TIME SERIES ANALYSIS METHODS FOR IDENTIFICATION OF GROUNDWATER IMPACTS FROM UNDERGROUND CONSTRUCTION PROJECTS IN URBAN AREAS

Construction of underground facilities often impacts groundwater levels, which may lead to subsidence-related damage, effect of objects with high nature-values, impact of groundwater and geothermal heat wells and mobilization of pollutants. These negative effects are in turn associated with potentially high costs. Effects on groundwater levels are generally difficult to identify in aquifers in urban areas that are highly perturbed by known and unknown human activity. Here, we compare the suitability of time series analysis methods to make traceable the extent to which specific infrastructure projects perturb the groundwater in space and time. Among these methods are traditional regression methods (Attanayake & Waterman, 2006), index-based methods (Heudorfer, Haaf, Stahl, & Barthel, 2019), transfer function noise models (TFN) (Von Asmuth, 2012), as well as machine learning methods such as long short-term memory (LSTM) and Nonlinear autoregressive exogenous models (NARX) (Wunsch, Liesch, & Broda, 2021). Results of these models are combined with cluster analysis to map groundwater impact in time and space to attribute magnitude of perturbations to different constructions sites. We show examples on how this approach can be used to reduce uncertainties associated with the cause of an effect. This allows for fair attribution of costs to the causer of impacts and confident decision analysis regarding timely mitigation measures at the correct location and time before high costs arise.

Attanayake, P. M., & Waterman, M. K. (2006). Identifying environmental impacts of underground construction. Hydrogeology Journal, 14(7), 1160-1170. doi:10.1007/s10040-006-0037-0

Heudorfer, B., Haaf, E., Stahl, K., & Barthel, R. (2019). Index-Based Characterization and Quantification of Groundwater Dynamics. Water Resources Research, 55(7), 5575-5592. doi:10.1029/2018wr024418

Von Asmuth, J. R. (2012). Groundwater System Identification Through Time Series Analysis.

Wunsch, A., Liesch, T., & Broda, S. (2021). Groundwater level forecasting with artificial neural networks: a comparison of long short-term memory (LSTM), convolutional neural networks (CNNs), and non-linear autoregressive networks with exogenous input (NARX). Hydrol. Earth Syst. Sci., 25(3), 1671-1687. doi:10.5194/hess-25-1671-2021