Abstract. We present curlew, an open-source python package for structural geological modelling using neural fields. This modelling framework incorporates various local constraints (value, gradient, tangent and (in)equalities) and tailored global loss functions to ensure data-consistent and geologically realistic predictions. Random Fourier Feature (RFF) encodings are used to improve model convergence and facilitate stochastic uncertainty quantification, while simultaneously improving the model's ability to learn naturally periodic features such as folds. These advances are integrated into a software framework that allows incremental construction of complex geological models through temporally-linked neural fields, each representing a specific deposition, intrusion or faulting event. Significantly, this framework allows semi-supervised learning to integrate diverse unlabelled datasets (e.g., geochemistry, petrophysics), reducing interpretation bias and potentially improving model robustness. We describe and demonstrate these various capabilities using synthetic examples and real data from a faulted stratigraphic digital outcrop model from Newcastle, Australia.

Following the demonstrated success of the CarbFix project in Iceland, subsurface carbon storage within basaltic formations presents a compelling strategy for permanent CO2 sequestration. Basaltic reservoirs facilitate rapid in situ mineral carbonation on timescales of years, much faster than the centuries typically needed to mineralize carbon in conventional sedimentary reservoirs. The Deccan Traps, one of the largest continental flood basalt provinces worldwide, has therefore been recognized as a prospective site for large-scale CO2 storage. However, substantial internal heterogeneity within basalt sequences, coupled with the inherent challenges of imaging deeper depths with conventional seismic techniques, creates considerable uncertainty in reservoir characterization and storage modeling. Here, we develop a digital 3D outcrop analogue characterizing Deccan basalt flow architecture across a 4×2 km2 area in Kumbharli Ghat (Western Ghats, Maharashtra, India). The study area was selected for its well-exposed flows, accessibility, high-quality road-cut outcrops, and proximity to Borehole Geophysical Research Laboratory (BGRL) wells, enabling high-resolution quantification for analogue studies and future correlation of BGRL’s one-dimensional subsurface core data with the outcrop model. This is the first drone-based photogrammetric (SfM-MVS; 1660 images) model of Deccan Traps to quantify vertical and lateral thickness variability and facies heterogeneity of flow in the study area. Flow boundaries have been mapped, isopach maps have been created, and probabilistic thicknesses have been calculated for seven distinct lava flows. Each flow displays a characteristic sequence, from a vesicular base, massive/columnar jointed core, vesicular roof zone, and capping breccia and/or red bole. Flow thickness varies by 16-60% both vertically and laterally, emphasizing the heterogeneity likely to be encountered in the subsurface. These thickness statistics exhibit both unimodal and bimodal frequency patterns, which have been linked to laterally extensive sheet flow and compound or ponded lobes, respectively. Within flows, fluid-free zone (FFZ; massive/jointed) and fluid-trapping zone (FTZ; vesicular, brecciated, or red bole) intervals constitute 54 ± 7% and 46 ± 7% of flow thickness, respectively, suggesting substantial compartmentalization of potential CCS reservoirs. Micro-CT reveals both isolated vesicles and locally connected pores, while petrographic analysis highlights locally abundant zeolites and secondary clays, underscoring reactive and transport heterogeneity. Our results thus suggest that macro- and micro-scale heterogeneity will exert a first-order control on injectivity, storage capacity, connectivity, and formation pressure.

Abstract. 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.

Global mining operations cause significant vegetation disturbance, yet their cumulative footprint remains poorly quantified. Remote monitoring of these changes over time is needed for environmental accounting and to estimate aggregated impacts across the thousands of mines that underpin global metal supply chains. In this contribution, we integrate vegetation index time series with a domain-adapted breakpoint detection approach, Seasonal Harmonic Anomaly Break Analysis (SHABA), to detect multiple abrupt and subtle vegetation changes associated with mining activities. We demonstrate the potential of this methodology with six geographically distributed mine sites: Cardinal River, Roșia Poieni, Trident, Carajás, Vametco, and Worsley. The resulting breakpoints accurately quantify cumulative changes associated with mine expansion and rehabilitation efforts and indicate where secondary changes in surrounding areas might occur. The mines’ footprint trajectories are varied and complex. In some of the cases illustrated in this study, primary footprints are dominated by stepwise vegetation clearing (2.11–5.77% yr 1), but also show early regrowth (1.58–2.68%yr 1). consistent with natural succession or rehabilitation. Secondary changes outside the direct mine footprint vary widely, but are temporally correlated with mine expansion. Based on these results, we hypothesise that the SHABA workflow could provide a robust approach to automated, global-scale quantification of changes in mine footprint over time and to better dialogue among stakeholders.

