Hyunggu Jun

The increasing exploitation of the subsurface has increased the demand for advanced time-lapse seismic monitoring technologies, which are critical for effective underground monitoring. Although machine-learning-based data processing techniques have significantly improved data repeatability, research on quantifying the uncertainty in the predictions generated by these models remains insufficient. This study proposed a machine-learning model for the quantitative evaluation of uncertainties in time-lapse seismic data. The aleatoric and epistemic uncertainties were independently quantified, providing a more detailed and objective assessment of prediction reliability and enabling the formulation of guidelines for additional data processing procedures to further enhance the repeatability of the time-lapse seismic data. The model was trained and validated using synthetic data generated based on the Sleipner monitoring data set and field data from Pohang, Korea. In synthetic experiments, low uncertainty values were obtained when time-lapse seismic differences were associated with actual subsurface changes, whereas elevated uncertainty was observed when the seismic differences were unrelated to true subsurface variations, confirming that the quantified uncertainties provided meaningful criteria for time-lapse seismic interpretation. In field-data applications, regions exhibiting time-lapse differences consistently showed high uncertainty, indicating that the observed differences might not be attributed to temporal subsurface changes. Furthermore, a geologic change indicator was introduced, defined as the relationship between the quantified uncertainties and data differences, which successfully identified the presence or absence of subsurface changes.

Jongpil Won, Hyunggu Jun. Econ. Environ. Geol. 2024;57:681-99. https://doi.org/10.9719/EEG.2024.57.6.681

Time‐lapse seismic data processing is an important technique for observing subsurface changes over time. The conventional time‐lapse seismic exploration has been conducted using a large‐scale exploration system. However, for efficient monitoring of shallow subsurface, time‐lapse monitoring based on the small‐scale exploration system is required. Small‐scale exploration system using a sparker source offers high vertical resolution and cost efficiency, but it faces challenges, such as inconsistent waveforms of sparker sources, inaccurate positioning information and a low signal‐to‐noise ratio. Therefore, this study proposes a data processing workflow to preserve the signal and enhance the repeatability of small‐scale time‐lapse seismic data acquired using a sparker source. The proposed workflow has three stages: pre‐stack, post‐stack and machine learning–based data processing. Conventional seismic data processing methods were applied to enhance the quality of the sparker seismic data during the pre‐stack data processing stage. In the post‐stack processing stage, the positions and energy correction were performed, and the machine learning–based data processing stage attenuated random noise and applied a matched filter. The data processing was performed using only the seismic signals recorded near the seafloor, and the results confirmed the improvement in the repeatability of the entire seismic profile, including that of the target area. According to the repeatability quantification results, the predictability increased and the normalized root mean square decreased during data processing, indicating improved repeatability. In particular, the repeatability of the data was greatly improved through vertical correction, energy correction and matched filtering approaches. The processing results demonstrate that the data processing method proposed in this study can effectively enhance the repeatability of high‐resolution time‐lapse seismic data. Consequently, this approach could contribute to a more accurate understanding of temporal changes in subsurface structure and material properties.

SUMMARY: To address complex subsurface structures, a high-resolution velocity model must be constructed. Conventionally, algorithms such as full waveform inversion (FWI) have been used to derive accurate high-resolution velocity structures, but obstacles such as high computational costs remain. Therefore, we propose a high-resolution U-NET (HR U-NET) machine learning model to derive a high-resolution velocity model from a low-resolution velocity model. The low-resolution velocity model and migration data obtained through the corresponding velocity information were used as input data for training. In addition, we tried to improve the accuracy of the high-resolution velocity model by using prior information containing accurate velocity values. A prior model generated through geophysical logging data and a weight model including the reliability information of the prior model were also utilized. Therefore, the HR U-NET model was trained using the low-resolution velocity model, the migration data, the prior model and the weight model. Numerical experiments conducted using synthetic and field data demonstrated that the proposed model could accurately construct a high-resolution velocity model and verified that the prior model and weight model play important roles in the training process. Additionally, we confirmed that the proposed method derived almost similar results using only 8.2 percent of the computational cost of the conventional inversion method. In other words, there is an advantage that it is possible to predict high-resolution velocity information more efficiently in terms of computational cost.

Jongpil Won, Jungkyun Shin, Jiho Ha, Hyunggu Jun. Econ. Environ. Geol. 2024;57:51-71. https://doi.org/10.9719/EEG.2024.57.1.51

Data acquired through seismic surveys suffers from information loss for various reasons. Among these, shallow gas affects signals beneath it and causes distortion of seismic wave signals. These distorted signals often cause difficulties in interpreting seismic data. Various preceding research has been conducted to mitigate this issue, and this study presents an interpolation method using self-supervised machine learning. The dataset we use consists of small-scale marine seismic survey data acquired from Youngil Bay near eastern continental margin of the Korean Peninsula. 2D planes are randomly extracted from 3D seismic data to generate training data of various sizes, which are then used to create an artificial neural network model. This study utilizes the U-Net++ architecture with an attention module to overcome the limitations associated with U-Net and to focus and emphasize repetitive patterns. In the training process, both the L1 loss function and the structural similarity index (SSIM) loss function are used as hybrid loss function. This enables the network to recognize both the overall patterns of the data and the key patterns to focus on. The neural network model created through this study shows successful performance in detecting patterns in the original data and interpolating missing values of seismic data. The output data generated through the proposed network improves stratigraphic continuity by interpolating the missing values of the original data and restoring the lost structure.

Seismic random noise is one of the main factors that degrade the quality of seismic data. Therefore, seismic random noise attenuation should be performed appropriately through several stages during seismic data processing, and this requires sufficient experience and knowledge because the proper hyperparameters need to be determined based on the features of the noise in the target seismic data. Recently, machine learning–based seismic noise attenuation has been widely studied because it provides suitable results by learning noise features from noisy data, unlike conventional physics‐based approaches. There are many important factors in machine learning, and, here, we focus on the loss functions of machine learning in terms of seismic random noise attenuation. The most widely used loss function is l₂, but we train a model with various kinds of single and multiple loss functions and attenuate seismic random noise. We analyse the efficiency of loss functions by comparing the noise‐attenuated results of synthetic and field seismic data qualitatively and quantitatively. Our analysis indicates that the multiple loss function with the l₁ norm can be a proper choice for random noise suppression of seismic data.