Mineralogical analysis is critical for understanding the origins and properties of various rock types. Recent advancements in artificial intelligence technologies, particularly supervised deep learning, have transformed qualitative and quantitative mineral analysis. However, current deep-learning approaches require high-quality annotated datasets to acquire and identify different mineralogical characteristics and properties. This is exacerbated by the time-consuming and error-prone labeling processes for the datasets. With self-supervised architectures matching supervised approaches in many computer vision tasks, it is timely to investigate the potential of Self-Supervised Learning (SSL) models in the geosciences. As a result, we use a self-supervised semantic segmentation model to identify and characterize minerals in thin sections, with the model attempting to obtain categories of interest from images without the need for human intervention in annotating the minerals. In this study, we adopted a Self-Supervised Transformer architecture and proposed SelfMin to automatically segment out pyrite minerals from other background gangue minerals in the thin section. Our proposed method achieved 80% in the mean Intersection over Union (mIoU) metric, indicating the model's ability to accurately segment minerals that were not labeled during the annotation process. This work describes the first use of self-supervised deep learning in mineralogical analysis. Further application of this proposed method would allow a robust and efficient advanced qualitative and quantitative mineralogical analysis. It also demonstrates how this technique can be implemented to avoid the need for a large volume of high-quality labeled datasets in other image-based deep learning geosciences analyses.

How to cite: Koeshidayatullah, A. and Ferreira, I.: SelfMin: Self-Supervised Deep Learning for Advanced Mineralogical Analysis, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-2189, https://doi.org/10.5194/egusphere-egu23-2189, 2023.

Characterizing the types of crystalline structures that form in different environments helps us to better interpret the geologic record and deepens our understanding of mineral stability. To this end, the Dolivo – Dobrovol’sky symmetry index provides a convenient way to quantify the statistical trends in the symmetry of minerals over time (Bermanec et al. 2022). Behavior of the Dolivo - Dobrovol’sky symmetry index was investigated for different paragenetic modes of minerals (Hazen and Morrison 2022). Two datasets were used and compared (code and datasets are available on https://github.com/NoaVidovic/pgm-mineral-pairs-pg). The first one used only minerals, and each mineral was considered just once. In contrast, in the mineral–paragenetic mode pair dataset, minerals were counted once for each of the paragenetic modes in which they occurred.

The paragenetic mode dataset incorporates a number of properties associated with each of more than 60 modes of formation, including relative age and order of that mode’s first appearance, estimated minimum and maximum temperature and pressure of formation, and duration. Paragenetic mode order does not substantially affect the symmetry index of minerals. However, some trends are evident when inspecting the properties of given paragenetic modes. The symmetry indices show a strong correlation with the maximum temperature, maximum pressure, and minimum pressure of paragenetic modes they belong to (Hazen et al. 2022) with correlation coefficients of 69%, 84% and 95%, respectively when using the mineral dataset. These trends show that minerals formed at higher temperature display higher overall symmetry. Trends for pressure are enigmatic: correlations show that minerals formed at higher minimum pressure tend to favor lower symmetry, whereas minerals formed at higher maximum pressure tend to favor higher symmetry.

When using the mineral–paragenetic mode dataset, the correlation coefficients are significantly lower at 42%, 30% and 89% for maximum temperature, maximum pressure, and minimum pressure, respectively. The lower correlation coefficients obtained using the mineral–paragenetic mode pairs might indicate that the paragenetic mode is not as important in terms of trends in symmetry as initially thought. On the other hand, considering a much higher correlation coefficient for the mineral dataset, perhaps there is a more dominant effect where certain P-T conditions tend to favor certain types of symmetry at equilibrium.

How to cite: Bermanec, M., Vidović, N., Gavryliv, L., M. Hazen, R., R. Hummer, D., M. Morrison, S., Prabhu, A., and R. Williams, J.: Symmetry statistics of mineral – paragenetic mode pairs, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8924, https://doi.org/10.5194/egusphere-egu23-8924, 2023.

Recent technological advances allow geoscientists to generate high-resolution (HR) imagery using a variety of different beam-forming mechanisms (e.g. visible light, X-Rays, or charged particles such as electrons and ions). One of the main limitations in producing HR data is the required acquisition time at high magnifications. For example, back-scattered electron (BSE) mapping of a standard petrographic thin section at a resolution of 50nm/pixel takes approximately 60 days and is associated with a storage requirement in the order of 700 GB. Deep-learning methods have proven effective for resolution enhancement in regular photographic images, and in this work we present an integrated image registration and upscaling workflow to enhance image resolution, using real-world BSE datasets.