The Eastern Erzgebirge (Germany) and Krusne hory (Czech Republic / Czechia) region hosts prolific Li-(Sn-W) deposits, all linked to late-stage magma evolution and magmatic-hydrothermal alteration within a Caldera collapse system. Whereas the geochemical controls are relatively well understood, tectonic controls on magma emplacement are not. Here, we aim to explain the tectonic controls on trans-crustal caldera-forming magmatic systems, and link these to more local controls on fertile magmatism. This is achieved by compiling and reviewing available geological, geochronological, geophysical, and structural data, and integrating them to derive a framework for late-to post-Variscan tectonics and magmatism. Specifically, we link the main faults in the vicinity of the Altenberg-Teplice and Tharandt calderas with the western middle Pennsylvanian (similar to 314-312 Ma) Bohemian basin system, to propose a major transtensional linkage structure between the Elbe Shear Zone and Pfahl or Danube Shear Zones. We propose that these transtensional pull-apart basins and dextral strike-slip fault systems do not only localize crustal-scale magmatic systems and associated calderas, but also exert a more local control on intra-caldera intrusive stocks that are host to greisen-type Li-(Sn-W) ore deposits in the Eastern Erzgebirge / Krusne hory region.

Hyperspectral analysis of carbonate rocks provides a novel method to recognize not only calcite/dolomite alterations, but also to differentiate dolomite fabrics. Coupled with digital outcrop models, hyperspectral data provide an integrated representation of the geometric and mineralogical characteristics of exposed dolomite geobodies at sub-seismic resolution and over large (seismic-scale) extent. This facilitates the continuous, unbiased, and data-driven assessment of the spatial distribution of dolomites, dolomite types and properties. Here we integrate hyperspectral attributes, geochemical data, fracture analysis, tectonic, and thermal histories to constrain the process and timing of dolomitization and the compositional and textural heterogeneity at cm-scale. Our results suggest that the km-scale strata-bound dolomitized layers of the Arab-D member formed in an overall regressive system tract. near the surface (T ~ 30 °C) by refluxing of slightly evaporated seawater (-1.0 to 0‰ SMOW). With undolomitized shallow transgressive mudstone/wackestone layers forming baffles restricting downward fluid flow, the dolomitization process apparently was repetitive and linked to high frequency cycles with preferential dolomitization of cycle-top grainstone facies. Multiple reflux events during high-frequency cycle deposition led to the alternating dolomite/calcite layering. Thus, a classical one-time-dolomitize-all end-of-sequence reflux system is not indicated. The early-formed metastable dolomites were then recrystallized during burial and finally overprinted by a hot (80 °C or more) deep-seated fluid with a composition of up to 6.5‰ SMOW. This fluid was channeled by a NW-SE oriented regional fracture trend, which originated from a Late Cretaceous plate-wide structural event related to the Alpine I tectonic deformation. As the dolomite fabric was altered, porosity and permeability became enhanced. The temperatures derived from clumped isotope analysis, thermal history, and the Alpine I related fracture conduits consistently suggest a latest Cretaceous origin for the final burial dolomite maturation and textural overprinting.

