SZSteven E. Zhang

Steven E. Zhang

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Steven E. Zhang has a total of 54 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    The Effects of Smoothing-Induced Weak Annotation in Mineral Prospectivity Mapping

    by Steven E. Zhang, Mohammad Parsa

    The meaning of targets in data-driven mineral prospectivity mapping (DD-MPM) is foremost dependent on the quality of data. The resolving power of annotated data depends on the ratio of annotated area or volume and target footprint. At sufficiently large-scales of DD-MPM, it is impossible to match annotated area and covariate resolution to the footprint of mineral deposits. Consequently, information exposed to data modelling is unrestricted to between a target and its features. By definition, this type of annotation is a form of weak annotation, whose predictor–predictand relationship is weakened geostatistically through smoothing (a high ratio of annotated area or volume to target footprint). Annotation strength in DD-MPM is unstudied, although it is a research frontier in adjacent domains such as medical image segmentation. In this study, we present experimental results that reveal the spatial and aspatial significance of smoothing-induced weak annotation. Major findings include: (a) model performance improves with increasing smoothing, which is consistent with the linearization of the predictor–predictand relationship; (b) prospective area decreases with increasing smoothing, as smoothing weakens the signature of positive samples; (c) multifocal workflow uncertainty increases, caused by the same mechanism as in (b); and (d) the advantage of nonlinear and more expressive models is diminished with increasing smoothing. Therefore, although annotation strategy in DD-MPM seems innocuous, it has profound effects on the meaning of target area and exploration models. These effects impact: (a) the in-field validation of targets or model deployment expectations; (b) the mathematical meaning of trained models and targets; (c) the choice of model complexity; and (d) comparative analysis of workflow design (e.g., algorithms).

  • Open Access

    Targeting Sensitivity and Selectivity in Data-Driven Mineral Prospectivity Mapping in Response to Mineral System-Based Geodata Annotation: Experimental Studies Using Pan-Canadian Copper Annotation and Philosophical Implications

    by Steven E. Zhang, Mohammad Parsa, Laura-Jane Eagleson, Christopher J. M. Lawley, Dianne Mitchinson, Renato Cumani

    The nature of targets extracted through data-driven mineral prospectivity mapping (DD-MPM) is dependent on geodata annotation. However, geodata annotation is noisy because mineral systems are complex systems and the definition of positivity changes with annotator, jurisdiction and time. Consequently, historically annotated data (HAD), which are the exhaustive collection of labeled samples (sites) that have accumulated over the history of exploration, is a noisy and subjective source of ground truth. However, the spatial sensitivity and selectivity of DD-MPM products to annotation is quantitatively unknown. Although HAD are always used at large scales of MPM, the cost-to-benefit analysis of improving and augmenting them is unclear because of inherent limitations of DD-MPM, including: (1) scientific confounds that ironically bias targeting toward the known as opposed to truly greenfield; (2) research design challenges that create circular logic during validation; and (3) geoscience and implementation uncertainty that make DD-MPM products heuristic. We conducted two experiments to: (1) verify the distribution of noisy positive samples in HAD and test whether a minority of samples carry a majority of information; and (2) assess the effectiveness of mineral system constraints by annotating maximally separable mineral systems of a commodity and relating their target overlap with class separability. Using a HAD database containing all copper samples in Canada, we observed that: (1) about 14% of the cleanest samples in HAD generated 71% of all spatial targets; and (2) fully separable mineral systems generated spatially overlapping targets, which is a hypothesis-generating finding. Together, these results (a) provide scientific evidence for annotation noise, its characteristics and impact; and (b) highlight the value of HAD and the need for more rigorous frameworks of mineral system-based DD-MPM.


  • Expanded models of Canadian critical mineral prospectivity maps for magmatic Ni (±Cu ±Co ±PGE), Mississippi–Valley Type Pb-Zn and clast-dominated Pb-Zn

    by S. E. Zhang, M. Parsa

    Ce rapport présente des modèles de prospectivité étendus à l’échelle canadienne pour les systèmes minéraux Ni magmatique (±Cu ±Co ±ÉGP), Pb-Zn de type Mississippi-Valley (MVT) et Pb-Zn à prédominance clastique (PC). Ces modèles sont des agrégations d’ensembles profonds, conçus pour refléter la diversité des approches entre praticiens et pour intégrer l’incertitude induite par les flux de travail. Contrairement aux modèles inauguraux, ils ne reposent pas sur un seul flux de travail, mais sur des propriétés statistiques d’ensembles larges et décorrélés. Par rapport aux modèles initiaux, ces modèles étendus offrent une expansion des zones cibles en terrains vierges (‘greenfield’), apportant ainsi de nouvelles perspectives de ciblage. Pour le Ni magmatique, les covariables ont été mises à jour avec une résolution spatiale plus élevée, et le système d’indexation spatiale a été standardisé. Les ensembles de données contenant les résultats de prédiction pour le Ni magmatique sont disponibles pour le Canada, tandis que les résultats pour les cibles Pb-Zn de type MVT et PC couvrent l’Amérique du Nord et l’Australie. Des échantillons positifs pour le Ni magmatique, ré-indexés selon le système H3, sont également inclus dans cette publication en forme de données.

