Drone-based magnetic survey systems are becoming innovative geophysical exploration tools capable of acquiring high-resolution geomagnetic data in diverse environments. However, fluxgate magnetometers mounted on drones suffer from signal degradation due to the platform's inherent magnetism, electromagnetic interference from electronic equipment, flight speed, vibrations and attitude changes during surveys. This study aims to analyze the major noise sources in drone systems and to compare and analyze magnetic compensation algorithms, originally proposed for shipborne and airborne platforms, to optimize them for drone applications. For this purpose, survey data, including figure-eight correction flights, were collected and analyzed. The acquired magnetic data underwent preprocessing, including FFT analysis to identify frequency characteristics, low-pass filtering, and sampling correction. Subsequently, rotation matrices were comparatively analyzed for drone attitude correction. Based on this, three magnetic correction methods were evaluated. The results indicated that the Improved correction method of shipborne three-component magnetometer showed the highest accuracy and a residual distribution closest to a normal distribution for the vector three-component data, while the calibration method of satellite vector magnetometer demonstrated the best performance for scalar magnetic data. We confirmed that similar magnetic anomalies are exhibited along the additional parallel survey lines. It was also observed that noise tends to increase with faster flight speeds. This is interpreted as the drone's attitude becoming unstable at higher speeds, leading to irregular increases in the motor current supplied to the rotors to maintain stability, which in turn generates electromagnetic noise recorded by the sensor. This study contributes to improving the quality of drone-based magnetic survey data by comparing correction methods, applied various platforms and presenting a systematic data processing algorithm. It is expected to contribute to the practical application of drone magnetometer systems in diverse fields such as resource exploration and environmental investigation in the future.
Offshore wind power has recently become a key component in global energy transition strategies, increasing the importance of acquiring accurate marine subsurface information for structural design. For the safe and cost-effective development of offshore wind farms, marine geophysical surveys are essential for resolving the spatial characteristics of the seabed and shallow sub-seabed at high resolution. In Korea, standards for marine geophysical investigations specific to offshore wind farm development remain limited; therefore, identifying and analyzing international standards is a priority. Accordingly, this study reviews the international standards for marine geophysical survey methods in offshore wind farm development and systematically compiles the technical specifications of the latest survey equipment. The aim is to provide baseline information to support future technological advancements and the standardization of marine geophysical surveys in Korea.
Distributed Acoustic Sensing (DAS) systems are a promising technology for microseismic monitoring due to their high spatial resolution and broad detection coverage. However, real-time monitoring generates large volumes of data and therefore require efficient data management, signal detection, and classification. This study aimed to develop efficient data processing procedures and optimal machine learning models for near-real-time microseismic monitoring. Using DAS data from the Utah FORGE geothermal project, we systematically applied traditional data processing procedures step-by-step and performed signal detection based on the VGG-19 and YOLO v11 models to compare and analyze their performance. A quantitative performance evaluation was conducted using confusion matrices, accompanied by processing-time analysis for each data processing procedure, and model-selection considerations for monitoring applications were examined. Additionally, signal-detection models were applied to data processed with machine-learning-based denoising, and the results were compared to evaluate model performance. This study confirms that developing a practical near-real-time monitoring system requires an optimal combination of lightweight machine learning models and traditional data processing procedures, with comprehensive consideration of model performance, processing time, and data-preparation efficiency.
Distributed acoustic sensing (DAS) systems utilize optical fiber cables as seismic receivers by emitting light pulses and detecting the resulting minute deformations or vibrations along the fiber through an interrogator unit, thereby enabling the measurement of both strain magnitude and spatial location. While DAS exhibits high sensitivity and can capture even weak seismic signals, it often suffers from low signal-to-noise ratios due to various types of noise, including distinctive patterns not typically found in conventional seismic data. Therefore, target-oriented processing strategies are essential for effective noise suppression. In this study, we focus on common-mode noise, also referred to as horizontal stripe noise, which is caused by external factors such as mechanical vibrations acting on the interrogator. This noise appears as uniform horizontal patterns simultaneously recorded across all channels at specific times and must be removed prior to applying standard seismic processing workflows. Although conventional frequency-wavenumber domain filters can attenuate such noise, their performance often depends heavily on manually selected parameters and may result in unintended signal loss. To address these limitations, we propose a novel method that minimizes signal degradation while effectively isolating common-mode noise. Because common-mode noise exhibits spatial invariance across channels, randomly shuffling the traces disrupts coherent signal components while leaving the noise pattern unchanged. A time-frequency domain filter is then applied to the randomized data to suppress the now-randomized signals and extract the spatially coherent noise. The process is iteratively repeated to enhance robustness. The proposed method is validated on both synthetic and field DAS datasets obtained from a borehole installation, demonstrating effective suppression of common-mode noise with minimal impact on the underlying seismic signals.
Seismic data regularization is a critical step that improves the quality of subsequent data processing. We present a case study applying a deep learning (DL)-based regularization method using Delaunay tessellation proposed by Yeeh (2024) to small-scale, 3D ocean bottom node (OBN) data acquired from irregularly positioned sources in Yeongil Bay, Pohang. The DL model for regularization was trained solely on the preprocessed field data. We describe the data preprocessing and training data selection strategies used to enhance model performance. The trained DL model regularized the irregularly located source positions to a 10-meter spacing, which is denser than the originally acquired spacing of 25-50 meters. Except for a few receiver nodes lacking nearby sources in near-offset regions where the curvature of arrival times varied, the main signals were well predicted. The results also effectively suppressed random noise. Because this method enables both the interpolation and regularization of irregularly acquired 3D seismic data using only field data, it is highly adaptable for future geophysical applications.