Recent research has highlighted significant correlations between hyperspectral data and the petrophysical properties of geological formations. Petrophysics acts as the link between geology and geophysics, and is crucial for constraining geophysical inversions, regional characterisation and mineral exploration. In this study, we employ various machine learning methods to predict P-Wave velocities using hyperspectral borehole data, with a focus on cross-validation between different boreholes in the same region. Our dataset includes 4022 paired observations of P-wave velocities, obtained from downhole sonic logging in one borehole, and corresponding hyperspectral data spanning visible-near (380–970 nm), shortwave (970–2500 nm), midwave (2700–5300 nm), and long-wave (7700–12300 nm) spectra, averaged over a 10 cm x 5 cm area. We utilised principal component analysis (PCA) for dimensionality reduction. The initial PCA stage extracted 10 principal components from each sensor type, which were then integrated. A subsequent PCA stage was conducted to reduce inter-sensor correlation, yielding 10 composite features that represent the variability across the complete VNIR-LWIR spectrum. To validate our model, we conducted tests using 1160 pairs of analogous measurements from a different borehole within the same geological region. The model demonstrated impressive predictive capabilities, particularly with Support Vector Regression (SVR) and Artificial Neural Networks (ANN). The test set yielded R2 scores of 0.758 for SVR and 0.811 for ANN, indicating strong predictive accuracy. Building upon this success, our future work will expand the scope of prediction to include various other petrophysical properties critical to geophysical characterization and mineral exploration, such as S-Wave velocity, magnetic susceptibility, and rock density, properties which are critical for geophysical characterization and mineral exploration.

Editor’s note: The aim of the Geology and Mining series is to introduce early career professionals and students to various aspects of mineral exploration, development, and mining in order to share the experiences and insight of each author on the myriad of topics involved with the mineral industry and the ways in which geoscientists contribute to each. Abstract We outline the potential to adopt geometallurgical concepts during early mineral exploration, particularly during scoping studies, rather than later during feasibility studies or exploitation when costs are higher. The approach is rooted in the increasing capabilities of drill core scanning technologies. Continuous drill core scanning data can now be generated efficiently and at reasonable cost. Validating and calibrating these data with high-resolution quantitative imaging of a suite of localized test samples, e.g., from scanning electron microscopy-based image analysis, allow the mineralogy and microfabric of drill core to be quantified. This quantitative information can then be used for more accurate geologic domaining of a potential orebody. The resulting geologic domain model then provides the basis for sample selection and blending that is essential for representative beneficiation test work. These test results can then be combined with emerging particle-based process modeling techniques that are predictive and can be designed to help understand and tackle metallurgical challenges in unlocking a mineral resource. This will assist in defining geometallurgical domains, using both geologic and technological constraints. However, this ambition is currently limited by several knowledge gaps. Arguably the most crucial issue concerns the forecasting of comminution responses, including particle sizes and compositions, based on the measured mineralogy and microfabric of the ores. Other challenges relate to the resolution and speed of available core scanning technologies and the incorporation of physical constraints into particle-based beneficiation models. Once these issues have been resolved, we expect substantial improvements in the efficiency and predictive power of geometallurgy, which should enable its application during earlier stages of exploration, with greater reliability at each decision stage during a development.

Hyperspectral data provide rich information on both the mineralogical and fine-scale textural properties of rocks, which also control their petrophysical characteristics. We propose that some physical rock properties can be predicted directly from hyperspectral data, improving petrophysical characterisation and reducing the need for often laborious measurements. In this contribution we explore correlations between hyperspectral and petrophysical data using a deep convolutional neural network. Our model learns relevant features from high-dimensioned hyperspectral data to predict slowness, density, and gamma-ray values using training and testing data from Spremberg, Germany. Our results show that, with careful preprocessing and thorough data cleaning, differences in resolution can be overcome to learn the relationship between hyperspectral data and petrophysics. Using a test dataset from a spatially independent borehole, we generated a pixel-resolution (≈1 mm2) model of the petrophysical properties and resampled it to match the measured logs. This test indicated substantial accuracy, with R2 scores and root-mean-squared errors (RMSEs) of 0.7 and 16.55 µs m−1, 0.86 and 0.06 g cm−3, and 0.90 and 15.29 API for the slowness, density, and gamma-ray predictions respectively. We also analysed the Shapley values of our model to gain deeper insights into its predictions. These findings lay the groundwork for building deep learning models that predict physical and mechanical rock properties from hyperspectral data. Such models could provide the high-resolution but large-extent data needed to bridge the different scales of mechanical and petrophysical characterisation.

