JCJefter Caldeira

Jefter Caldeira

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Jefter Caldeira has a total of 11 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Towards robust fracture mapping: benchmarking automatic fracture mapping in 2D outcrop imagery

    by Ayoub Fatihi, Jefter Caldeira, Tom Beucler, Samuel T. Thiele, Anindita Samsu

    Analysis of fracture geometries, orientation, distribution, and connectivity is commonly conducted to characterise the mechanical and hydraulic properties of a fractured rock mass and to interpret its deformation history. These studies increasingly leverage high-resolution drone imagery (e.g., orthomosaics or orthophotographs). However, consistently and accurately extracting fracture traces from these large datasets remains a persistent challenge. In this contribution, we present a harmonised benchmarking dataset, FraXet, for pixel-wise fracture segmentation of combined high-resolution RGB orthophotographs and digital elevation models (DEMs) (ground sampling distance ranging from approximately 0.5 to 32 mm). FraXet curates images from three publicly available datasets, totalling 8953 256 × 256 RGB + DEM patches spanning diverse lithologies and imaging conditions (ground sampling distance, illumination, etc.) to systematically assess the effectiveness of conventional image-processing approaches for fracture extraction (Canny, Sobel, Gabor, Sato, and phase congruency) and two deep-learning (DL) models (U-Net and SegFormer). Quantitative comparisons using image-quality (e.g., MSE, PSNR), segmentation (e.g., Precision, Recall, F1, IoU), and new task-specific error metrics show that the deep models substantially outperform classical filters (F1 ≈ 0.3–0.5 vs. ± 0.29) and produce smoother, more continuous fracture traces. Training on the combined dataset (M_all) improves cross-site generalisation compared to models trained on individual sub-datasets used in this study. Probability maps derived from the DL approaches enable confidence-based triage and visualisation of model uncertainty. This work establishes a unified benchmark, curated dataset, and reproducible baseline for developing robust automated fracture-detection tools.

  • Open Access

    AC1

    by Ayoub Fatihi, Jefter Caldeira, Tom Beucler, Samuel T. Thiele, Anindita Samsu

    Extracting consistent and accurate fracture traces from large volumes of high-resolution imagery remains a persistent challenge in structural analysis. We present a harmonised benchmarking dataset, FraXet, for pixel-wise fracture segmentation in high-resolution RGB orthophotos and digital elevation models (DEMs). FraXet curates images from three publicly available datasets, totalling 8953 256 × 256 RGB+DEM patches spanning diverse lithologies and imaging conditions. We use this dataset to systematically assess traditional image-processing filters (Canny, Sobel, Gabor, Sato, phase congruency) and two deep-learning (DL) models, U-Net and SegFormer, for per-pixel fracture detection. Quantitative comparison using image-quality (e.g., MSE, PSNR), segmentation (e.g., Precision, Recall, F1, IoU) and proposed similarity FracSim metrics suggest that the deep models substantially outperform classical filters (F1 ≈ 03 −0.5 vs ≤ 0.29), giving smoother, more continuous fracture traces with reduced noise. Training on the combined dataset (M_all) improves cross-site generalisation relative to models trained on the individual sub-datasets. Challenges remain in handling annotation misalignments, illumination artifacts, and thin traces. More importantly, probability maps derived from the DL approaches enable confidence-based triage and visualisation of model uncertainty. This work thus establishes a unified benchmark, curated dataset, and reproducible baseline to support further development of robust automated tools for fracture detection.

