ausblenden:
Schlagwörter:
-
Zusammenfassung:
Quantifying impacts on the environment and human health is a critical requirement for geological subsurface utilisation projects. In practice, an easily accessible interface for operators and regulators is needed so that risks can be monitored, managed, and mitigated. The primary goal of this work was to create an environmental hazards quantification toolkit as part of a risk assessment for in-situ coal conversion at two European study areas: the Kardia lignite mine in Greece and the Máza-Váralja hard coal deposit in Hungary, with complex geological settings. A substantial rock volume is extracted during this operation, and a contaminant pool is potentially left behind, which may put the freshwater aquifers and existing infrastructure at the surface at risk. The data-driven, predictive tool is outlined exemplary in this paper for the Kardia contaminant transport model. Three input parameters were varied in a previous scenario analysis: the hydraulic conductivity, as well as the solute dispersivity and retardation coefficient. Numerical models are computationally intensive, so the number of simulations that can be performed for scenario analyses is limited. The presented approach overcomes these limitations by instead using surrogate models to determine the probability and severity of each hazard. Different surrogates based on look-up tables or machine learning algorithms were tested for their simplicity, goodness of fit, and efficiency. The best performing surrogate was then used to develop an interactive dashboard for visualising the hazard probability distributions. The machine learning surrogates performed best on the data with coefficients of determination R2>0.98, and were able to make the predictions quasi-instantaneously. The retardation coefficient was identified as the most influential parameter, which was also visualised using the toolkit dashboard. It showed that the median values for the contaminant concentrations in the nearby aquifer varied by five orders of magnitude depending on whether the lower or upper retardation range was chosen. The flexibility of this approach to update parameter uncertainties as needed can significantly increase the quality of predictions and the value of risk assessments. In principle, this newly developed tool can be used as a basis for similar hazard quantification activities.
