbon storage research due to its potential of imaging and monitoring carbon storage in real time, significantly improving efficiency and safety of carbon storage operations. We present a deep-learning full waveform inversion method that after the neural network has been trained can image CO2 saturation and its uncertainty in real time. Our deep learning inversion method is based on the U-Net architecture with the neural network trained on pairs of synthetic seismic data and CO2 saturation models. Accordingly, our training establishes a mapping relationship between seismic data and CO2 saturation models and once fully trained directly estimates CO2 saturation as a function of subsurface location. We further quantify uncertainties of CO2 saturation estimates using the Monte Carlo dropout method and a bootstrap aggregating method. For this proof-ofconcept study, the CO2 training models and data are derived from the Kimberlina 1.2 model, a hypothetical 3D geological carbon storage model that is constructed based on various geological and hydrological data from the Southern San Joaquin Basin, California. We perform deep-learning inversion experiments using noise-free and noisy training and test data sets and compare the results. Our modeling experiments show that 1) the deep-learning inversion can estimate 2D distributions of CO2 fairly well even in the presence of Gaussian random noise and 2) both CO2 saturation imaging and uncertainty quantification can be done in real time. Our results suggest that the deep-learning inversion method can serve as a robust real-time monitoring tool for geological carbon storage and/or other time varying reservoir/aquifer properties that result from injection, extraction, and/or other subsurface transport phenomena.

The Kimberlina synthetic multiphysics dataset for CO2 monitoring investigations

“…We note the strong response generated by the introduction of the CO 2 plume. As mentioned previously, these data served as the test data for the ML algorithms described in Wu and Lin (2019) and Um et al (2022).…”

Deep learning multiphysics network for imaging CO2 saturation and estimating uncertainty in geological carbon storage

“…For example, when only one type of geophysical data is available, the encoder for that type of data is built but the other two encoders are not. In this case (Figure 2), the DL network reduces to the classic U‐Net architecture (Ronneberger et al., 2015) and can be used as a DL single‐physics network for imaging CO 2 saturation (Um et al., 2022). The details of their implementation will be described later.…”

“…Recently, deep learning (DL) imaging has drawn attention in computational geophysics as it overcomes some of the main drawbacks that traditional inversion exhibits (Araya‐Polo et al., 2018; Colombo et al., 2020; Kaur et al., 2021; Li & Yang, 2021; Puzyrev, 2019; Um et al., 2022; Wu & Lin, 2019; Yang & Ma, 2019; Yang et al., 2022; Zhang & Alkhalifah, 2019; Zhang & Lin, 2020). A deep neural network is trained such that it can learn complex non‐linear correlations between earth models and corresponding geophysical data.…”

“…Nonetheless, using DL imaging does not remove the computational challenges of the multiphysics CO 2 monitoring problem. For example, in order to generate a realistic set of DL training models and data, one needs to simulate a number of CO 2 flow models and their geophysical counterparts by solving their governing partial differential equations (Um et al., 2022; Zeng et al., 2021). The flow simulation cost may be reduced by simply inserting many different shapes of CO 2 bodies into a background model (Puzyrev, 2019; Yang et al., 2022).…”

“…The network architectures are designed to exploit seismic, electromagnetic and gravity data separately as well as together. To directly image CO 2 saturation rather than geophysical proxy properties (e.g., P wave velocity, electrical resistivity or density), the network is trained on pairs of CO 2 saturation label models and associated geophysical monitoring data (Um et al., 2022). The CO 2 saturation label models are generated by simulating CO 2 injection and flow over a three‐dimensional CO 2 storage model that was constructed based on real geologic and hydrogeologic data.…”