Application of a data-driven approach for well field modelling


Work package WP 5.2 “Combination of Managed Aquifer Recharge (MAR) and adjusted conventional treatment processes for an Integrated Water Resources Management“ within the European Project TECHNEAU (“Technology enabled universal access to safe water”) investigates bank filtration (BF) + post-treatment as a MAR technique to provide sustainable and safe drinking water supply. One of the tasks within the project is the testing of a data-driven approach for the identification (pattern recognition) and quantification of the key processes that drive the groundwater (GW) dynamics in observation wells (OW) near well fields of a BF waterworks. For this BUSSE (2010) used a multivariate statistical method (principal component analysis - PCA) with daily GW level time series of 41 OWs and was able to identify four processes that explained 95% of the total variance in the data set. On the one hand GW recharge (58.9%) and its temporal delay (3.3%) explain 62% of the GW level fluctuations within the study period. On the other hand any discernible impact of waterworks abstractions is limited to one of the three well fields with the highest production rate (29.8% of explained variance). In addition the infiltration of a marshy ditch into the GW accounts for another 2.9% of the GW level fluctuations. Regarding the ability to identify driving forces for GW level fluctuations the main advantage for using PCA compared to process-driven GW flow modelling is that the driving forces for GW level fluctuations can be identified and quantified without requiring exact knowledge about the structural properties of the subsurface (e.g. aquifer transmissivities) and its input parameters (e.g. GW recharge, production rates). Note that the latter do not enter the PCA directly but are used for spatiotemporal interpretation of the results, which also requires some expertise. In addition, it is recommended to perform a sensitivity analysis of the PCA results in a next step, so that it can be tested whether the processes identified above are robust in case of changing input parameters such as: - Reduced spatiotemporal resolution - Study period with different boundary conditions (e.g. pumping regime). The contents of this report were presented to the involved experts from the Berliner Wasserbetriebe (BWB). In agreement with their recommendations it was decided to focus further research within follow-up projects on the (i) sensitivity analysis of the PCA results and (ii) to apply nonlinear approaches for identification and quantification of processes that drive GW quality dynamics within the study area.

Kompetenzzentrum Wasser Berlin gGmbH
Rustler, M.
Rustler, M.
Data scientist

My research interests include reproducible research, data management and programming (R & Python).