Europe holds significant potential for crucial metals essential for renewable energy and digital advancements. However, there is a need for a deeper understanding of the European subsurface, coupled with a requirement to reduce the environmental impact of exploration activities. We employ hyperspectral scanning of legacy drill cores to develop innovative indicators of mineralization in the Central European Kupferschiefer district, home to Europe's largest copper and silver resources. Hyperspectral imaging, capturing spectral reflectance and emission across numerous spectral bands, allows for non-invasive, high-resolution mineral mapping of drill cores. Initial findings showcased the technique's ability to identify critical redox boundaries and alteration minerals, aiding ore deposit characterization. In the framework of two research projects, we scanned 2400 meters of drill core from 87 boreholes across the Spremberg–Graustein Kupferschiefer deposit in Germany. For data acquisition, we deployed a drill-core scanner with a full suite of hyperspectral sensors covering the visible and near-infrared (400 to 970 nm), shortwave (970 to 2500 nm), mid-wave (2700 to 5300 nm), and longwave infrared (7700 to 12300 nm) ranges. The collected data were processed through a novel, open-source software pipeline, which enables i) real-time correction, processing, and analysis, ii) efficient data management and storage, and iii) comprehensive visualization and integrative interpretation of the hyperspectral drill core data. We upscaled mineral abundances across all of the scanned drill cores using a supervised learning model trained on quantitative mineralogical data from select samples. Initial analyses, particularly the visual alignment of hyperspectral derivatives at the base of the Kupferschiefer marker horizon, indicate geographical patterns in dolomite content and correlations between carbonate, clay, and mica compositions and copper grade. The hyperspectral data will eventually be integrated with geological, geophysical, and geochemical constraints to create accurate 3D subsurface and 4D mineral system models, aimed at enhancing our understanding of geological processes and resource management strategies in similar geological settings worldwide. Acknowledgements: This research has received funding from the European Union’s Horizon Europe research and innovation programme under grant agreement nº 101058483 (VECTOR), and from the Geological Survey of Saxony (Sächsisches Landesamt für Umwelt, Landwirtschaft und Geologie, LfULG) under agreement nº 4-0912014LFULG01-88.

Heterogeneous structures and diverse volcanic, hydrothermal, and geomorphological processes hinder characterisation of the mechanical properties of volcanic rock masses. Laboratory experiments can provide accurate rock property measurements, but are limited by sample scale and labor-intensive procedures. In this contribution, we expand on previous research linking the hyperspectral fingerprints of rocks to their physical and mechanical properties. We acquired a unique dataset characterising the visible-near (VNIR), shortwave (SWIR), midwave (MWIR), and longwave (LWIR) infrared reflectance of samples from eight basaltic to andesitic volcanoes. Several machine learning models were then trained to predict density, porosity, uniaxial compressive strength (UCS), and Young's modulus (E) from these spectral data. Significantly, nonlinear techniques such as multilayer perceptron (MLP) models were able to explain up to 80 % of the variance in density and porosity, and 65 %-70 % of the variance in UCS and E. Shapley value analysis, a tool from explainable AI, highlights the dominant contribution of VNIR-SWIR absorptions that can be attributed to hydrothermal alteration, and MWIR-LWIR features sensitive to volcanic glass content, fabric, and/or surface roughness. These results demonstrate that hyperspectral imaging can serve as a robust proxy for rock physical and mechanical properties, potentially offering an efficient, scalable method for characterising large areas of exposed volcanic rock. The integration of these data with geomechanical models could enhance hazard assessment, infrastructure development, and resource utilisation in volcanic regions.

Abstract. Tensor fields, as spatial derivatives of scalar or vector potentials, offer powerful insight into subsurface structures in geophysics. However, accurately interpolating these measurements–such as those from full-tensor potential field gradiometry–remains difficult, especially when data are sparse or irregularly sampled. We present a physics-informed spatial neural network that treats tensors according to their nature as derivatives of an underlying scalar field, enabling consistent, high-fidelity interpolation across the entire domain. By leveraging the differentiable nature of neural networks, our method not only honours the physical constraints inherent to potential fields but also reconstructs the scalar and vector fields that generate the observed tensors. We demonstrate the approach on synthetic gravity gradiometry data and real full-tensor magnetic data from Geyer, Germany. Results show significant improvements in interpolation accuracy, structural continuity, and uncertainty quantification compared to conventional methods.

