Accurate characterization of fracture evolution in jointed rock is essential for understanding its mechanical behavior and failure mechanisms. This study proposes a unified framework that integrates Digital Image Correlation (DIC) and deep-learning-based crack detection to quantitatively link deformation localization with crack evolution in anchored jointed rock. Uniaxial compression tests were conducted on rock-like specimens with different joint inclinations and anchoring conditions to capture full-field deformation and fracture responses. DIC results indicate that anchoring conditions significantly influence strain localization and crack propagation. Compared with unanchored specimens, both free and restrained anchoring reduce strain concentration, with restrained anchoring exhibiting a stronger suppression effect and delaying crack coalescence. To enable automated crack identification, an improved YOLOv8n-DWConv-C2FN model was developed and trained using image data derived from DIC-monitored experiments. The model provides quantitative crack localization information through bounding boxes, enabling the extraction of crack distribution and morphology features. By combining the detected crack regions with DIC-derived strain localization fields, spatial consistency metrics (R1, R2) and a composite index (CPCA) were proposed to quantitatively evaluate their correspondence. The results demonstrate that the proposed approach effectively correlates deformation localization with crack evolution, with higher consistency observed for dominant crack propagation paths. The proposed methodology provides a quantitative pathway for integrating deformation-field measurements with data-driven crack recognition, enabling more reliable fracture monitoring and analysis in rock mechanics and geotechnical engineering.
The identification of geochemical weak anomalies in data-gap or minimally explored areas-where labeled mineral occurrence data are scarce or completely absent-poses a significant challenge to conventional supervised methods that rely on known positive samples for training. This study proposes a fully unsupervised framework that combines optimized Multiple-Point Statistics (MPS) with a Fusion Convolutional Autoencoder (FCAE), applied to geochemical data from 1:50,000-scale stream sediments in the Gaoqiao-Jiugong (Edongnan) area in southeastern Hubei, China. Under the principle of multifractal scale invariance, the MPS training image (TI) construction strategy was optimized to capture high-order spatial structural patterns, generating geologically consistent spatial distributions while preserving multi-scale heterogeneity and mitigating the over-smoothing effects inherent in conventional interpolation methods. These MPS outputs serve as high-quality input representations for subsequent unsupervised FCAE, extracting composite weak anomalies through reconstruction error analysis without any pre-labeled mineral occurrence data. Application results demonstrate that the MPS-generated maps exhibit superior structural conformity with known geological features compared with IDW and Kriging methods. Quantitative validation based on the maximum percentile ranking within a 500 m buffer around 18 known mineral occurrences (2 deposits, 14 occurrences, 2 low-potential showings) shows that MPS-FCAE achieves a mean percentile of 0.60% in high-potential occurrences, outperforming the IDW-FCAE counterpart (0.79%), with prominent improvements in deposit-scale anomaly detection (0.77% vs. 1.58%) and a maximum 45-fold enhancement for weak Au and Hg anomalies. Crucially, MPS-FCAE correctly downgraded two low-potential showings to percentiles above 6%, while IDW-FCAE retained them within the top 3% due to the bullseye effect. The composite anomalies derived from FCAE are associated with known Au deposits and reveal previously unidentified exploration targets in data-gap regions, while yielding fewer false-positive anomalies than conventional manual anomaly delineation methods. These findings demonstrate that the proposed unsupervised MPS-FCAE framework effectively addresses the issues of inaccurate spatial structural characterization and insensitive detection of weak anomalies in medium-scale (1:50,000) geochemical surveys, offering a label-independent, transferable methodology for mineral exploration in sparsely sampled terrains.