The proposed workflow requires the acquisition of one, or multiple, HR regions within a region that is imaged at low-resolution (LR). Next, close to pixel-accurate image registration is performed by using the successive implementation of two concepts: i) first the precise location of the HR region within the LR region is determined by using a Fast Fourier Transform algorithm (Lewis, 2005), and ii) final image registration is achieved by iteratively calculating a deformation matrix that, using Newton’s method of optimization, is aiming to minimize an error function describing the differences between both images (Tudisco et al., 2017).

Subsequently, matching HR and LR image pairs are fed into a Generative Adversarial Network (GAN) that learns to produce HR images from the LR counterparts. A GAN consists of two neural networks, a generator and a discriminator. The generator produces synthetic HR data based on LR input, and the discriminator attempts to classify the data as either real HR or synthetic HR. The two networks are trained together in an adversarial process, with the goal of the generator producing synthetic data that the discriminator cannot distinguish from real data.

We demonstrate our method on a variety of large real-world datasets and show that it effectively increases the resolution of full-size BSE maps up to a factor of four, while being able to resolve important features. The upscaling of BSE data, with a factor of four, is associated with a 90% reduction in beamtime and a factor 16 reduction in storage requirements. Image registration, preprocessing, and model training on a high-performance workstation takes 12-24 hours. Having a trained model, inference can be done using a regular laptop.

[1] Lewis, J. P. "Fast normalized cross-correlation, Industrial Light and Magic." unpublished (2005).

[2] Tudisco, Erika, et al. "An extension of digital volume correlation for multimodality image registration." Measurement Science and Technology 28.9 (2017): 095401

How to cite: van Melick, H. and Plümper, O.: Resolution enhancement using deep learning methods: an integrated workflow applied to real-world Back-Scattered Electron (BSE) data, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-15285, https://doi.org/10.5194/egusphere-egu23-15285, 2023.

As geochemical data enable understanding of the Earth system and help to address critical societal issues the organisation thereof is important. Questions asked about processes affecting our environment and geological past become more complex and interdisciplinary in nature as well as multidimensional. To help answer these questions within the geochemistry research capabilities and data compilations are required to be comprehensive and both human and machine readable. Various international organisations are building infrastructure to capture and distribute geochemical data in a consistent manner adhering to the FAIR principles. 

Since May 2021 the OneGeochemistry initiative has officially started efforts towards aligning these organisations’ data frameworks in order to standardise how geochemical data is reported around the globe. In November 2022 the OneGeochemistry initiative applied and was granted to become the OneGeochemistry CODATA Working Group as part of the International Science Councils Committee on Data. The initiative has now also been endorsed by the Geochemical Society, the European Association of Geochemistry and the Working Group has been endorsed by the IUGS Commission on Global Geochemical Baselines. Coordination of the OneGeochemistry initiative is funded through the WorldFAIR project where it is one of the work packages in the larger ‘WorldFAIR: Global cooperation on FAIR data policy and practice’ project. A FAIR Implementation Profile analyses of the geochemistry communities of Australia (AusGeochem), USA (EarthChem, AstroMat) and Europe (GEOROC-DIGIS, EPOS-MSL, NFDI4EARTH) resulted in recognition of the need for common vocabularies for geochemistry data reporting as one of the most important actions to undertake towards international geochemistry data interoperability. A task adopted by EarthChem-DIGIS(GEOROC)-GFZ(DataSystems) collaboration and Research Vocabularies Australia.

Here we will present an overview of the current OneGeochemistry initiative and its preliminary outcomes with regards to FAIR Implementation Profiles and processes that will help enable geochemical data interoperability between various stakeholders.

How to cite: Prent, A., Wyborn, L., Klöcking, M., Lehnert, K., Elger, K., Hezel, D., Profeta, L., ter Maat, G., Farrington, R., and Rawling, T.: The OneGeochemistry initiative as a CODATA Working Group; bringing together international geochemical data systems for easy data discovery, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-16338, https://doi.org/10.5194/egusphere-egu23-16338, 2023.