The new generation of satellite hyperspectral (HS) sensors provides remarkable potential for regional-scale mineralogical mapping. However, as with any satellite sensor, mapping results are dependent on a typically complex correction procedure needed to remove atmospheric, topographic and geometric distortions before accurate reflectance spectra can be retrieved. These are typically applied by the satellite operators but use different approaches that can yield different results. In this study, we conduct a comparative analysis of PRISMA, EnMAP, and EMIT hyperspectral satellite data, alongside airborne data acquired by the HyMap sensor, to investigate the consistency between these datasets and their suitability for geological mapping. Two sites in Namibia were selected for this comparison, the Marinkas-Quellen and Epembe carbonatite complexes, based on their geological significance, relatively good exposure, arid climate and data availability. We conducted qualitative and three different quantitative comparisons of the hyperspectral data from these sites. These included correlative comparisons of (1) the reflectance values across the visible-near infrared (VNIR) to shortwave infrared (SWIR) spectral ranges, (2) established spectral indices sensitive to minerals we expect in each of the scenes, and (3) spectral abundances estimated using linear unmixing. The results highlighted a notable shift in inter-sensor consistency between the VNIR and SWIR spectral ranges, with the VNIR range being more similar between the compared sensors than the SWIR. Our qualitative comparisons suggest that the SWIR spectra from the EnMAP and EMIT sensors are the most interpretable (show the most distinct absorption features) but that latent features (i.e., endmember abundances) from the HyMap and PRISMA sensors are consistent with geological variations. We conclude that our results reinforce the need for accurate radiometric and topographic corrections, especially for the SWIR range most commonly used for geological mapping.

We combine 2 m resolution airborne (HySpex) and 30 m resolution satellite (EnMAP) hyperspectral data to address the challenge of mixed pixels in satellite imagery. Endmembers are manually selected from HySpex data, and Non-negative Least Squares (NNLS) spectral unmixing is applied to generate high-resolution spectral abundance maps. These maps are then resampled to match EnMAP's spatial resolution and used to predict an endmember library from the EnMAP scene. This predicted library is then used for unmixing the EnMAP data over a broader area. When compared to spectral abundance maps generated from direct endmember selection from EnMAP alone, the unmixing results using the predicted library closely align with the high-resolution output, despite some land cover changes over time. In contrast, the spectral abundance maps from low-resolution endmembers lack detail. We discuss the implications of our approach for improved spatial and temporal mapping.

Hyperspectral imaging is gaining widespread use in the resource sector, with applications in mineral exploration, geometallurgy and mine mapping. However, the sheer size of many hyperspectral datasets (>1 Tb) and associated correction, visualisation and analysis challenges can limit the integration of this technique into time-critical exploration and mining workflows. In this contribution, we propose and demonstrate a novel open-source workflow for rapidly processing hyperspectral data acquired on exploration drillcores. The resulting products are adaptable to the varied needs of geologists, geophysicists and geological engineers, facilitating better integration of hyperspectral data during decision making. These tools are applied to process hyperspectral data of 6.4 km of exploration drill cores from Stonepark (Ireland), Collinstown (Ireland) and Spremberg (Germany). The results are presented via an open-source web-viewing platform that we have developed to facilitate easy on and off-site access to hyperspectral data and its derivatives. We suggest that maximum value can be extracted from hyperspectral data if it is acquired shortly after drilling and processed on-site in real time, so that results can be quickly validated and used to inform time-critical decisions on sample selection, geological interpretation (logging) and drillhole continuation or termination. This timeliness and accessibility is key to ensure rapid data availability for decision makers during mineral exploration and exploitation. Finally, we discuss several remaining challenges that limit the real-time integration of hyperspectral drill core scanning data, and explore some opportunities that may arise as these rich datasets become more widely collected.