Convolutional Neural Networks (CNNs) offer powerful image classification capabilities in geology, where visual interpretation of sedimentary structures is essential for reconstructing depositional environments. However, their “black-box” nature can limit trust and slow adoption in geoscientific workflows. This study presents a CNN-based framework for classifying eleven physical and biogenic sedimentary structures from core images, while treating explainability as a core objective. Three architectures (EfficientNet-B2, MobileNet-V3, and ResNet-50) were trained and evaluated on a geologically diverse annotated dataset. EfficientNet-B2 achieved the highest overall accuracy (97.7%), followed by MobileNet-V3 (96.8%) and ResNet-50 (95.8%). Blind-test evaluation on 401 unseen images showed that the CNN models retained substantial predictive capability under domain-shift conditions, with accuracies of 79.05% for EfficientNet-B2, 78.8% for MobileNet-V3, and 78.3% for ResNet-50. These results indicate good overall generalization, while the remaining errors were mainly associated with geologically similar sedimentary structures. Explainability was assessed using Grad-CAM++ and, crucially, was evaluated quantitatively using (i) deletion-based faithfulness, measuring the drop in target-class probability after masking the most salient regions, and (ii) stability under controlled perturbations (brightness changes, noise, blur, and small spatial shifts), measuring similarity between explanation maps. Across models, Grad-CAM++ consistently emphasized geologically meaningful features (e.g., pebble boundaries, inclined laminae, and biogenic textures) while de-emphasizing non-diagnostic artifacts. ResNet-50 produced the most faithful and stable explanations, showing the steepest probability decay under deletion and the highest robustness of saliency patterns under perturbations, despite slightly lower classification accuracy. Overall, these results demonstrate that high classification accuracy does not necessarily imply high explainability quality, and that explanation reliability metrics provide a practical route to more transparent and trustworthy sedimentary structure classification for real-world core analysis.
The Marmousi2 model is a widely used elastic benchmark in exploration geophysics, cited in over 600 papers. However, no resistivity model has been published for Marmousi2, which limits its use in controlled-source electromagnetic (CSEM) and joint seismic–EM inversion research. We present Marmousi2-Res, a resistivity companion to Marmousi2 derived from its elastic properties using Gassmann fluid substitution and Archie’s law. The model was classified into seven lithologies at the pixel level, and porosity was recovered for all ten hydrocarbon zones by back-calculating from the Gassmann equation and the density relation. Zones that share common wet-rock properties or mineralogy were grouped to better constrain the inversion. Resistivity was then calculated from Archie’s equation using temperature-corrected brine resistivity from the Arps relation. We generated resistivity grids for six uniform water saturations (Sw=15% to 100%) and two split-saturation scenarios with distinct gas- and oil-zone saturations. The mean P-wave velocity error after round-trip fluid substitution is 4 m/s (<0.4%), and the recovered matrix density is constrained to 2.56–2.62 g/cm3. A sensitivity comparison shows that resistivity contrast increases from 2.0× to 11.1× as hydrocarbon saturation increases from 30% to 70%, while velocity anomalies change by only ∼2% over the same range. Averaged over all zones, resistivity is approximately 263 times more sensitive than P-wave velocity to this saturation change. 1D CSEM modelling shows that the reservoirs buried 600–1600 m below the seafloor are detectable with standard marine CSEM acquisition, while the deepest zones at approximately 2500 m are marginally detectable at low frequency and become undetectable where a salt tongue overlies them. The model, lithology maps, rock-physics tables, and all code are publicly available at https://github.com/YogiIdaBagus/marmousi2-resistivity and archived at https://doi.org/10.5281/zenodo.21754423.
The time-domain induced polarization (TDIP) method is frequently used in mineral resource exploration. Surface exploration frequently encounters topography, such as slope, mountains, and valleys, which significantly hinders the accurate data interpretation. Furthermore, it is impossible to ignore the negative effects of target anomalous entities, which often display arbitrary anisotropy. Therefore, the development of a high-precision modeling technique for TDIP models in complex situations is essential. To address the challenges posed by arbitrary anisotropy and topography, this research integrates adaptive finite element methods with unstructured meshes. An adaptive modeling algorithm is proposed for TDIP models in complex geological settings. First, we simulate a spherical anomaly's TDIP responses under various mesh levels on flat topography. The findings show that the apparent chargeability are greatly impacted by mesh quality. We then investigate the impact of slope topography and anisotropic parameters on TDIP data based on the adaptive mesh level, and the results indicate that they have an undeniable impact on TDIP data. Finally, an anisotropic metal deposit model beneath a mountain valley is investigated with unstructured adaptive meshes. These models highlight the precision and adaptability of our established methodology, demonstrating its high accuracy and exceptional adaptability for anisotropic problems and complex topographies.
