A 10.5 km² 3D seismic survey was acquired over the Kylylahti mine area (Outokumpu mineral district, eastern Finland) as a part of the COGITO-MIN (COst-effective Geophysical Imaging Techniques for supporting Ongoing MINeral exploration in Europe) project, which aimed at the development of cost-effective geophysical imaging methods for mineral exploration. The cost-effectiveness in our case was related to the fact that an active-source 3D seismic survey was accomplished by using the receiver spread originally designed for a 3D passive survey. The 3D array recorded Vibroseis and dynamite shots from an active-source 2D seismic survey, from a vertical seismic profiling experiment survey, as well as some additional “random” Vibroseis and dynamite shots made to complement the 3D source distribution. The resulting 3D survey was characterized by irregular shooting geometry and relatively large receiver intervals (50 m). Using this dataset, we evaluate the effectiveness of the standard time-imaging approach (post-stack and pre-stack time migration) compared to depth imaging (standard and specialized Kirchhoff pre-stack depth migration, KPreSDM). Standard time-domain processing and imaging failed to convincingly portray the first ~1500 m of the subsurface, which was the primary interest of the survey. With a standard KPreSDM, we managed to obtain a good image of the base of the Kylylahti formation bordering the extent of the mineralization-hosting Outokumpu assemblage rocks, but otherwise the image was very noisy in the shallower section. The specialized KPreSDM approach (i.e., coherency-based Fresnel volume migration) resulted in a much cleaner image of the shallow, steeply dipping events, as well as some additional deeper reflectors, possibly representing repetition of the contact between the Outokumpu assemblage and the surrounding Kalevian metasediments at depth.

We present a method for automatic detection and classification of seismic events from continuous ambient‐noise (AN) recordings using an unsupervised machine‐learning (ML) approach. We combine classic and recently developed array‐processing techniques with ML enabling the use of unsupervised techniques in the routine processing of continuous data. We test our method on a dataset from a large‐number (large‐$N$⁠) array, which was deployed over the Kylylahti underground mine (Finland), and show the potential to automatically process and cluster the volumes of AN data. Automatic sorting of detected events into different classes allows faster data analysis and facilitates the selection of desired parts of the wavefield for imaging (e.g., using seismic interferometry) and monitoring. First, using array‐processing techniques, we obtain directivity, location, velocity, and frequency representations of AN data. Next, we transform these representations into vector‐shaped matrices. The transformed data are input into a clustering algorithm (called $k$‐means) to define groups of similar events, and optimization methods are used to obtain the optimal number of clusters (called elbow and silhouette tests). We use these techniques to obtain the optimal number of classes that characterize the AN recordings and consequently assign the proper class membership (cluster) to each data sample. For the Kylylahti AN, the unsupervised clustering produced 40 clusters. After visual inspection of events belonging to different clusters that were quality controlled by the silhouette method, we confirm the reliability of 10 clusters with a prediction accuracy higher than 90%. The obtained division into separate seismic‐event classes proves the feasibility of the unsupervised ML approach to advance the automation of processing and the utilization of array AN data. Our workflow is very flexible and can be easily adapted for other input features and classification algorithms.

The main issues related to passive-source reflection imaging with seismic interferometry (SI) are inadequate acquisition parameters for sufficient spatial wavefield sampling and vulnerability of surface arrays to the dominant influence of the omnipresent surface-wave sources. Additionally, long recordings provide large data volumes that require robust and efficient processing methods. We address these problems by developing a two-step wavefield evaluation and event detection (TWEED) method of body waves in recorded ambient noise. TWEED evaluates the spatiotemporal characteristics of noise recordings by simultaneous analysis of adjacent receiver lines. We test our method on synthetic data representing transient ambient-noise sources at the surface and in the deeper subsurface. We discriminate between basic types of seismic events by using three adjacent receiver lines. Subsequently, we apply TWEED to 600 h of ambient noise acquired with an approximately 1000-receiver array deployed over an active underground mine in Eastern Finland. We develop the detection of body-wave events related to mine blasts and other routine mining activities using a representative 1 h noise panel. Using TWEED, we successfully detect 1093 body-wave events in the full data set. To increase the computational efficiency, we use slowness parameters derived from the first step of TWEED as input to a support vector machine (SVM) algorithm. Using this approach, we detect 94% of the TWEED-evaluated body-wave events indicating the possibility to limit the illumination analysis to only one step, and therefore increase the time efficiency at the price of lower detection rate. However, TWEED on a small volume of the recorded data followed by SVM on the rest of the data could be efficiently used for a quick and robust (real-time) scanning for body-wave energy in large data volumes for subsequent application of SI for retrieval of reflections.

Passive seismic experiments have been proposed as a cost-effective and non-invasive alternative to controlled-source seismology, allowing body–wave reflections based on seismic interferometry principles to be retrieved. However, from the huge volume of the recorded ambient noise, only selected time periods (noise panels) are contributing constructively to the retrieval of reflections. We address the issue of automatic scanning of ambient noise data recorded by a large-N array in search of body–wave energy (body–wave events) utilizing a convolutional neural network (CNN). It consists of computing first both amplitude and frequency attribute values at each receiver station for all divided portions of the recorded signal (noise panels). The created 2-D attribute maps are then converted to images and used to extract spatial and temporal patterns associated with the body–wave energy present in the data to build binary CNN-based classifiers. The ensemble of two multi-headed CNN models trained separately on the frequency and amplitude attribute maps demonstrates better generalization ability than each of its participating networks. We also compare the prediction performance of our deep learning (DL) framework with a conventional machine learning (ML) algorithm called XGBoost. The DL-based solution applied to 240 h of ambient seismic noise data recorded by the Kylylahti array in Finland demonstrates high detection accuracy and the superiority over the ML-based one. The ensemble of CNN-based models managed to find almost three times more verified body–wave events in the full unlabelled dataset than it was provided at the training stage. Moreover, the high-level abstraction features extracted at the deeper convolution layers can be used to perform unsupervised clustering of the classified panels with respect to their visual characteristics.

For the first time, we apply a full-scale 3D seismic virtual-source survey (VSS) for the purpose of near-mine mineral exploration. The data were acquired directly above the Kylylahti underground mine in Finland. Recorded ambient noise (AN) data are characterized using power spectral density (PSD) and beamforming. Data have the most energy at frequencies 25–90 Hz, and arrivals with velocities higher than 4 km s⁻¹ have a wide range of azimuths. Based on the PSD and beamforming results, we created 10 d subset of AN recordings that were dominated by multi-azimuth high-velocity arrivals. We use an illumination diagnosis technique and location procedure to show that the AN recordings associated with high apparent velocities are related to body-wave events. Next, we produce 994 virtual-source gathers by applying seismic interferometry processing by cross-correlating AN at all receivers, resulting in full 3D VSS. We apply standard 3D time-domain reflection seismic data processing and imaging using both a selectively stacked subset and full passive data, and we validate the results against a pre-existing detailed geological information and 3D active-source survey data processed in the same way as the passive data. The resulting post-stack migrated sections show agreement of reflections between the passive and active data and indicate that VSS provides images where the active-source data are not available due to terrain restrictions. We conclude that while the all-noise approach provides some higher-quality reflections related to the inner geological contacts within the target formation and the general dipping trend of the formation, the selected subset is most efficient in resolving the base of formation.