  • Open Access

    Comment on egusphere-2026-1097

    by Ayoub Fatihi, Jefter Caldeira, Tom Beucler, Samuel T. Thiele, Anindita Samsu

    Extracting consistent and accurate fracture traces from large volumes of high-resolution imagery remains a persistent challenge in structural analysis. We present a harmonised benchmarking dataset, FraXet, for pixel-wise fracture segmentation in high-resolution RGB orthophotos and digital elevation models (DEMs). FraXet curates images from three publicly available datasets, totalling 8953 256 × 256 RGB+DEM patches spanning diverse lithologies and imaging conditions. We use this dataset to systematically assess traditional image-processing filters (Canny, Sobel, Gabor, Sato, phase congruency) and two deep-learning (DL) models, U-Net and SegFormer, for per-pixel fracture detection. Quantitative comparison using image-quality (e.g., MSE, PSNR), segmentation (e.g., Precision, Recall, F1, IoU) and proposed similarity FracSim metrics suggest that the deep models substantially outperform classical filters (F1 ≈ 03 −0.5 vs ≤ 0.29), giving smoother, more continuous fracture traces with reduced noise. Training on the combined dataset (M_all) improves cross-site generalisation relative to models trained on the individual sub-datasets. Challenges remain in handling annotation misalignments, illumination artifacts, and thin traces. More importantly, probability maps derived from the DL approaches enable confidence-based triage and visualisation of model uncertainty. This work thus establishes a unified benchmark, curated dataset, and reproducible baseline to support further development of robust automated tools for fracture detection.

  • Open Access

    Strike-slip fault zone architecture in folded Upper Malm limestones at Gorges de l’Orbe, Central Internal Jura, Switzerland

    by Jefter Caldeira, Anindita Samsu, Ana Tanaka, Loïc Bazalgette

    Strike-slip fault zones often play a key role in controlling brittle deformation patterns in fold-and-thrust belts. Yet, the distribution of associated fractures and their interactions with fold-related structures remain insufficiently understood across scales and as part of the deformation history. In the Jura Mountains, previous work in the Central Internal Jura, notably at Creux-du-Van, proposed a multi-scale hierarchy of strike-slip faults and associated fracture networks based on sub-seismic structures spanning meter- to kilometer-scale. This study extends the multi-scale structural analysis to a regionally mappable ~12 km long N–S striking sinistral fault zone, the Suchet Fault. The fault is exposed along the E–W-oriented Gorges de l’Orbe, where incision into Upper Malm limestones enables largely continuous outcrop-scale access to fault-related damage zones.To address the multi-scale character of this system, we combine field-based structural mapping at 1:5,000 scale and three-dimensional structural interpretation of publicly available aerial LiDAR data (SwissALTI3D) with systematic fracture data acquisition along four scanlines totaling 157 m. Scanlines are positioned at varying distances from the fault core, defined by the presence of fault breccia and gouge lenses, and across western and eastern structural compartments, spanning fault-proximal to fault-distal domains. This configuration enables comparison between fracture populations associated with fault-zone architecture and those interpreted as fold-related or background fracturing.The Suchet Fault separates two contrasting structural domains. The western compartment is characterized by NW–SE striking fold trains, whereas the eastern compartment exhibits a comparatively flatter structural geometry. Within the vicinity of the fault trace, bedding orientations rotate progressively toward the N–S fault trend, with gentle eastward dips (~15°). Locally, near the fault core, bedding dips steepen and may reach up to 70°, indicating increased strain localization within the damage zone. LiDAR-based structural interpretation identifies three dominant fracture populations, with the NW–SE striking set displaying comparatively longer fractures than the N–S and NE–SW sets.Fault-slip indicators show dominant subvertical conjugate strike-slip pairs at outcrop scale, comprising sinistral N–S to NNE-SSW striking faults and dextral NW–SE striking faults. Preliminary paleostress inversion analysis indicates a strike-slip regime characterized by a subhorizontal NW-directed maximum principal stress and a subvertical intermediate stress, consistent with results from other sectors of the Central Internal Jura. Fracture density (P10) increases toward the fault core, with values close to the brecciated core notably higher than those measured beyond 100 m.This study emphasizes the need for robust fracture set definition and sequencing as a basis for structural analysis, including paleostress orientations and spatial variations in fracture intensity and anisotropy. This study evaluates whether structural patterns and paleostress behaviors identified at Creux-du-Van are comparable to those observed in the Gorges de l’Orbe area, at the scale of larger strike-slip fault zones. It also considers the potential regional implications of these structural features for fracture-controlled fluid flow in the Jura Mountains and potentially downstream, in the geothermal reservoirs beneath the Molasse Basin.