Machine learning has been shown to be a highly effective method for classifying geochemistry data, such as mineral forming environments and rock tectonics. However, it can be difficult to understand the decision-making processes of these models. To address this issue, we propose the use of Decision Boundary Maps (DBMs) as a visualization tool for interpreting machine learning models. These maps project high-dimensional geochemistry data onto a 2D plane and depict the decision boundaries in the projected space, providing a visual representation of the algorithm's decision-making processes. In addition, DBMs can reveal trends, correlations, and outliers in the data, helping to interpret the results obtained from machine learning-based geochemistry data classification. Seeing the positions of data points, rather than just class labels, is especially valuable because samples in geological categories often follow a sequence, such as a magmatic to hydrothermal transition. Observing the positions of data points allows for the identification of trends from one class to an adjacent class.

How to cite: Wang, Y., Qiu, K., Telea, A., Hou, Z., and Yu, H.: Interpreting Machine Learning Models for Geochemistry Data Classification using Decision Boundary Maps, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-10228, https://doi.org/10.5194/egusphere-egu23-10228, 2023.

The diverse suite of trace elements incorporated into apatite in ore-forming systems has important applications in petrogenesis studies of mineral deposits. Trace element variations in apatite can be used to distinguish between different types of rocks as well as discriminating between deposit types, and thus have potential as mineral exploration tools. Such classification approaches commonly employ two-variable scatterplots of apatite trace element compositional data. While such diagrams offer easy and convenient visualization of compositional trends, they often struggle to effectively distinguish deposit types because they do not employ all the high-dimensional (i.e. multi-element) information accessible from high-quality apatite trace element analysis. To address this issue, we employ, for the first time, a supervised machine learning-based approach (eXtreme Gradient Boosting, XGBoost) to correlate apatite compositions with ore deposit type, utilizing high-dimensional information. We evaluated 8629 apatite trace element data from five deposit types (porphyry, skarn, orogenic Au, iron oxide copper gold, and iron oxide-apatite) along with unmineralized apatite to discriminate between apatite in mineralized vs unmineralized systems. We could show that the XGBoost classifier efficiently and accurately classifies high-dimensional apatite trace element data according to the ore deposit type (overall accuracy: 94% and F1 score: 89%). Interpretation of the model using the SHAPley Additive exPlanations tool (SHAP) shows that Th, U, Eu and Nd are the most indicative elements for classifying deposit types using apatite trace element chemistry. Our approach has broad implications for the understanding of the sources, chemistry and evolution of melts and hydrothermal fluids resulting in ore deposit formation.

Keywords: Machine learning; apatite; Trace elements; Ore deposit type; XGBoost

How to cite: Zhou, T., Qiu, K., and Wang, Y.: From Trace Elements to Petrogenesis: A Machine Learning Approach to Determine Ore Deposit Type from Trace Elements Analysis of Apatite, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-4857, https://doi.org/10.5194/egusphere-egu23-4857, 2023.

In the last decades, the use of Raman spectroscopy on dispersed carbonaceous material in rocks has become a promising tool for geothermometry and thermal maturity assessment. In diagenesis the main problem is linked to organofacies composition and the need of time-consuming optical classification (Henry et al., 2019, Sanders et al., 2022). In this work, three different methods of clustering analysis on Raman spectra were tested as a potential approach to recognize the three main organofacies (amorphous organic matter, translucid and opaque phytoclasts) that characterize a set of 27 organic-rich samples from the Lower Toarcian source rock interval (Schistes Carton) of the Paris Basin (France).

Raman analyses were performed on concentrated organic matter obtained by acid attacks, with around 60 counts for each sample. Principal Component Analysis (PCA) was applied to reduce the dimensionality of each dataset on a 2-D score-plot. Unsupervised clustering was then performed by using three different clustering algorithms: k-means, Gaussian Mixture Models (GMM), and Density-Based Spatial Clustering for Applications with Noise (DBSCAN). The main task of these algorithms is to correctly assign the number of clusters, their size, orientation, and distribution in the score-plot that related to the heterogeneities in organofacies composition.

Results show the best performances are achieved through the application of GMM clustering that can successfully determine cluster’s geometry and optimal numbers with an accuracy mostly higher than 80% for the translucid phytoclasts group, that is the target for thermal maturity assessment.  This is a preliminary attempt showing promising application for unsupervised learning techniques coupled with Raman spectroscopy that could be applied in industrial routinely organic matter characterization or in the analysis of big dataset in both Earth and planetary sciences.

Henry, D. G., Jarvis, I., Gillmore, G., & Stephenson, M., 2019. Raman spectroscopy as a tool to determine the thermal maturity of organic matter: Application to sedimentary, metamorphic and structural geology. Earth-Science Reviews 198, 102936.

