The hydrogeological connection between tunnel excavation and springs is a key aspect to be assessed during the preliminary design phase, both for environmental and socio-economic reasons. Based on a preliminary hydrogeological survey and environmental monitoring of the springs, the need to anticipate potential impacts at an early stage influences both the feasibility of the project and the design of mitigation measures in more critical areas.
A data-driven Machine Learning (ML) approach, designed to incorporate the complexity of the relationships between various physical and hydrogeological parameters contributing to the risk of spring impact due to tunneling, was calibrated using a dataset from detailed hydrogeological monitoring conducted alongside the excavation of two major tunnels in the Apennines (Italy), involving sedimentary and carbonate karst aquifers.
The approach shows good scores for model evaluation, and here we present the results of a further validation against datasets from two additional sites: one in the Alps (Brenner Base Tunnel, within a crystalline aquifer) and another in the Apennines (Bologna-Florence highway pass variant, within a turbiditic sedimentary setting). The application of the method to these new sites—one of which features a geological setting different from the original validation dataset— shows good scores again, demonstrating its potential for broader generalization.
The results were compared with those of the Drawdown Hazard Index (DHI), demonstrating that both methods can effectively identify risk while also highlighting the sensitivity of the latter. Specifically, by adjusting the various thresholds, highly accurate results can be achieved. Conversely, the ML models generate outputs that are not subject to modification or classification into discrete categories.