Internship Colloquium WOUTER COOMANS

Piping alongside the Pannerdens Kanaal; groundwater modeling with MUDFLOW (GWV)​

A large amount of the world population is living in areas prone to floods, and this amount is expected to increase even further in the future. There are many types of floods, which are potentially all very devastating, but this research concerns river floods caused by the fail mechanism piping. This fail mechanism can occur during periods of high water and causes dikes to become undermined, weaken them, and may even cause them to collapse. Even though there is a lot of knowledge already available about piping in the Netherlands, there are still some uncertainties that require additional insight. One of these uncertainties concerns the resistance of the top layer of the flood plains and its effect during a high water event. A previous study with the regional AMIGO model failed to realistically capture available measurement data, due to the scale of the model and being on the edge of the model area.

For this reason, three smaller sub-models are made in MODFLOW (GWV) with a finer spatial scale for the locations Westervoort, Pannerden and Lobith alongside the Pannerdens Kanaal. Data is available from extensive soil research and drillings, but also from sources such as REGIS. Using the combination of this data allows for adaptations to be made to a great amount of model assumptions and parameters to improve model performance. Due to a lack of a high water event during the available measurement period, the focus is to get the model fit as good as possible, to be able to use it when an actual high water event is occurring or expected.

All three sub-models in MODFLOW showed significant improvements compared to the use of the AMIGO model. The best results were provided by the Westervoort model, with a RMSE of 0.1068 and 0.2507 for respectively the calibration and validation. There was however a very slight underestimation of the peaks during high water events. This means it has to be taken into account when looking at piping, because it would subsequently also underestimate the risk of piping.
The model for the location at Pannerden provided the worst results out of the 3 sub-models. This was partly caused by work being done on a secondary channel next to the Pannerdens Kanaal. There were also a lot of broken monitoring wells at this location that need to be replaced. The calibration on a part of the year 2016 showed potential (RMSE=0.1819, NSE = 0.9068), but it was not possible yet to validate the model due to measurement data that was not yet made available.
For the model location at Lobith the calibration run was reasonably good, with a RMSE of 0.1834 and a NSE of 0.8741. The validation however, showed a big decrease in model performance (RMSE = 0.4332, NSE = 0.613). Overall, the modelling in MODFLOW proved to be very data hungry compared to 2d-modelling. It does provide a lot more insight in spatial variability, because it can take old river beds into account in the subsoil. These can easily be missed by a 2d-model.

We recommend to first replace broken monitoring wells, and improve others. The measurements should be continued until a high water event occurs to test the model performance. If these prove to be valuable, we recommend to explore the option of using an USG-model which allows further refining of model grids. Finally, the location at Pannerden needs further intensive investigation concerning the placement and depth of monitoring wells, and a potential influence of regional groundwater flows.