First Break - Volume 44, Issue 1, 2026

Distributed Acoustic Sensing (DAS) provides dense spatial, robust and cost-effective seismic monitoring. However, automated analysis of DAS recordings remains challenging due to their variable and inherently two-dimensional structure, densely sampled in space and time. While explicit methods often struggle with this complexity, machine learning schemes are well-suited to handle such data but are limited by the need for large, labelled datasets, which are rarely available for DAS. Furthermore, unique properties of different DAS datasets make it necessary to have a representative labelled dataset for each new application, which limits the utility of pre-trained models. To overcome these limitations, we propose a fully automated, generalisable framework that synthesizes realistic, labelled DAS datasets using cycle-consistent adversarial networks. We demonstrate the benefits of such datasets by training a seismic phase arrival picker for microseismic DAS images by comparing two models: one trained on purely synthetic data and one trained on domain-translated data, using our proposed method. The latter model provides more accurate predictions by effectively reducing model generalisation issues. The end-to-end pipeline is fully automatable and provides a scalable machine learning tool for DAS-based seismic analysis.

Equinor and its partners Ørsted and Nordsøfonden recently conducted the Kalundborg carbon capture and storage (CCS) 3D seismic survey, acquired by Geofizyka Toruń S.A. (GT), to evaluate the suitability of subsurface sites for CO2 storage in Denmark’s coastal transition zone.

This paper describes how challenges specific to transition zone (TZ) in 3D seismic surveys can be overcome in a cost-effective manner, by adapting land seismic equipment to marine conditions.

With a strong focus on compliance with environmental regulations, the project successfully delivered high-quality 3D seismic data by integrating land-nodal technology across both onshore and offshore settings. The innovative application of buoy-mounted nodes, combined with an adaptive acquisition design, enabled seamless data acquisition and processing while minimising operational complexity. This approach sets a new benchmark for transition-zone seismic surveys.

Switzerland benefits from strong social and political support for energy transition policies aimed at eliminating around 12 million tons of CO2 per year to achieve carbon neutrality by 2050. Geothermal energy is widely regarded by society and by both federal and cantonal authorities as a key contributor to this objective. While Switzerland is a global leader in the deployment of borehole heat pumps, significantly reducing fossil fuel use for heating in domestic, industrial, and agricultural sectors, efforts to exploit high-temperature rocks and fluids in the deep subsurface over the past three decades have delivered mixed results.

Outcomes from numerous deep wells drilled since 1988, including Reinach, St Moritz, Bulle, Weissbad, Thônex, Saillon, Yverdon, Basel, Triemli, St. Gallen, GEo-01, GEo-02, and more recent projects such as Lavey-les-Bains, Venzel/La Côte, and Yverdon II, show that deep geothermal potential has not yet been exploited as initially expected. Nevertheless, successes such as GEo-01 in the Geneva Basin and the repurposing of wells at Triemli, Weissbad, and Thônex demonstrate that geothermal resources exist and that giving initially disappointing wells a productive second life is a viable strategy.

This paper reviews four decades of deep geothermal projects in Switzerland and distills the main lessons learnt. Two key insights emerge: first, a viable geothermal industry requires a mature geo-energy exploration mindset supported by substantial upfront investment; second, the Swiss subsurface remains insufficiently understood, particularly regarding the controls on porosity and permeability at basin and reservoir scales. Addressing these gaps demands a co-ordinated regional exploration strategy rather than isolated, one-shot projects.

We introduce a multiparameter full-waveform inversion (FWI) approach for land seismic imaging that jointly estimates subsurface velocity and reflectivity. The method employs a scale-separation strategy to recover background velocity and high-resolution reflectivity, extending imaging beyond the depth limits of head and diving waves. To enhance robustness against noise and low-frequency limitations, we incorporate a dynamic matching objective function that aligns observed and modelled phases through multidimensional correlation, reducing sensitivity to amplitude mismatches and low signal-to-noise ratios. The reflectivity kernel further improves high-wavenumber content, delivering superior resolution and structural coherence compared to conventional imaging. Application to a Midland Basin land dataset demonstrates the method’s potential for high-resolution imaging in complex geological settings.