Sanders, M. M., Jubb, A. M., Hackley, P. C., & Peters, K. E., 2022. Molecular mechanisms of solid bitumen and vitrinite reflectance suppression explored using hydrous pyrolysis of artificial source rock. Organic Geochemistry 165, 104371.

How to cite: Schito, A., Vergara Sassarini, N. A., Gasparrini, M., Michel, P., and Corrado, S.: Unsupervised Clustering applied on Raman spectra of dispersed carbonaceous material: a case history from the Paris Basin (France), EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-14683, https://doi.org/10.5194/egusphere-egu23-14683, 2023.

Many macroscopic transport properties and physical processes, such as the flow of fluids through a porous medium, are directly controlled by its microstructure, specifically the presence and connectivity of individual pore spaces at micron and submicron scales. Reconstructing and evaluating the material properties of porous media plays a key role across many engineering disciplines from subsurface storage (e.g., CO2 and hydrogen) to geothermal energy and reservoir characterization. As such, the rapid and reliable characterization, evaluation, and simulation of complex pore microstructures is required not only to enhance our understanding of the fundamental processes occurring at the pore scale, but to also better estimate their material behavior on a larger scale.

These material behaviors are inherently volumetric and therefore cannot be accurately modelled using two-dimensional (2D) data alone. As a result, the accuracy of reconstruction techniques used to extract these morphological properties and spatial distributions is in part determined by the quality of available three-dimensional (3D) microstructural datasets. However, in comparison to their 3D counterparts, 2D imaging techniques are typically more cost efficient, easier to collect, and higher resolution. Our goal of generating statistically accurate 3D reconstructions of complex pore microstructural distributions based on high resolution 2D datasets is essential to bridging this dimensionality gap.

Newly explored 2D-to-3D reconstruction techniques based on deep-learning (DL) algorithms offer an alternative means of generating robust and statistically representative digital 3D rock reconstructions by measuring some spatial morphological properties and statistical microstructural descriptors (SMDs) of porous media samples from high-resolution 2D datasets. These DL models are highly flexible and capable of capturing a variety of complex microstructural features given representative 2D training datasets. In this paper, we implement a newly developed deep Generative Adversarial Network (GAN), known as SliceGAN, to synthesize novel binary digital 3D reconstructions using high-resolution 2D back-scattered electron (BSE) images obtained from thin-sections oriented in the x-, y- & z-direction.

Our trained model is capable of accurately reconstructing complex 3D microstructural features of porous media through capturing the underlying (micro-)structural and morphological properties contained in the original sample (2D) thin-sections. To demonstrate the effectiveness of our trained model, we conducted a comparative analysis between the generated 3D reconstructions and real sample datasets by evaluating morphological properties (volume fraction, surface area, equivalent diameter, pore orientations, etc.) as well as the widely popular SMD the two-point correlation function (S₂ (r) ). The resulting reconstructions are virtually indistinguishable, both visually and statistically, from the real sample. Our research paves the way for quickly and accurately describing complex heterogenous media for the prediction of transport processes, for example, carbon and hydrogen storage and extraction.

How to cite: Vogel, H., Amiri, H., Arias, A., Plümper, O., and Ohl, M.: 3-D Reconstructions of Porous Media from 2-D input via Generative Adversarial Networks (GANs), EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-15241, https://doi.org/10.5194/egusphere-egu23-15241, 2023.

Reconstructing the emergence of the modern continental crust is crucial to understand the evolution of the crust, the onset of plate tectonics, elemental cycling, and long-term climate. However, it remains highly contentious about when and how the major subaerial continental crust emerged over time. Here, we used a machine learning (ML) model (XGBoost) to reveal the exposed history of continental crust over the last 3.8 billion years ago (Ga). First, we compiled ~10,000 modern subaerial or submarine basalts with major and trace elements to train the ML model. Then, the trained ML model (with resampling) was utilized to predict and calculate the mean proportions of subaerially erupted continental basaltic rocks since 3.8 Ga. The result suggested that the subaerial proportion only reached about 50% at ~2.5 Ga, indicating the exposure of the continental crust was far from the present-day level at the end of the Archean era. On the other hand, since ~1.8 Ga, the subaerial proportion of the continental crust exhibited a dynamic balance between ~60% and 80%, reaching the present-day level.

How to cite: Liu, C., Keller, C. B., Liu, X., and Zhang, Z.: The emergence of Continental Crust Revealed by Machine Learning, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-4648, https://doi.org/10.5194/egusphere-egu23-4648, 2023.