Geochemical anomaly identification plays a critical role in mineral exploration because mineralization-related geochemical patterns provide important evidence for delineating prospective ore-forming zones. However, geochemical survey data are commonly affected by heterogeneous background and complex inter-element relationships. Mineralization-related anomalies are typically sparse and weak, making them difficult to distinguish from surrounding background variations. These characteristics limit the effectiveness of conventional statistical methods and supervised learning models, particularly when reliable mineralization labels are scarce. To address these challenges, this study proposes a prior-guided self-supervised contrastive Transformer network for sparse geochemical anomaly identification (CTGA). The framework learns discriminative inter-element representations directly from unlabeled geochemical data through geochemically constrained contrastive representation learning, while a known mineral occurrence is used only during inference as a prototype to guide similarity-based anomaly retrieval in the learned feature space. Spatial refinement is further applied to improve anomaly continuity and suppress isolated false-positive responses. The proposed framework was evaluated in the Zhaishang–Liba gold ore cluster area in the Western Qinling Mountains, China. Experimental results demonstrate that CTGA produces geologically meaningful anomaly patterns with high spatial consistency to known mineralization and outperforms representative comparative methods, achieving an AUC of 95.21% under the adopted evaluation protocol. These results indicate that the proposed framework provides an effective solution for geochemical anomaly identification in mineral exploration scenarios where labeled mineralization information is extremely limited.
The Open Neural Network Exchange (ONNX) is widely promoted as a framework-independent solution for deep learning model portability, but systematic evidence of its reliability across languages, frameworks, and deployment environments remains limited for Earth Observation (EO) workflows. This paper introduces an eight-step evaluation protocol for assessing ONNX inference portability of EO deep learning models, demonstrated through two case studies under documented software conditions. For a TempCNN crop-type classifier, native-to-ONNX deviation was within 10−6 RMSE for both PyTorch and Flux exports, and cross-runtime outputs across Python, R, and Julia were bitwise-identical under the CUDAExecutionProvider. For a ConvLSTM precipitation nowcasting model, three conversion pathways were tested: the Keras 3-on-PyTorch pathway (KPT) passed all eight protocol steps, including a directly computed RMSEport=2.81×10−7; the other two pathways required workarounds or were blocked from native re-import. ONNX portability is architecture-dependent in the tested configurations. Static convolutional models built from common operators port cleanly, while recurrent models with explicit temporal loops expose distinct failure modes that require toolchain-specific validation. The protocol’s demonstrated scope covers these two architectures and software conditions; its behavior on other model families should be confirmed through fresh application.
Clay minerals, known for their distinctive properties such as plasticity, adsorption, and catalytic behavior, play a vital role in hydrocarbon exploration, production, and reservoir characterization. Traditional methods for quantifying clay content typically rely on core analysis and mineralogical logs, which are costly and often prone to uncertainty due to their indirect nature. This study introduces a novel machine learning-based approach to estimate the weight percentage of clay (WCL%) in Kitchen Lights Unit 1 (KLU 1), located in the North Cook Inlet Field, Alaska, USA. The methodology integrates six machine learning models: three individual boosting algorithms CatBoost (CB), Gradient Boosting (GB), and Extreme Gradient Boosting (XGB) alongside three hybrid stacking models. The hybrid models employ combinations of CB, GB, and XGB as base regressors, with Multiple Variable Regression (MVR) as the meta-model. A curated dataset combining well-log measurements and X-ray diffraction (XRD) analysis is used to train and evaluate these models. The hybrid model 3 (HM_3) outperform all others, with the best model achieving an R2 of 0.9658 and a mean squared error (MSE) of 0.7331. In contrast, the CB model yields the lowest performance (R2 = 0.931, MSE = 1.480) among the boosting models. Overall, HM_1, HM_1 (GS), HM_2, HM_3, GB, and XGB all deliver R2 values exceeding 0.95, demonstrating their robustness in estimating WCL (%) from conventional well logs. SHAP analysis identified Thermal Neutron Porosity (TNPH) as the most influential feature. Finally, the proposed methodology is validated in the Montney Formation, British Columbia, confirming its generalizability across diverse geological settings.
