Magdalena Markovic
Assistant Professor
Research Interests
Make your next move.
Prepare your application
Review drafts before you send them.
Explore related searches
About
Magdalena Markovic is an Assistant Professor at Uppsala University, Sweden. Her research focuses on seismic imaging, geological carbon storage, and the exploration of critical raw materials. Recent publications include advancements in multi-scale unsupervised learning for seismic applications and investigations of iron oxide deposits. Markovic's work contributes significantly to the fields of geophysics and geotechnical engineering.
Articles (16)
Year:
2026
Surface fibre-optic distributed acoustic sensing (S-DAS) typically requires trenching to achieve adequate coupling, which can be challenging and costly in hardrock environments. As an alternative, untrenched S-DAS offers significant time and cost savings, though at the expense of data quality. In this study, we systematically evaluated the impact of untrenched deployment on DAS data quality through a multi-method comparison with conventional sensors, including three-component geophones and vertical-component accelerometers. All recorded data were converted to particle velocity to enable direct amplitude comparisons across arrays. Phase fidelity was assessed using the surface-wave method, cross-power spectral analysis, and ambient noise interferometry. Untrenched S-DAS recorded amplitude levels that differed from those of conventional sensors, particularly above 50 Hz. However, within the surface-wave band (up to 35 Hz), the data quality was sufficient to derive reliable shear-wave velocity profiles, especially when the data were resorted to common-receiver gathers. Sensitivity and coherence decreased significantly at higher frequencies and larger offsets, and a systematic time delay was observed for DAS in the surface-wave band. Ambient noise interferometry was ineffective for the untrenched DAS array, largely due to variable channel coupling and system-specific noise. This study provides a systematic field comparison of untrenched S-DAS and conventional sensors in hardrock settings, outlining both the limitations and the practical potential of this cost-effective deployment method.
Year:
2026
Year:
2026
High‐resolution seismic data are essential for interpreting thin‐layered stratigraphy and subtle structures within hardrock media, as this information can lead to better exploration decisions in the mining sector. Conventional resolution enhancement techniques, such as spectral broadening and supervised deep‐learning techniques, often rely on oversimplified assumptions or require high‐resolution training labels. These limitations restrict their applicability in real seismic data processing, especially for hardrock seismic data, which are typically characterized by high velocities, strong heterogeneity and short reflector continuity due to complex emplacement contacts and geological settings. We propose an unsupervised seismic resolution enhancement framework that integrates physics‐guided and attention‐based mechanisms. The framework is designed to address the progressive loss of high‐frequency information in seismic exploration and the limitations of conventional resolution enhancement methods, which struggle to balance imaging fidelity with geological interpretability. The proposed network incorporates coordinate attention blocks and the lightweight Vision Transformer, enabling more effective capture of spatial dependencies and long‐range significant features in complex geological settings. Specifically, our approach utilizes a physics‐constrained deconvolutional loss function, where the predicted reflectivity is regularized by an adaptive sparsity prior and convolved with a wavelet to synthesize seismic traces that are consistent with the observed data. In addition, a robust Charbonnier penalty ensures stable physical fitting, while anisotropic total variation regularization improves lateral continuity. Following this design, the model achieves end‐to‐end recovery of high‐resolution seismic information without requiring high‐resolution labels, thereby explicitly embedding physical constraints into the learning process. Testing results on synthetic and field datasets from two different regions demonstrate that the proposed method significantly enhances vertical resolution, reflector sharpness and lateral continuity, enabling more precise delineation of subtle stratigraphic features within the target intervals. Compared with spectral enhancement and conventional deep‐learning methods, our approach achieves higher seismic reconstruction fidelity and more interpretable reflectivity, providing an option that combines robustness with interpretability for high‐resolution imaging in complex geological conditions.
