Measuring and predicting the ζ-potential of anthropogenic TiO2 nanoparticles in surface waters

We propose a novel approach for determining and predicting the ζ-potential of nanoparticles in surface waters based on the water composition and environmental parameters. Applying the dialysis bag method, five different types of TiO₂ nanoparticles representing the most common TiO₂ particles in commercial products were exposed in situ to a set of representative surface waters. The ζ-potentials of these environmentally coated particles ranged from −58 to 13 mV and were used together with water composition data to train models for predicting the ζ-potential from the water composition. With an average root mean square error of 3.6 mV for 50 generated models, the XGBoost models outperformed random forest and various linear models. We explored these models using parameter importance and shap values. Furthermore, we characterized the surface coating of a selection of samples using XPS and TG-QMS. Using these techniques, we could confirm the presence of an organic coating and explore the connections between ζ-potential values and environmental coating. As expected from batch experiment studies, the concentration of divalent cations is the most important factor for predicting the ζ-potential of environmentally coated nanoparticles in surface waters. We found that the quality of dissolved organic matter has a significant effect, whereas pH and dissolved organic matter content are less important. This study demonstrates the potential of our in situ exposure method combined with multivariate analysis to explore the fate of nanoparticles in aquatic environments.