Large-scale 3D reconstruction from UAV imagery is essential for remote sensing and environmental monitoring, yet modern neural rendering methods such as 3D Gaussian Splatting (3DGS) are computationally intensive due to redundant views. Reducing input images while preserving reconstruction accuracy and completeness remains a practical challenge. We present a scalable image subset selection framework that integrates transformer-based visual place recognition (PairVPR) with a facility-location selection strategy to identify and remove redundant views while maintaining spatial coverage. We first validate the approach on six standard benchmark scenes (five Mip-NeRF360 scenes and the Tanks&Temples Truck scene): under a fixed image budget, PairVPR-based selection attains the best average PSNR, SSIM, and LPIPS among all evaluated VPR selectors. Building on this, we present a case study on a real-world glacier UAV survey of 589 images, in which pruning to a compact subset reduces overall runtime by more than half at the most aggressive setting, while incurring only a modest loss in reconstruction quality, and in which PairVPR-based selection again attains the best reconstruction quality among the evaluated VPR selectors. By combining learned similarity with a coverage-driven selection process, the method operates with linear memory complexity and adapts to large image collections, offering a practical solution for large-scale geospatial mapping and monitoring workflows.
This study develops a hybrid convolutional neural network (CNN) ensemble-learning framework for landslide susceptibility prediction in Uttaradit Province, Thailand. Three hybrid models, CNN-RF, CNN-LightGBM, and CNN-XGBoost, are trained using a 30-year landslide inventory and 14 environmental conditioning factors covering topographic, geological, hydrological, ecological, and anthropogenic domains. Models are evaluated through buffer-based spatial cross-validation with a 500 m exclusion radius to reduce bias from spatial autocorrelation. Performance is assessed using AUROC, F1-score, Precision, Recall, and Accuracy. Additional diagnostics, including train–test gap, Moran’s I on out-of-fold residuals, and expected calibration error (ECE), assess overfitting, residual spatial autocorrelation, and calibration. Boosting-based hybrid models outperform the bagging-based hybrid in predictive performance and spatial discrimination. Among the tested models, CNN-LightGBM achieves the best trade-off between predictive performance and efficiency, with an AUROC of 0.9699, an accuracy of 0.9696, and the lowest computational cost. The proposed hybrid deep-ensemble learning framework, supported by spatially rigorous validation, provides an effective and reproducible approach for landslide susceptibility mapping in tropical mountainous environments.
