Towards Benchmarking Time Series Analysis with Process-Based Groundwater Models. The Case of Hydrogeological Impacts of Tunnel Construction

Abstract

This thesis evaluates time series modelling for investigating groundwater impacts due to tunnelling in an urban environment based on a process-based, benchmark groundwater model. The Haga site in Gothenburg, Sweden, as part of the Västlänken infrastructure project was used as case study area. Utilizing datasets from various sources, including climate data from SMHI, head observations from the Haga site, and MODFLOW simulations. The study employs MODFLOW simulations and time series analysis to simulate and evaluate GW dynamics. Through Python-based transfer function noise modelling (TFN) using the Pastas package, the study constructs time series models to assess potential tunnel leakage and its impact on GW levels. The thesis emphasizes the importance of accurate data collection and precise modelling techniques to correctly calibrate the model to show the effects on GW systems. The calibration of the MODFLOW model showed good correlation with observed groundwater data, but urban complexities and model limitations caused discrepancies. Refinements, such as improved calibration techniques and improving the representation of groundwater recharge, could enhance model accuracy. The TFN models demonstrated strong performance, especially with added stress data of tunnel leakage. This indicates that the ability of TFN models to investigate groundwater impacts can be benchmarked with a groundwater flow model. However, this study also highlighted challenges due to data scarcity, leading to mismatches between groundwater observations and simulations with the benchmark model, which in future studies could be addressed with more advanced calibration and integration techniques. Future research should focus on refining these models and investigating the skill of TFN models when more groundwater impacts are present.