  • Open Access

    Towards robust fracture mapping: benchmarking automatic fracture mapping in 2D outcrop imagery

    by Ayoub Fatihi, Jefter Caldeira, Tom Beucler, Samuel T. Thiele, Anindita Samsu

    Extracting consistent and accurate fracture traces from large volumes of high-resolution imagery remains a persistent challenge in structural analysis. We present a harmonised benchmarking dataset, FraXet, for pixel-wise fracture segmentation in high-resolution RGB orthophotos and digital elevation models (DEMs). FraXet curates images from three publicly available datasets, totalling 8953 256 × 256 RGB+DEM patches spanning diverse lithologies and imaging conditions. We use this dataset to systematically assess traditional image-processing filters (Canny, Sobel, Gabor, Sato, phase congruency) and two deep-learning (DL) models, U-Net and SegFormer, for per-pixel fracture detection. Quantitative comparison using image-quality (e.g., MSE, PSNR), segmentation (e.g., Precision, Recall, F1, IoU) and proposed similarity FracSim metrics suggest that the deep models substantially outperform classical filters (F1 ≈ 03 −0.5 vs ≤ 0.29), giving smoother, more continuous fracture traces with reduced noise. Training on the combined dataset (M_all) improves cross-site generalisation relative to models trained on the individual sub-datasets. Challenges remain in handling annotation misalignments, illumination artifacts, and thin traces. More importantly, probability maps derived from the DL approaches enable confidence-based triage and visualisation of model uncertainty. This work thus establishes a unified benchmark, curated dataset, and reproducible baseline to support further development of robust automated tools for fracture detection.

  • Open Access

    Seamless Fracture Characterization with Virtual Outcrop Models: Fracture Geometry and Vertical Persistence at Creux-du-Van, Swiss Jura Mountains  

    by Jefter Caldeira, Anindita Samsu

    Deep geothermal energy is emerging as a key component of Switzerland’s Energy Strategy 2050. In the Canton of Vaud, projects target natural hydrothermal systems at depths of 800 to 1,000 meters, with temperatures ranging from 35 to 100°C, in Upper Malm limestone reservoirs. These limestones, with low matrix porosity and permeability, rely on well-connected fracture networks and karstic features to enhance fluid flow. However, subsurface data from seismic surveys and well data do not fully cover scale intervals that are relevant for reservoir characterization and modeling. To address this limitation, we use a virtual outcrop model (VOM) to characterize fractures in 3D, explore its potential to bridge the length-scale gap, and compare fracture patterns and kinematics between scales of observation. At Creux-du-Van, a unique continuous, 3D exposure of fractured and gently folded Upper Malm limestones in the Central Internal Jura Fold and Thrust Belt provides an exceptional opportunity for fracture characterization at the decimeter to hundreds of meters scale, allowing comparisons with previous structural interpretations at the regional, 1:500,000 to 1:25,000 scale. Approximately 700 fractures were interpreted from high-resolution VOMs, i.e., point clouds (~110 points/m²), of Creux-du-Van, derived from terrestrial LiDAR scanning. These fractures were classified based on vertical persistence, which is a relative measure for the extent to which they propagate across mechano-stratigraphic boundaries. Geometric parameters such as orientation, dimensions (length, width, and aspect ratio), and spacing were also quantified. Field-based structural analysis complements the digital dataset by providing kinematic and chronological interpretations of brittle structures linked to the tectonic evolution of the fold and thrust belt.The lengths of regional strike-slip fault traces span four orders of magnitude, ranging from tens of meters to tens of thousands of meters, with a median of 177 meters. Their orientation, kinematic, and length relationships align with the multi-scale Riedel shear model (Ruhland, 1973) forming the regional structural framework. Fractures in the VOM span scales from 0.1 to hundreds of meters, with a median length of 5 meters. Regional faults and VOM-derived fractures show an overlap in length distributions and consistency in fracture orientations and kinematics. Both align with the NW-SE compression inferred from field-based kinematic data and regional restorations associated with the Jura shortening event, demonstrating seamless characterization of brittle features across scales. This study seeks to further investigate the role of mechanical boundaries (e.g., stratigraphic boundaries, regional structures) in controlling reservoir compartmentalization. It also showcases the potential of outcrop analogues such as Creux-du-Van to support 3D characterization and analyses of fracture properties such as length distributions, orientation, and vertical persistence, ultimately contributing to the advancement of structural modelling of subsurface reservoirs for sustainable energy solutions.

