Using Natural Language Processing to estimate the grey water footprint of mines

Despite water playing a critical role in nearly every stage of mining activities, substantial uncertainty remains about the extent to which mine operations pollute downstream water bodies. In this study, we develop and parameterize a transferable model to estimate the grey water footprint (GWF) of mine sites. The GWF represents the volume of water required to dilute mine-derived pollutants to safe levels in receiving waters, accounting for both pollutant release rates and natural background concentrations. Applying the GWF concept to systematically evaluate the water quality impact of      large scale mine operations is challenging due to the diversity of pollutants and emission pathways, and because relevant data is scarce, uncertain, and dispersed across numerous text sources. We address this challenge by combining natural language processing and probabilistic estimation. NLP is used to infer from publicly available documents plausible concentration ranges and treatment or immobilization efficiencies across processing steps. We then reduce parameter dimensionality and propagate uncertainties through sensitivity analysis and Monte Carlo simulations. We demonstrate the model’s practical utility by applying it to a representative copper mining site. The strength of our approach lies in its versatility: it adapts to available data at the site level while producing outputs that are readily comparable across sites and linkable to mine typologies, supporting more effective water and pollutant management strategies.