Abstract: Most of the World’s germanium (Ge) is mined from Ge-rich lignite, where it is commonly associated with elevated arsenic (As), tungsten (W) and beryllium (Be) contents. Over the past decade, new evidence showing that World-class Ge deposits result from the interaction of hydrothermal fluids with organic matter in coal seams has emerged. Yet, the chemical state of Ge and associated metals in lignite remains poorly understood. We used Mega-pixel Synchrotron X-ray Fluorescence (MSXRF), X-ray Absorption Near Edge Structure (XANES) and Extended X-ray Absorption Fine Structure (EXAFS) to characterize the oxidation states and chemical bonding environment of Ge, As, and W in two world-class Mesozoic Ge-in-lignite deposits (Lincang, Yunnan, southwestern China; Wulantuga, Inner Mongolia, northeastern China); in lignite-bearing uranium (U) ores from the Beverley deposit (South Australia) hosted in Eocene sandstones; and in lignite and preserved wood in late Oligocene-Miocene fluviatile sediments (Gore, Southland, New Zealand). The aim was to improve our understanding of the enrichment mechanism of Ge in lignite and better evaluate the environmental mobility of Ge and some of the associated metals (specifically As and W) in lignite ores. In all samples, chemical maps show that Ge is distributed homogeneously (down to 2μm) within the organic matter. XANES and EXAFS data show that Ge exists in the tetravalent oxidation state and in a distorted octahedral coordination with O, consistent with complexing of Ge by organic ligands. In some pyrite-bearing samples, a minor fraction of Ge is also present as Ge(IV) in association with pyrite. In contrast, As displays a more complex speciation pattern, sometimes even in a single sample, including As(III), As(V), and As(−I/+II) in solid solution in sulfides. Arsenic in sulfides occurs in anionic and cationic forms, i.e., it shows both the common substitution for S22− and the substitution for Fe recently discovered in some hydrothermal pyrites. Tungsten was present as W(VI) in distorted octahedral (3+3) coordination. The EXAFS data confirm the absence or minor contribution of individual W-rich minerals such as scheelite or ferberite to W mass balance in the studied samples. These data show that Ge, W, and probably some As are scavenged via formation of insoluble, oxygen-bridged metal organic complexes in lignite. Destruction of the organic ligands responsible for fixing Ge and W (As) in these lignites is required for liberating the metals, e.g. from waste materials. Geochemical modelling suggests that Ge, W, Be and As all can be extracted from granitic rocks by dilute, low temperature hydrothermal fluids. Germanium is transported mainly as the tetrahedral Ge(OH)4(aq) complex, but fixed as an octahedral oxy-bridged organic complex. The same situation is valid for W, which is transported at the tetrahedral tungstate ion, but most likely scavenged via formation of a 6-coordinated metal-organic species. The Ge-Be-W±As association in Ge-rich coals reflects the source of the metals as well as related scavenging mechanisms.

Abstract: The crucial role which groundwater resource plays in our environment and the overall well-being of all living things can not be underestimated. Nonetheless, mismanagement of resources, over-exploitation, inadequate supply of surface water and pollution have led to severe drought and an overall drop in groundwater resources’ levels over the past decades. To address this, a groundwater flow model and several mathematical data-driven models have been developed for forecasting groundwater levels. However, there is a problem of unavailability and scarcity of the on-site input data needed by the data-driven models to forecast the groundwater level. Furthermore, as a result of the dynamics and stochastic characteristics of groundwater, there is a need for an appropriate, accurate and reliable forecasting model to solve these challenges. Over the years, the broad application of Machine Learning (ML) and Artificial Intelligence (AI) models are gaining attraction as an alternative solution for forecasting groundwater levels. Against this background, this article provides an overview of forecasting methods for predicting groundwater levels. Also, this article uses ML models such as Regressions Models, Deep Auto-Regressive models, and Nonlinear Autoregressive Neural Networks with External Input (NARX) to forecast groundwater levels using the groundwater region 10 at Karst belt in South Africa as a case study. This was done using Python Mx. Version 1.9.1., and MATLAB R2022a machine learning toolboxes. Moreover, the Coefficient of Determination (R2);, Root Mean Square Error (RMSE), Mutual Information gain, Mean Absolute Percentage Error (MAPE), Mean Squared Error (MSE), Mean Absolute Error (MAE), and the Mean Absolute Scaled Error (MASE)) models were the forecasting statistical performance metrics used to assess the predictive performance of these models. The results obtained showed that NARX and Support Vector Machine (SVM) have higher performance metrics and accuracy compared to other models used in this study.

Abstract: Predicting groundwater levels is pivotal in curbing overexploitation and ensuring effective water resource governance. However, groundwater level prediction is intricate, driven by dynamic nonlinear factors. To comprehend the dynamic interaction among these drivers, leveraging machine learning models can provide valuable insights. The drastic increase in computational capabilities has catalysed a substantial surge in the utilisation of machine learning-based solutions for effective groundwater management. The performance of these models highly depends on the selection of hyperparameters. The optimisation of hyperparameters is a complex process that often requires application-specific expertise for a skillful prediction. To mitigate the challenge posed by hyperparameter tuning’s problem-specific nature, we present an innovative approach by introducing the automated machine learning (AutoML-GWL) framework. This framework is specifically designed for precise groundwater level mapping. It seamlessly integrates the selection of best machine learning model and adeptly fine-tunes its hyperparameters by using Bayesian optimisation. We used long time series (1997-2018) data of precipitation, temperature, evaporation, soil type, relative humidity, and lag of groundwater level as input features to train the AutoML-GWL model while considering the influence of Land Use Land Cover (LULC) as a contextual factor. Among these input features, the lag of groundwater level emerged as the most relevant input feature. Once the model is trained, it performs well over the unseen data with a strong correlation of coefficient (R = 0.90), low root mean square error (RMSE = 1.22), and minimal bias = 0.23. Further, we compared the performance of the proposed AutoML-GWL with sixteen benchmark algorithms comprising baseline and novel algorithms. The AutoML-GWL outperforms all the benchmark algorithms. Furthermore, the proposed algorithm ranked first in Friedman’s statistical test, confirming its reliability. Moreover, we conducted a spatial distribution and uncertainty analysis for the proposed algorithm. The outcomes of this analysis affirmed that the AutoML-GWL can effectively manage data with spatial variations and demonstrates remarkable stability when faced with small uncertainties in the input parameters. This study holds significant promise in revolutionising groundwater management practices by establishing an automated framework for simulating groundwater levels for sustainable water resource management.