  • Open Access

    Fracture networks in the Muschelkalk aquifer of the external northen foreland of the Central Alps (CH/DE); implications for permeability, CO2 storage and geothermal potential

    by Alannah C. Brett, Jefter Caldeira, Anindita Samsu, Larryn W. Diamond, Herfried Madritsch

    Switzerland aims to reach net-zero CO2 emissions by 2050 (BAFU, 2021). Implementation of geological storage of CO2 and geothermal heat mining to help achieve this aim requires reservoir properties to be assessed. The structural and permeability architectures of the target reservoirs are essential input for numerical models used to assess storage and production potential, minimize fluid injection and extraction uncertainties, and reduce exploration risks. The so-called Muschelkalk aquifer (Triassic) including the Schinznach Formation is regarded as one of the key potential aquifers for gas storage and hydrothermal geothermal systems in Switzerland (Chevalier et al., 2010). While the matrix permeability of the Schinznach Formation is relatively well known (Diamond et al., 2019), magnitude and distribution of its fracture permeability and structural controls on these fractures are poorly understood. This study aims to assess the style and intensity of natural fracture networks in the Muschelkalk aquifer at sub-seismic scale and explain their regional variability. Outcrop analogues in the Wutach Gorge of southern Germany are used to improve understanding of lateral and vertical variability of fracture networks, including how they are influenced by regional structures. The Wutach Gorge is within the Tabular Jura and provides cliff exposures along 5–12 km-long E–W and N–S transects, aligning with and crossing fault strands of the Freiburg–Bonndorf–Bodensee Fault zone.Insights from field observations contribute to ongoing work that supports the proposed pilot CO2 injection test into the Schinznach Formation via an existing exploration borehole at Trüllikon in northern canton Zurich. A feasibility study (Diamond et al., 2023) assessed the reservoir properties at Trüllikon by building discrete fracture network models and computing their permeabilities from a combination of rock-matrix properties, vertical drill hole fracture logs, a horizontal fracture log from another nearby drill hole, and results from hydraulic tests. This multidisciplinary approach should provide a more robust basis for exploration for CO2 storage sites and geothermal energy. REFERENCES Chevalier, G., Diamond, L. W., & Leu, W. (2010). Potential for deep geological sequestration of CO2 in Switzerland: a first appraisal. Swiss Journal of Geosciences, 103, 427-455.Diamond, L. W., Alt-Epping, P., Brett, A.C., Aschwanden, L. and Wanner, C. (2023) Geochemical–hydrogeological study of a proposed CO2 injection pilot at Trüllikon, Switzerland. Report 2023-7 submitted to the Swiss Geological Survey (swisstopo). Rock Water Interaction, University of Bern, 87 pp. https://doi.org/10.5281/zenodo.10938102Diamond L.W., Aschwanden, L., Adams, A., and Egli, D. (2019) Revised potential of the Upper Muschelkalk Formation (Central Swiss Plateau) for CO2 storage and geothermal electricity. Slides of an oral presentation at the SCCER-SoE Annual Conference at EPFL-Lausanne, 4th Sept. 2019. 13 pp. http://static.seismo.ethz.ch/sccer-soe/Annual_Conference_2019/AC19_S3a_08_Diamond.pdfBAFU (2021) Switzerland Long-Term Climate Strategy. 4 pp. https://www.bafu.admin.ch/dam/bafu/en/dokumente/klima/fachinfo-daten/langfristige-klimastrategie-der-schweiz.pdf.download.pdf/Switzerland's%20Long-Term%20Climate%20Strategy.pdf