Flooding in the Teesta River Basin, northern Bangladesh, frequently threatens lives, agriculture, and infrastructure due to intense monsoonal rainfall, transboundary flows, and low-lying floodplains. This study presents an integrated probabilistic flood risk assessment by combining statistical flood frequency analysis, 1D–2D hydrodynamic modelling (HEC-RAS), and GIS-based socio-economic vulnerability assessment using the Analytic Hierarchy Process (AHP). Long-term discharge and water level records were analyzed using six probability distributions: Normal, Log-normal, Pearson Type-III, Log-Pearson Type-III, Gumbel, and Generalized Extreme Value. Based on the goodness-of-fit tests (Chi-square, Kolmogorov-Smirnov, and Anderson-Darling), the Generalized Extreme Value and Pearson Type-III distributions were selected for the upstream and downstream stations, respectively, to estimate design discharges and water levels for the 50-, 100-, 200-, and 500-year return periods, along with the 5th and 95th percentile confidence intervals. Calibration and validation of the 1D model were conducted at Kaunia station, yielding RMSE of 0.183–0.273 m, NSE of 0.917–0.967, and PBIAS of −0.46 to +0.22, indicating excellent predictive performance. The 2D model was validated using Receiver Operating Characteristic analysis, achieving an Area Under the Curve value of 0.864. Coupled 1D–2D simulations revealed 34–40% of the basin inundated across return periods, with maximum depths exceeding 7 m in main channels. AHP-based vulnerability mapping, integrating nine socio-economic and physical indicators, identified moderate to high vulnerability across most of the study area. Combined flood hazard and vulnerability analysis indicated that over 20% of the basin falls under high and very high flood risk, with the most affected upazilas (sub-districts) including Lalmonirhat Sadar, Kaunia, Gangachara, Dimla, and Aditmari. This integrated framework highlights the combined influence of hydrological extremes and socio-economic exposure on flood risk, offering a transferable approach for data-limited deltaic basins to support targeted mitigation, land-use planning, and resilience enhancement.
Recent machine learning methods automate volcanic event detection, yet many struggle to generalize across event types and often require extensive labelled, eruption-specific training data. We propose an unsupervised diffusion-based approach using a Denoising Diffusion Probabilistic Model (DDPM), formulating the detection as anomaly identification: volcanic events are deviations from the learned normal behaviour of continuous seismic data, represented by characteristic tremor features derived from the ambient seismic field. The method applies a forward diffusion that injects noise into observed seismic signals and a reverse denoising that reconstructs the clean counterpart. During this reconstruction, anomalous components are effectively replaced by the model's learned representation of normal seismic behaviour. We compute anomaly scores as residuals between reconstructed and original signals, highlighting abnormal/eruptive activity. Leveraging the generative capacity of DDPMs, the model captures complex variability in the ambient seismic field and its characteristic functions. Evaluations on real-world volcanic datasets show effective detection of anomalous events with state-of-the-art performance. Additional tests on benchmark multivariate anomaly datasets also outperform prior methods. Under a field-standard 48-h forecasting protocol, the label-free model attains AUC up to 0.87 on phreatic systems, matching a supervised state-of-the-art forecaster without using eruption labels. These findings show that diffusion-based models offer a robust framework for automated volcanic monitoring, providing anomaly evidence localised in time and across input channels and demonstrating short-term forecasting skill.
The Hackett River zone, located in the northeastern part of the Slave Geological Province, Nunavut, Canada, is a prolific metallogenic zone hosting two principal styles of mineralization: volcanogenic massive sulfide (VMS) and banded iron formation (BIF)-related gold mineralization. The region has been the focus of both legacy and modern lake sediment geochemical surveys, with the modern program involving the reanalysis of archived samples for an expanded suite of elements. Despite the modernization of this geochemical database, delineating mineralization-related anomalies associated with the distinct mineralization styles remains challenging. The first challenge arises from the multivariate and compositional nature of regional geochemical data, which requires appropriate statistical treatment to address the closure effect and accurately identify meaningful multi-element geochemical associations. The second challenge relates to the geological complexity of the study area, characterized by heterogeneous lithologies, deformation patterns, and overlapping geochemical signatures that complicate conventional anomaly detection approaches. To address these challenges, this study reformulates geochemical anomaly delineation as a bagging-based positive–unlabeled learning problem integrated with compositional data analysis, deep convolutional neural networks, generative adversarial networks, and large language models. An improved cross-validation framework was further implemented to rigorously evaluate the robustness of the generated anomaly maps. The resulting geochemical anomaly maps successfully reduce the exploration search space and demonstrate strong spatial correspondence with the distribution of favourable bedrock units and known mineralized trends, indicating that the proposed framework is effective for the interpretation of regional lake sediment geochemical survey datasets.
