EGU23-7060, updated on 09 Jan 2024
EGU General Assembly 2023
© Author(s) 2024. This work is distributed under
the Creative Commons Attribution 4.0 License.

Estimating Pathogen Removal in Bank Filtration – A Methodology for the Construction of Surrogate Models to Assist Decision Making

Dustin Knabe1, Aronne Dell'Oca2, Alberto Guadagnini2, Monica Riva2, and Irina Engelhardt1
Dustin Knabe et al.
  • 1Technische Universität Berlin, Department of Hydrogeology, Berlin, Germany (
  • 2Politecnico di Milano, Dipartimento di Ingegneria Civile e Ambientale, Milano, Italy

Induced bank filtration is a known method for sustainable drinking water production in regions with limited groundwater resources. However, this method is at risk from surface water contaminations, e.g., by pathogens. Numerical models simulating pathogen fate in groundwater are typically too complex to be used as standard tools by waterworks managers or environmental agencies. To mitigate this problem, we present a methodology for the construction of easy-to-use reduced order models as surrogates for complex numerical reactive transport models for pathogens and pathogen indicators in induced bank filtration.

First, a streamlined one-dimensional numerical model was set up for the reactive transport of pathogens and pathogen indicators in induced bank filtration. Processes in the model include advection-dispersion, inactivation, attachment to and detachment from the sediment as well as straining and the presence of a clogging layer. Model parameters are divided into two groups: Group A includes site specific parameters for which values are typically available (with limited uncertainty) for management- and engineer-level users (e.g., grain size, distance of extraction well to the river); Group B includes process parameters which are typically affected by high uncertainty (e.g., inactivation and detachment coefficients).

We rely on an extensive dataset for coliforms and somatic coliphages collected over a 16-month period at an active induced bank filtration site. Stochastic inversion is used to assess uncertainty for model parameters of Group B (constrained on the dataset), while those of Group A are set to the values of the specific site. Principal component analysis (PCA) is applied to reduce the dimensionality of model parameter space and correlation amongst the uncertain parameters of Group B. A surrogate model is then constructed through generalized polynomial chaos expansion (gPCE). In this, the value range of Group A parameters is based on typical scenarios for induced bank filtration sites, while the range of the PCA-reduced Group B parameters is set to the uncertainty identified in the stochastic inversion.

The surrogate model allows to evaluate, at significantly reduced computational cost, the removal of coliforms and somatic coliphages in induced bank filtration based on user-defined values for parameters of Group A, but also including the uncertainty stemming from parameters of Group B. The surrogate model estimates for removal are in good agreement with observed removals for coliforms and somatic coliphages at the monitored site and with other (albeit limited) datasets from induced bank filtration sites found in the published literature. At this stage, the obtained surrogate model can be considered as a prototype. The assessment of its full potential requires additional extensive validation against other field sites. In general, surrogate models together with the overall methodological framework we propose, can be seen as a promising tool to assist informed decisions about pathogen transport at induced bank filtration sites.

How to cite: Knabe, D., Dell'Oca, A., Guadagnini, A., Riva, M., and Engelhardt, I.: Estimating Pathogen Removal in Bank Filtration – A Methodology for the Construction of Surrogate Models to Assist Decision Making, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-7060,, 2023.