Approaches to hazard-oriented groundwater management based on multivariate analysis of groundwater quality

Abstract

Drinking water extracted near rivers in alluvial aquifers is subject to potential microbial contamination due to rapidly infiltrating river water during high discharge events. The heterogeneity of river-groundwater interaction and hydrogeological characteristics of the aquifer renders a complex pattern of groundwater quality. The quality of the extracted drinking water can be managed using decision support and HACCP (Hazard Analysis and Critical Control Point) systems, but the detection of potential contamination remains a complex task to master. The methodology proposed herein uses a combination of high-resolution measurements and multivariate statistical analyses to characterise actual groundwater quality and detect potential contamination. The aim of this project was to improve the protection of riverine groundwater extraction wells and to increase the degrees of freedom available to the management of fluvial planes with drinking-water production and aquifer recharge by river-groundwater interaction. The monitoring network was set up in the Reinacherheide in North-west Switzerland and encompassed the depth-oriented installation of multiparameter instruments, a surface-water monitoring station and a flow-through cell with an automated sampler and high-precision measurement instruments. The parameters recorded included temperature, electrical conductivity, spectral absorption coefficient, particle density and turbidity. Two of the observation wells were equipped with a telemetry system and the flow cell could be controlled remotely. The well-field encompassed eight groundwater extraction wells. The optimal choice of observation wells and indicator parameters was assessed using principal component analysis of groundwater head, temperature and electrical conductivity time-series to detect the influence of, for example, river-water infiltration or river-stage fluctuations on the time-series recorded in the groundwater observation wells. Groundwater head was susceptible to pressure waves induced by both river-stage fluctuations and groundwater extraction. Temperature time-series showed only weak responses to high discharge events. Electrical conductivity, however, showed a distance-driven response pattern to high discharge events. To further assess the representative strength of individual groundwater quality indicator parameters for identifying microbial contamination, a bi-weekly and a high-resolution sampling campaign were carried out. The results showed high faecal-indicator bacteria densities (E. coli and Enterococcus sp.) at the beginning of high discharge events, followed by a rapid decrease, leading to a strong hit-and-miss characteristic in the bi-weekly sampling campaign. The third approach applied used the neural network-based combination of self-organizing maps and Sammon's projection (SOM-SM) to detect shifts in groundwater quality system states. The nonlinear analysis was carried out with groundwater head, temperature and electrical conductivity time-series from six observation wells. The subsequent shading of the projected trajectory of system states with independent time-series (spectral absorption coefficient and particle density) allowed the identification of critical system states, when actual groundwater quality decreased and contamination of the extraction wells was imminent. The time at which the changes in system state occurred and were detected were used as potential warning indicators for the water supplier. The effects of altered groundwater extraction (as a consequence of the SOM-SM warning) were then simulated using a groundwater flow model. The outcome of the SOM-SM analysis is, thus, proposed as an interface between the monitoring system and extraction-well management system. The proposed approach incorporates hydrogeological knowledge and the analysis of prevalent conditions concerning river-groundwater interaction with real-time telemetric data transfer, data-base management and nonlinear statistical analysis to detect deterioration in actual groundwater quality due to rapidly infiltrating river water. As the SOM-SM is not based on threshold values and independent of indicator parameters, the approach can be transferred to other sites with similar characteristics.