This study investigates the dependency and correlation of the Cole-Cole model (CCM) parameters using Bayesian inference to invert spectral induced polarization (SIP) data. The aim is to improve the understanding of subsurface properties and provide a reliable interpretation of the estimated subsurface models by analyzing parameter interdependencies in detail. We present a novel 2.5D inversion framework specifically developed for spectral induced polarization (SIP) data. This approach leverages Python-based libraries and advanced statistical inversion techniques, employing an affine-invariant ensemble sampling-based Markov chain Monte Carlo (McMC) algorithm. To evaluate the efficacy of our methodology, we conduct synthetic modeling and McMC inversion across various subsurface scenarios, including a homogeneous Earth model, a two-layer medium, and a model containing two buried anomalies within a homogeneous background. Our methodology allows for the extraction of CCM parameters that represent electrical properties, providing a deeper insight into complex geological structures. Visualizations of McMC chains and corner plots illustrate the interdependencies among CCM parameters, demonstrating the convergence and reliability of parameter estimates. Through validation against synthetic models, we highlight the accuracy and effectiveness of our methodology. Overall, our study showcases the potential of Bayesian inversion in improving geophysical data interpretation and deepening our understanding of the interdependencies among CCM parameters in estimated models across a range of simplified geological scenarios.
Thematic maps are indispensable for environmental monitoring, resource management, and policy-making, yet their value depends critically on rigorous accuracy assessment. AcATaMa (Accuracy Assessment of Thematic Maps) is a free, open-source QGIS plugin that delivers an integrated workflow for thematic map accuracy assessment and sample-based area estimation, enabling users to follow international good practice guidelines within a single, accessible environment. The plugin integrates three core components: sampling design (simple random, stratified random, and systematic), response design (a multi-window interface for simultaneous review of multiple reference data sources), and analysis (error matrices, accuracy metrics, area-adjusted estimates, and confidence intervals). By unifying these components, AcATaMa addresses a critical gap: no prior open-source tool provides a complete, end-to-end accuracy-assessment workflow within a desktop GIS environment. Its effectiveness is demonstrated through a case study assessing Colombia's National Forest Change Map for 2018–2019. Global adoption, over 160,000 downloads and application in 86 peer-reviewed studies across diverse domains, confirms the tool's broad scientific relevance and societal impact.
Accurate spatial prediction from sparse measurements is a fundamental challenge in geoscience applications, including reservoir characterization, environmental monitoring, and subsurface modeling. This study introduces a kriging-informed machine learning framework that integrates geostatistical methods with modern neural architectures to enhance prediction accuracy under data-scarce conditions. The approach augments sparse real measurements with a dense ordinary-kriging grid to generate synthetic training samples, constructing a spatial prior k(x,y) evaluated at arbitrary locations via inverse-distance weighting over kriged nodes. Input features combine centered coordinates (Xc,Yc), low-order polynomial terms (Xc2,Yc2,XcYc), and the kriging-based spatial prior k(x,y). Three established methods including Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Multilayer Perceptron (MLP) are benchmarked against the Kolmogorov–Arnold Network (KAN), an advanced model that replaces fixed activations with learnable univariate basis functions. Models are trained on 60% of data using a multi-fidelity strategy (synthetic + real with 10× replication) and evaluated on a held-out 40% test set. KAN achieves superior performance: test R2=0.934 (95% CI: [0.843, 0.967]), RMSE =133.89 (95% CI: [102.27, 163.69]), representing a 31% error reduction versus ordinary kriging alone (R2=0.859). Sensitivity analysis across synthetic-to-real ratios (3:1 to 10:1) confirms robust performance (R2=0.934±0.002), and five-fold cross-validation yields consistent generalization (R2=0.876±0.098). The framework is generalizable to spatial estimation tasks with sparse measurements and demonstrates that hybrid geostatistical-machine learning approaches achieve accurate, stable predictions from limited field data.