  • Modèles étendus de cartes de prospectivité des minéraux critiques au Canada pour le Ni magmatique (±Cu ±Co ±ÉGP), le Pb-Zn de type Mississippi-Valley et le Pb-Zn à prédominance clastique

    by S. E. Zhang, M. Parsa

    Ce rapport présente des modèles de prospectivité étendus à l’échelle canadienne pour les systèmes minéraux Ni magmatique (±Cu ±Co ±ÉGP), Pb-Zn de type Mississippi-Valley (MVT) et Pb-Zn à prédominance clastique (PC). Ces modèles sont des agrégations d’ensembles profonds, conçus pour refléter la diversité des approches entre praticiens et pour intégrer l’incertitude induite par les flux de travail. Contrairement aux modèles inauguraux, ils ne reposent pas sur un seul flux de travail, mais sur des propriétés statistiques d’ensembles larges et décorrélés. Par rapport aux modèles initiaux, ces modèles étendus offrent une expansion des zones cibles en terrains vierges (‘greenfield’), apportant ainsi de nouvelles perspectives de ciblage. Pour le Ni magmatique, les covariables ont été mises à jour avec une résolution spatiale plus élevée, et le système d’indexation spatiale a été standardisé. Les ensembles de données contenant les résultats de prédiction pour le Ni magmatique sont disponibles pour le Canada, tandis que les résultats pour les cibles Pb-Zn de type MVT et PC couvrent l’Amérique du Nord et l’Australie. Des échantillons positifs pour le Ni magmatique, ré-indexés selon le système H3, sont également inclus dans cette publication en forme de données.


  • Open Access

    Recursive Annotation for Negative Labeling in Data-Driven Mineral Prospectivity Mapping

    by Steven E. Zhang, Daniel Coutts, Mohammad Parsa, Renato Cumani, Aaron Thompson

    Data engineering is a challenge in mineral prospectivity mapping using supervised data-driven methods because of a general paucity of true samples, both positive and negative. Naïve random sampling of unlabeled samples as presumptive negative samples is a common method, but it creates label crossover, which is the condition that otherwise true positives would be reversed in class designation. For mineral prospectivity mapping, label crossover has two main effects: (1) it artificially decreases the area of positive prospectivity because label crossover is asymmetrical; and (2) it increases model variance by increasing model complexity and, therefore, weakens the spatial continuity of predictions. These effects reduce the realism and trustworthiness of spatial analytic products (e.g., maps), which hinders their adoption. Here, we propose a data-driven and recursive annotation method based on positive and unlabeled learning to annotate negative samples. Our method progressively biases the negative labeling of unlabeled samples using prospectivity scores (as a class prior). The optimization of the number of recursions occurs through a joint maximization of the variable (aspatial) and spatial domain objectives of prospectivity mapping. Reducing label crossover improves the performance of a deep ensemble (of unique workflows) and increases spatial continuity of the resulting prospectivity map. Only two iterations of annotation were necessary for the refinement of a graphite prospectivity map in Canada. By its design, recursive annotation is broadly compatible with a range of workflows.


  • Open Access

    Mineral Prospectivity Modeling of Graphite Deposits and Occurrences in Canada

    by Steven E. Zhang, Christopher J. M. Lawley, Julie E. Bourdeau, Mohammad Parsa, Renato Cumani, Aaron Thompson

    Exploration for graphite in Canada is of economic, strategic and governance priority. In this study, we aimed to develop a reliable prospectivity map for graphite in Canada. Our approach mitigated multiple sources of workflow-induced uncertainty by propagating uncertainty due to the selection of negative labels, machine learning algorithms, feature space dimensionality, and hyperparameter tuning metrics. By averaging an ensemble of de-correlated models, we produced a single-merged model that clearly represents propagated uncertainty through a consensus map and an uncertainty map. These maps adhere to the metrological convention of "result plus/minus associated uncertainty" and are intuitive to use. Our ensemble demonstrated robustness, quickly converging to the consensus model, suggesting that new mineral prospectivity mapping (MPM) products using the same data would unlikely perturb our consensus model’s coverage. We conducted a maximally double-blind study, avoiding geoscientific knowledge during model generation to ensure impartial post-hoc analysis and interpretation. Therefore, our MPM products complement geoscientific knowledge-based exploration, because the targeting information provided in our MPM products constitute a maximally independent source. Our MPM products showed excellent spatial variability, aligning with existing knowledge of graphite deposits in Canada, indicating that combining data-driven rigor with independent interpretation enhances the robustness of our MPM products. Consequently, we believe our MPM products could effectively guide regional exploration of natural graphite in Canada.


