Machine learning-based prediction of gaps in groundwater time series in the overburden of the Morsleben radioactive waste repository, Germany

Understanding the movement of radionuclides over time is crucial for assessing the integrity of geological formations as barrier for a radionuclide waste repository. Long-term groundwater potential time series enable the modelling of flow and transport scenarios, which help to predict how radionuclides may migrate from the repository through the overburden into the biosphere, if the overburden as geological barrier should fail. The accuracy of numerical flow and transport models depend on the availability of reliable input data, such that long-term groundwater potential time series help to ensure that numerical flow and transport scenarios accurately represent the complex hydrogeological processes occurring over time.

However, in practice it is very common that, due to financial constraints, vandalism of measurement devices, and other logistical problems result in shorter and/or longer gaps in the ideally continuous groundwater monitoring time series. These gaps can significantly hinder the reliability and completeness of the dataset, making it challenging to perform accurate analyses.

In response to these challenges, we use machine-learning methods with monthly precipitation data from the German meteorological service (DWD), monthly groundwater recharge data generated from the hydrological model RUBINFLUX and continuous groundwater time series from state run monitoring wells as inputs to predict the missing gaps in the groundwater potential time series in the overburden of the radioactive waste repository Morsleben (ERAM).

This approach highlights the importance of continuity in the dataset for further studies, modelling, and safety assessments for radioactive waste repositories. Using machine learning techniques can help to reconstruct the missing data and provide a more comprehensive and continuous dataset for validating and calibrating numerical flow and transport models. 

 

References:

Bear, J., 1972. Dynamics of Fluids in Porous Media. American Elsevier, New York.

Langkutsch, U., Käbel, H., Margane, A., & Schwamm, G. (1998). Planfeststellungsverfahren zur Stillegung des Endlagers für radioaktive Abfälle Morsleben. 457. 

Peche, A., Kringel, R., Orilski, J., & Skiba, P. (2021). Hydrogeologische Modellbildung des ERA Morsleben. In Zwischenbericht Bundesanstalt für Geowissenschaften und Rohstoffe (BGR) im Auftrag der Bundesgesellschaft für Endlagerung (BGE).

Hölting, B., & Coldewey, W. G. (2013). Hydrogeologie. In Hydrogeologie. Spektrum Akademischer Verlag. https://doi.org/10.1007/978-3-8274-2354-2

Zepp, H., König, C., Kranl, J., Becker, M., Werth, B., & Rathje, M. (2017). Implizite Berechnung der Grundwasserneubildung (RUBINFLUX) im instationären Grundwasserströmungsmodell SPRING. Eine neue Methodik für regionale, räumlich hochaufgelöste Anwendungen. Grundwasser, 22(2), 113–126. 
https://doi.org/10.1007/S00767-017-0354-3