Prediction of non-convergent characteristics in a non-conforming geomaterials is very essential due to unpredictable nature of their behaviour. One of the important factors contributing to these materials mechanical behaviour is compositions (chemical and mineralogical). This paper presents soft computing techniques for the prediction of non-convergent behaviour in iron tailings considering compositional attributes. This was achieved by adopting artificial neural network (ANN) and adaptive neuro-fuzzy inference systems (ANFIS) to predict non-convergent behaviour in iron tailings. Both ANN and ANFIS successfully predicted non-convergent behaviour using chemical composition (silicon oxide SiO2, aluminum oxide Al2O3, iron oxide Fe2O3), mineralogical composition (quartz Qz, hematite He, kaolinite Ke) and combined compositions (SiO2, Al2O3, Fe2O3, Qz, He, Ke). Overall, ANFIS proved to be the most appropriate predictive models for transitional behaviour using chemical composition, mineralogical composition and combined compositions with an average performance indices R2, a20, a10, MAE, RMSE and MAPE of 0.94, 1.00, 0.95, 0.013, 0.029 and 2.96 respectively. Influence of sample preparations is observed in ANN model using chemical composition, mineralogical composition and combined compositions. However, sample preparations do not have effect in ANFIS model using chemical composition, mineralogical composition and combined compositions. This further confirms that ANFIS will predict non-convergent behaviour better than the ANN. Sensitivity analysis showed that hematite exert strongest control over the transitional behaviour followed by the quartz, iron oxide, silicon oxide, kaolinite and aluminum oxide.
Conventional mineralogical and geochemical laboratory-based analyses, though accurate, are often resource-intensive and destructive. Infrared Hyperspectral core imaging (HSI) enables rapid, cost-effective, and non-destructive characterization of mineralogical assemblages and lithologies, making it a valuable tool to complement and optimize traditional mineral characterization workflows. This study evaluates HSI's ability to estimate P2O5 concentrations directly from phosphate drill cores. We propose an integrated methodology that combines hyperspectral imaging across the Visible-Near Infrared (VNIR), Shortwave Infrared (SWIR), and Midwave Infrared (MWIR) spectral ranges with portable X-ray fluorescence (pXRF). This was applied to a ∼65-m-long drill core from a sedimentary phosphate deposit. Two strategies were tested: (i) an indirect method, estimating P2O5 content by quantifying carbonate fluorapatite (CFA), the primary phosphorus-bearing mineral in sedimentary phosphates; (ii) a direct method, by predicting P2O5 content using various machine learning regression models trained on a pXRF core dataset. The indirect estimation method, despite exhibiting a general correspondence with the pXRF-derived P2O5 profile, demonstrated low agreement of 25% and 62% in the SWIR and MWIR regions, respectively. Direct estimation of P2O5 using KNN on MWIR spectra was the most effective approach, offering significant potential for P2O5 chemical profiling, with ∼87% agreement and a low mean absolute error of 2.78%. These promising results highlight the potential of this geochemical method to generate orebody knowledge, supporting the integration of the geometallurgy approach into the phosphate mining value chain.
Nearshore–beach–dune systems serve as critical natural defenses against storm surge and wave action. Their morphological evolution is governed by dynamically coupled wave and wind-driven sediment transport processes. This study presents a novel algorithm that extends the 1D wave and sediment transport model CSHORE into a quasi-2D model, enabling simulations of long-term evolution of coastal morphology and shoreline. This algorithm is implemented using the Basic Model Interface (BMI) to establish a quasi-2D hydro-morphodynamic model BMI-CSHORE, which is verified against analytical solutions for shoreline change. The good agreement between model results and analytical solutions demonstrates the model capability to resolve both cross-shore and alongshore sediment transport and capture the shoreline change. Furthermore, BMI-CSHORE is coupled with the full 2D aeolian sediment transport model AeoLiS via BMI for simulating nearshore–beach–dune evolution. The integrated framework BMI-CA is applied to simulate morphodynamic evolution at Duck, North Carolina, USA. The presented methodology can be adopted to other 1D wave and sediment transport models to capture complex coastal processes in a quasi-2D fashion. Together, BMI-CSHORE and BMI-CA provide efficient and scalable tools for predicting multi-scale coastal morphodynamics.
