Development and calibration of protocols and methods for quantifying the hydrogeological interference risk of tunnelling in different geological settings

This PhD research, co-financed by Italferr S.p.A., addresses the hydrogeological impacts induced by tunnel excavation in mountainous regions, with particular focus on early-stage risk prediction during preliminary design. The main objective is to enhance the ability to anticipate potential interference with groundwater flow systems and springs, thereby supporting environmental risk management, cost–benefit evaluations, and the planning of effective mitigation measures. Following a comprehensive methodological review of analytical, empirical, parametric, numerical, and machine learning–based approaches, the most relevant parameters controlling groundwater inflows and alterations of natural discharge processes were identified. This analysis enabled the refinement of the existing Drawdown Hazard Index (DHI), improving the clarity of physical cause–effect relationships between hazard (tunnel inflow) and impact on environmental receptors. A data-driven predictive model based on the Random Forest algorithm was then developed. Hazard and vulnerability parameters, integrated with additional geomechanical and hydrogeological descriptors available at the preliminary design stage, were used as input variables. The model was trained and calibrated using two well-documented case studies: the Bologna–Florence high-speed railway tunnels and the Gran Sasso highway twin tunnels. Both provided reliable pre- and post-excavation hydrogeological monitoring data, ensuring robust calibration. Validation was performed on three additional case studies (Badia, San Giovanni Pass, and Brenner Base Tunnel), achieving very good predictive performance according to multiple metrics (Accuracy, Precision, Recall, F1-score, MCC, and AUC). SHAP analysis improved model transparency by quantifying parameter importance, highlighting Inflow Potential, Fracture Frequency, and Groundwater Flow System as the most influential factors. Finally, an innovative DNA-labelled silica-encapsulated nanoparticle tracer was successfully tested in a fractured arenitic aquifer, demonstrating strong potential for identifying preferential flow paths and validating tunnel–aquifer hydraulic connections.