IAMG Days 2021

The year 2021 will mark the 40th anniversary of the first work on contaminant source identification in aquifers: the Ph.D. thesis by Steven Gorelick at Stanford University. Forty years and close to one hundred sixty papers later, research on contaminant source identification has grown exponentially in numbers but not in quality. The most recent papers focus on the part of the problem that researchers know how to solve but leave aside those aspects that will make research valuable in practice. The talk will make a historical review on the subject and will identify the dos and don'ts to continuing investigation on the subject.

Informationization of geological and mineral resource works need large amount of compound talents with geological science knowledge and information technology. The key education problem of the compound talent at undergraduate stage is how to realize the knowledge fusion of geological science, computer technology and spatial information technology. Our cultivating target is focused on training student with ability of application system development. The corresponding course system and training plan are established based on configuration method of knowledge structure tree. Course system has three layers of discipline basic courses, main major courses and kernel major courses. Teaching method of “Real Case + Simulation Project” is set up based on software engineering and project engineering management. Multi-level integration practice teaching system is built up with customized geological information system software and application systems. Based on disciplinary foundation of geological information science, the education system for interdisciplinary knowledge fusion is constructed, and achieved results in the practice of new major of Spatial Information and Digital Technology.

In this talk we give an introduction to data-driven discovery in mineralogy with an emphasis on Mineral Ecology. Large and growing data resources on mineral species and their localities are employed to identify and parameterize frequency distributions of Earth’s mineral kingdom. We estimate Earth’s undiscovered, mineralogical diversity in a Bayesian framework and predict the percentage of observed mineral species that would differ if Earth’s history were replayed. The relative abundances are calculated numerically for all the mineral species in Earth’s crust, and can be used to characterize Earth in terms of mineralogy and uniqueness in the cosmos. We discuss chance versus necessity and to what extent the distribution of mineral species in Earth’s near-surface environment primarily is a consequence of deterministic factors (common to any Earth-like planet) or rather chance events (random, unlikely fortunes). The integration of various deep-time data enables data-driven discovery in the geosciences. The talk will conclude with some recent results regarding the classification of presolar silicon carbide grains found in meteorites and future goals for the modeling of mineral network data.

In this presentation, I will start with some remarks about Matheron’s theory of regionalized variables regarding robustness and efficiency in variogram estimation, as well as parametric variogram models allowing for transition from second-order to intrinsic stationarity, and transformed Gaussian random fields. I will then describe a large-scale problem of quantifying the uncertainty of global climate model outputs such as temperature or wind energy potential based on geostatistical simulations from models fitted to more than 1 billion space-time data values. I will advocate the visualization of the results based on apps available for smartphones with virtual reality (VR) mode. Finally, I will present ExaGeoStat, a high-performance unified software for geostatistics on manycore systems that has exascale computing capability, that is, one billion billion (10^18) calculations per second. It can run on one’s laptop, desktop, cluster, GPUs, and the supercomputers of tomorrow.

One of the main objectives in the analysis of data in geochemical exploration programs is the selection and application of suitable classification models to separate signals related to the effects of mineralisation from other processes or sources, as well as controls on variation in background geochemical distributions. The data addressed in these cases is commonly presented in the form of geochemical maps at various scales, sampling densities and sampling geometries. Various mathematical and statistical models have been applied to regional geochemical data to reveal subtle geochemical patterns related to the effects of different styles of mineralisation or variations in parent lithologies. Conventional statistical methods mostly have a number of limitations. This has encouraged investigation of less conventional alternatives to conventional parametric methods, including fractal modelling and geostatistical simulation to isolate such signals or patterns. It is however the case that the application of just one classification model provides a single classification map whose precision (stability) and efficiency are unknown. Classification models based on more than one approach permit assessment of such uncertainty, but the effect of different assumptions in the model may be difficult to assess. Precision can be assessed using simulation methods on a given modelling approach. This presentation will review some of the well-established as well as newly developed fractal/multifractal classification models and the relevant uncertainties, in addition to the future potential developments.

Forest covered area in the Bailing Cu-Pb deposit area suffered from metallic pollution caused by road dust resuspension and deposition. Portable X-ray fluorescence (pXRF) was used to determine elemental concentrations in the cut surface of tree stumps in the forest and soils at its location. Arsenic, Zn, and Fe concentrations in the cut surface of tree stumps and in soils displayed obvious zonation along the direction perpendicular to the road in the mining area. Arsenic and Zn concentrations are highest in the area nearest to the road and gradually decreases as the distance to the road increases, conforming to the power-law distribution. Micro X-ray fluorescence (micro-XRF) scanning images suggest that elements such as As and Zn mainly concentrate on the cut surface of tree stumps, indicating that pollution was caused by resuspension and deposition of road dust. Dust blown from roads is a major pathway of arsenic exposure in the Bailing Cu-Zn deposit area. Tree stumps are good “bioindicators” of metallic pollution caused by road dust resuspension and deposition, especially in mining areas with variable lithological and geological character of bedrock and soils, where distinguishing contaminated from uncontaminated areas is difficult.

Machine learning methods are increasingly used to spatially predict a target variable in the presence of auxiliary information available everywhere within the study region. Even though machine learning methods such as random forest show competitive prediction performance, they do not exactly honor the target variable's observed values at sampling locations by construction. On the other side, competitor geostatistical methods like regression-kriging perfectly match the target variable's observed values at sampling locations. In many geoscience applications, it is often desirable to perfectly match the observed values of the target variable at sampling locations. This talk will present a random forest-based approach in which the target variable is exactly conditioned to the data. The proposed method's effectiveness is illustrated on a real-world dataset comprising geochemical concentration data and compared with the regression-kriging and the classical random forest.

It is well known that parameters controlling water flow and solute transport are highly heterogeneous at all scales of measurement. Parameter spatial variability results in scale effects, that is, parameter values depend on the measurement support. If there is a difference of scales between observation and model scales there are two possible ways to consider it: either models are constructed with elements of a size similar to that at which the data were measured, or some upscaling rules must be defined. Drawing on field, laboratory and numerical studies I provide results of the analysis of heterogeneity of different parameters that control water flow and mass transport in a tropical soil. Experimental evidence is used to define upscaling rules for hydraulic conductivity, dispersivity and partition coefficient of potassium and chloride. An analysis of the use of sophisticated and basic techniques to perform stochastic upscaling is also presented and compared with some theoretical results.