**6. Discussion**

The coupled hydromechanical model of Lu et al. [5] was applied for slope stability assessment of the hillslope at the Dollendorfer Hardt (Bonn, Germany) with a relatively complex geometry and heterogeneity in material properties. The test site has a long history of investigations and slope analysis. Drilling, groundwater monitoring, laboratory tests, geophysical, and geomorphological surveys as well as a soil water content sensor network were used to design and parameterize the hydromechanical model. The soil water content obtained from the sensor network was successfully combined with the hydromechanical model to provide realistic initial conditions for the near-surface water content distribution. The ERT monitoring was only applied on a segmen<sup>t</sup> of the hillslope and was not used to define initial conditions for the simulations. As the ERT revealed very little changes in the water content distribution at greater depth even during summer time or heavy rainfall events, the sensor network covers all relevant shallow water dynamics. The results of the simulations show that both the lithological layers and the initial conditions play an important role in the redistribution of the pore water pressure and thus determine the position of the potentially unstable locations. The initially wetter conditions require significantly less rainfall than the drier initial conditions to reach potentially unstable conditions. Instabilities develop at locations that facilitate the accumulation of water due to the subsurface topography of less permeable layers. The obtained model allows us to study the influence of initial conditions and precipitation events on the slope stability.

For the test site itself, a significant mass movement seems unlikely as only potentially unstable locations close to the surface were identified in the model. However, the modeling results clearly identify the regions of most significant movement during long-lasting rainfall events at reasonable precipitation rates for the region in agreemen<sup>t</sup> with previous studies [38,40]. The hillslope is possibly moving in response to heavy precipitation but at a low rate of approximately 1–3 mm year<sup>−</sup><sup>1</sup> [40]. Modeling of more severe precipitation events would be required for a complete hazard assessment, also incorporating expected future changes in precipitation patterns due to climate change. Further, the model results indicate that the scarp is continuously unstable due to its high inclination. This is consistent with field observations of ruptures of small sizes that occur regularly. A slow extension of the scarp towards the top of the hill seems likely, potentially compromising hiking tracks.

In principle, the presented workflow for slope stability assessment can be extended towards a model-based early-warning system as almost all available data sources were incorporated into the model. So far, site-specific early warning systems are predominantly sensor-based with warning thresholds derived from conventional stability analysis e.g., [73–75]. Incorporating precipitation forecasts, a potentially critical state of a hillslope could be calculated based on the current water content distribution using the current state as initial conditions in the numerical model. However, computational demands are often the bottleneck to include near-real time complex numerical simulations into early warning systems. Therefore, it is challenging to calculate slope stability for a predicted rainfall based on the most recent soil water content measurements. A well-constrained model as presented in this work can be used to derive thresholds for implemented sensor networks or geophysical monitoring based on pre-calculated scenarios. Besides such a model-centered early-warning system, multiple data sources could be used for a more robust early-warning system by acknowledging the value of the in situ measurements itself. Data robustness could be improved through the combination of multiple different types of sensors [70]. In this work, soil water content sensors and ERT were combined to monitor the water content distribution in the subsurface. Through redundancy, erroneous measurements of single sensors or electrodes can be detected in order to reduce the risk of false alarms. Additionally, extensometers or GPS sensors could be used to capture

slope movement [75]. By combining precipitation forecasts, soil water content sensor networks and geophysical monitoring to capture water content dynamics and redistribution, and model-derived thresholds, it may be possible to establish a multi-level early warning system with high accuracy.

This study showed the rich possibilities that arise through the combination of various survey and monitoring methods with a hydromechanical model for slope stability. The complementary data sources allow for constraining the model in most aspects and reduce uncertainty in the model design. However, as seen in the comparison with the soil water content sensor network, there is a small scale heterogeneity that can not be resolved using geophysics as the contrast in geophysical parameters, e.g., resistivity or wave velocity, is too small. The high clay content in the soil made the application of ground penetrating radar, as another good methodology to detect near surface heterogeneity, impossible. On other hillslopes, this may be a useful addition to reveal near-surface heterogeneity. With additional near-surface soil sampling and extensive laboratory testing, it may be possible to describe the variability in parameters of the upper soil layers and to include the variability in the numerical model using geostatistical methods. For the studied test site, the assumption of homogeneous soil layers was considered to be adequate as slope topography and layering were the dominant factors controlling slope stability. The added value of ERT monitoring was not high in this study, but this method is expected to be more useful for slopes with stronger water content dynamics at greater depth. This would require an elaborate ERT setup with multiple electrodes and various electrode spacings. In more dynamic scenarios, material parameters and also soil layering may also change within an observation period. In the introduced framework, this could be included through an updated model.

In the future, an approach considering data assimilation seems desirable, especially in the context of a continuous monitoring of the test site. This should preferably be explored with a more sophisticated numerical model for slope stability analysis. In particular, a more advanced coupling of surface flow and infiltration into the soil would significantly improve the model. Field observations showed that strong precipitation events with more than 20 mm h−<sup>1</sup> only resulted in small increases in soil water content, since a large amount of the water ran off as surface water. Surface morphology supported this observation with visible runoff channels at the surface. Modeling surface runoff and ponding of water based on the surface morphology might alter infiltration dynamics along the slope during those rare heavy precipitation events. Daily water content dynamics along the hillslope are also altered by root-water uptake. While in the current study, net infiltration was considered, a more complex model for root-water uptake would change the water content distribution with depth and increase the heterogeneity of the water content distribution in the subsurface. If this process is adequately considered, the hydromechanical model could be operated with a daily or even higher resolution for the meteorological boundary conditions. In addition, differences between overgrown and bare parts of the slope could be incorporated into the model and possibly linked to the effective cohesion, as roots are known to contribute to the slope stability. With growing computational power, the full three-dimensional geometry of the hillslope and adjacent slopes could be modeled to eliminate simplifying assumptions and effects resulting from the geometrical reduction. An extension towards the calculation of plastic deformation would allow to determine slope movement and water dynamics for situations with a factor of safety larger than one, which not necessarily result in immediate rapid slope failure. Additional model verification could be achieved through measurements of slope movement.