  • Open Access

    A systematic framework for compilation of critical raw material lists and their importance for South Africa

    by Glen T. Nwaila, Julie E. Bourdeau, Steven E. Zhang, Nelson Chipangamate, Imraan Valodia, Muhammad Ahsan Mahboob, Thakaramahlaha Lehohla, Mulundumina Shimaponda-Nawa, Raymond J. Durrheim, Yousef Ghorbani


  • Open Access

    Workflow-Induced Uncertainty in Data-Driven Mineral Prospectivity Mapping

    by Steven E. Zhang, Christopher J. M. Lawley, Julie E. Bourdeau, Glen T. Nwaila, Yousef Ghorbani

    The primary goal of mineral prospectivity mapping (MPM) is to narrow the search for mineral resources by producing spatially selective maps. However, in the data-driven domain, MPM products vary depending on the workflow implemented. Although the data science framework is popular to guide the implementation of data-driven MPM tasks, and is intended to create objective and replicable workflows, this does not necessarily mean that maps derived from data science workflows are optimal in a spatial sense. In this study, we explore interactions between key components of a geodata science-based MPM workflow on the geospatial outcome, within the modeling stage by modulating: (1) feature space dimensionality, (2) the choice of machine learning algorithms, and (3) performance metrics that guide hyperparameter tuning. We specifically relate these variations in the data science workflow to the spatial selectivity of resulting maps using uncertainty propagation. Results demonstrate that typical geodata science-based MPM workflows contain substantial local minima, as it is highly probable for an arbitrary combination of workflow choices to produce highly discriminating models. In addition, variable domain metrics, which are key to guide the iterative implementation of the data science framework, exhibit inconsistent relationships with spatial selectivity. We refer to this class of uncertainty as workflow-induced uncertainty. Consequently, we propose that the canonical concept of scientific consensus from the greater experimental science framework should be adhered to, in order to quantify and mitigate against workflow-induced uncertainty as part of data-driven experimentation. Scientific consensus stipulates that the degree of consensus of experimental outcomes is the determinant in the reliability of findings. Indeed, we demonstrate that consensus through purposeful modulations of components of a data-driven MPM workflow is an effective method to understand and quantify workflow-induced uncertainty on MPM products. In other words, enlarging the search space for workflow design and experimenting with workflow components can result in more meaningful reductions in the physical search space for mineral resources.


  • Open Access

    Denoising of Geochemical Data using Deep Learning–Implications for Regional Surveys

    by Steven E. Zhang, Julie E. Bourdeau, Glen T. Nwaila, Mohammad Parsa, Yousef Ghorbani

    Regional geochemical surveys generate large amounts of data that can be used for a number of purposes such as to guide mineral exploration. Modern surveys are typically designed to permit quantification of data uncertainty through data quality metrics by using quality assurance and quality control (QA/QC) methods. However, these metrics, such as data accuracy and precision, are obtained through the data generation phase. Consequently, it is unclear how residual uncertainty in geochemical data can be minimized (denoised). This is a limitation to propagating uncertainty through downstream activities, particularly through complex models, which can result from the usage of artificial intelligence-based methods. This study aims to develop a deep learning-based method to examine and quantify uncertainty contained in geochemical survey data. Specifically, we demonstrate that: (1) autoencoders can reduce or modulate geochemical data uncertainty; (2) a reduction in uncertainty is observable in the spatial domain as a decrease of the nugget; and (3) a clear data reconstruction regime of the autoencoder can be identified that is strongly associated with data denoising, as opposed to the removal of useful events in data, such as meaningful geochemical anomalies. Our method to post-hoc denoising of geochemical data using deep learning is simple, clear and consistent, with the amount of denoising guided by highly interpretable metrics and existing frameworks of scientific data quality. Consequently, variably denoised data, as well as the original data, could be fed into a single downstream workflow (e.g., mapping, general data analysis or mineral prospectivity mapping), and the differences in the outcome can be subsequently quantified to propagate data uncertainty.