Year:
2026
Year:
2026
In hardrock or crystalline rock geological settings, due to low impedance contrast, reflected energy is usually weak. In addition, often stronger surface waves and noncoherent noise are observed including high‐frequency scattering noise, which seriously covers the useful reflection signal. Therefore, imaging of hardrock seismic data with a low signal‐to‐noise ratio (S/N) is challenging and requires tailored and cumbersome processing workflows. In this study, we propose an unsupervised learning‐based framework with frequency‐guided constraints for pre‐stack seismic data denoising. The proposed label‐free framework contains two input channels, noisy and time‐frequency‐domain data conditioned through a continuous wavelet transform (CWT) filter. The CWT filtered data provide richer feature representations guiding better the reconstruction of seismic signals. The proposed framework consists of several feature attention blocks with the soft attention mechanism to extract the spatial relationship between noisy and CWT filtered data and assign higher weights to significant features. To improve the denoising performance, we designed a hybrid loss function containing the log‐cosh function, amplitude‐weighted constraint, and frequency‐dynamic weighted constraint. We use one synthetic and two real pre‐stack seismic data sets from two mineral‐endowed regions in Sweden and Canada to test the effectiveness of the proposed network. Compared with the three benchmarks, our proposed framework shows stronger reflection signal recovery and is capable of better attenuating the complex noise. The proposed denoising workflow allows improved delineation of near‐surface structures and the mineral deposits targeted in one of the data sets.
Year:
2025
Geologic carbon storage in saline acquirers is a feasible and scalable way of reducing atmospheric carbon dioxide (CO 2 ). Since 2022, Denmark has stepped up site characterization at five suitable onshore locations for this purpose with a particular focus on four-way domal structures. A dual-element recording system incorporating landstreamer and wireless recorders was innovated and upscaled in a cost- and time-effective way for this purpose at the Rødby structure (one of the five sites). The dual-element recording allows a better imaging of the near-surface structures but also because of its broadband nature, it helps to retain higher resolution for imaging toplap structures and smaller faults in the area. The landstreamer data better image fault structures offsetting the Bunter Sandstone Formation, which is the primary reservoir, and the overlying geological seals such as Fjerritslev, Ørslev and Falster Formations. The Vedsted Formation, which a secondary seal in the region, appears unfaulted in the landstreamer data. The landstreamer data also reveal glacial valleys, near the surface, which are a source of groundwater. The two complementary datasets, from the landstreamer and nodal data, help to de-risk geological carbon storage in Denmark and is a solution we recommend to be adapted for onshore sites elsewhere in the world specially where the logistical acquisition challenges are not significant.
Year:
2025
Potential of multicomponent seismic surveys for direct targeting of metallic mineral deposits through reflection and diffraction imaging is studied in this work. The study site is at an iron-oxide deposit site in central Sweden, which has recently undergone extensive exploration work due to the increased global demand for iron ores as well as rare earth elements often associated with these deposits. The multicomponent survey consists of 150 3C recorders, connected to 10 Hz geophones with 10 m spacing. Alongside the 3C recording, a collocated broadband 1C microelectromechanical system-based survey also is conducted at every 5 m receiver spacing. The seismic source is a broadband vertical component vibrotruck, which generated three linear sweeps (2–200 Hz) at every 5 m. Given the low fold of 3C data set, we develop a processing workflow for P-P and S-S static corrections using the high-fold 1C broadband data set, which significantly improved the coherency of identified seismic features. We evaluate the final unmigrated stacks, where vertical component data indicate clear P-wavefield images of the known iron-oxide mineralization and two sets of crosscutting reflections. From the horizontal components, we are able to identify S-wavefield reflections from the mineralization only in the transverse component data. Moreover, in the transverse component data, we also identify a strong S-wavefield diffraction signal possibly indicating a magnetite-rich skarn deposit never imaged before nor in the vertical component data. We determine the value of multicomponent seismic technology for targeting large- and small-scale deposits and their potential to become a frequent part of mineral exploration workflow.
Year:
2025
Year:
2024
Year:
2023
Year:
2023
Year:
2022
Year:
2022
Year:
2020
Year:
2020
Year:
2020
Collaborators (4)
Myrto Papadopoulou
Assistant Professor
Uppsala University
Alireza MALEHMIR
Professor in applied geophysics
Uppsala University
Christopher Juhlin
Professor
Uppsala University
George Apostolopoulos
Professor of Applied Geophysics
National Technical University of Athens
Make your next move.
Prepare your application
Review drafts before you send them.