We argue that traditional 2D hyperspectral imaging is not adapted to many modern challenges. With the rise of high spatial resolution, hyperspectral sensors mounted on different platforms (e.g. drones, terrestrial, satellites) and innovative applications (e.g. urban mapping, mining monitoring), projections, occlusions, perspective effects and data processing limit the use of 2D hyperspectral imaging. We propose that 3D hyperclouds, in which Lidar or photogrammetric point clouds are augmented with hyperspectral attributes, can address numerous of these challenges. We demonstrate the benefits of hyperclouds and dedicated machine learning architectures with several realistic examples.

Diagenetic alteration commonly overprints depositional fabrics in carbonate-dominated sediments and impact reservoir quality. Dolomitization is a prevalent diagenetic process observed in subsurface reservoirs that profoundly alters the depositional precursor's pore network, thereby influencing subsurface storage capacity and fluid flow behavior. Typical workflows to characterize the dolomitized sequences, textures, degree and extent of dolomitization rely on mapping, spatial sampling, and time-consuming geochemical, petrographic and petrophysical analysis. In this study, we propose a hyperspectral data-driven workflow for identifying dolomitized horizons and extracting sample sets optimized to characterize textural and chemical variations. Hyperspectral imaging (HSI) data was acquired with 1.5 mm spatial sampling along a 50 m long core drilled behind an outcrop of the Late Jurassic Jubaila-Arab sequence in Wadi Daqlah, Saudi Arabia. Spectral features in the visible (VNIR), shortwave (SWIR), mid-wave (MWIR), and long-wave (LWIR) infrared regions were then used to classify carbonate mineralogy, allowing for the rapid identification of dolomitized zones, and k-means clustering applied exclusively to the dolomitized areas used to identify intra-dolomite variations and suggest representative sample locations. Petrographic and geochemical analyses were carried out on these samples, revealing that clusters identified with the hyperspectral data represent four distinct diagenetic fabrics. These results demonstrate the value of HSI for objective and data-driven sampling, reducing the number of samples required for petrographic, geochemical and geophysical analysis and hence time and costs required to spatially characterize diagenetic alteration.

The increasing use of deep learning techniques has reduced interpretation time and, ideally, reduced interpreter bias by automatically deriving geological maps from digital outcrop models. However, accurate validation of these automated mapping approaches is a significant challenge due to the subjective nature of geological mapping and the difficulty in collecting quantitative validation data. Additionally, many state-of-the-art deep learning methods are limited to 2D image data, which is insufficient for 3D digital outcrops, such as hyperclouds. To address these challenges, we present Tinto, a multi-sensor benchmark digital outcrop dataset designed to facilitate the development and validation of deep learning approaches for geological mapping, especially for non-structured 3D data like point clouds. Tinto comprises two complementary sets: 1) a real digital outcrop model from Corta Atalaya (Spain), with spectral attributes and ground-truth data, and 2) a synthetic twin that uses latent features in the original datasets to reconstruct realistic spectral data (including sensor noise and processing artifacts) from the ground-truth. The point cloud is dense and contains 3,242,964 labeled points. We used these datasets to explore the abilities of different deep learning approaches for automated geological mapping. By making Tinto publicly available, we hope to foster the development and adaptation of new deep learning tools for 3D applications in Earth sciences. The dataset can be accessed through this link: https://doi.org/10.14278/rodare.2256.

The Puga valley, in Ladakh, contains one of India's most prospective geothermal systems. Substantial geophysical and geochemical research has been conducted to characterise this system, though uncertainties regarding the subsurface reservoir's geometry and permeability structure remain a barrier to its development. In this contribution, we aim to fill some of these knowledge gaps by integrating new geological data and structural analyses with previously published geochemical and geophysical interpretations, and derive an integrated conceptual model of the geothermal system. Using digital outcrop techniques and field mapping, we identify and characterise several important structures (faults and foliations) that facilitate fluid flow in the otherwise impermeable Tso Morari gneiss. Petrological and field evidence for outcropping hydrothermally altered lithologies, may have formed in a geothermal system analogous to the active one, are also presented. Based on these observations and a simplified finite-element model, we suggest that tectonic and topographic stresses likely control reservoir architecture and connectivity. Lastly, we caution that geomorphological evidence for neotectonic movement on faults at Puga indicate the need for seismic hazard assessment prior to exploitation of the geothermal system, and identify potential parallels between Puga and the Yangbajing geothermal field in China.