Introduction

The Gebäudeversicherung Bern (GVB) experiences annual losses from natural hazard of approximately 60 million CHF, primarily from windstorms, hail, and flooding. A comprehensive risk analysis for 2023 and 2024 was conducted to address this, focusing on the risk: hazard x damage potential x vulnerability. While GVB has a solid understanding of hazards and damage potential, a knowledge gap remains in assessing vulnerability. This project aims to reduce this gap by developing a systematic approach to evaluate and mitigate vulnerability.

Methodology

To address the knowledge gap, hazard overviews and vulnerability indicators were developed for hail, windstorms, and flooding. This was achieved through literature reviews, expert interviews, and existing vulnerability indicators. A physical vulnerability index was created based on these sources to quantify the vulnerability of buildings.

Additionally, a catalog of behavioral guidelines was created for property owners based on their vulnerability index results. Risk was calculated using spatial data from hazard maps for hail, wind, and flooding.

A clickable prototype was developed and tested through 20 customer interviews to evaluate its effectiveness. Feedback was integrated into an updated prototype, which was further tested for usability.

Results

Sixteen key indicators were identified for assessing building vulnerability. These included:

1. Number of above-ground floors
2. Number of underground floors
3. Residential use of the basement
4. Presence of raised ground floor
5. Entry doors at ground level
6. Windows and doors in the basement
7. Value accumulation in the basement
8. Roof material
9. Solar installations
10. Roof overhangs
11. External shading systems
12. Number of motorized sunshades
13. Year of construction
14. Pool covers
15. Skylights and roof domes
16. Presence of backflow valve

Responses were categorized into vulnerability scores from 1 (low vulnerability) to 5 (high vulnerability). The unweighted sum of these indicators produced a score from 6 to 50, which classified properties into three risk categories. Behavioral recommendations were then provided based on the score.

For example, if a building had plastic skylights, the recommendation was to install protective grids and replace damaged parts. These actions provide both hail protection and fall-through prevention, and GVB offers financial support for these measures.

The prototype received an average rating of 8/10 from users, though feedback indicated that some found it difficult to distinguish between hazard and risk, prompting a greater focus on risk in future iterations. Based on the positive feedback, the tool will be implemented on the GVB customer platform by the end of 2024.

Conclusion

The vulnerability indicators developed in this project provide a solid foundation for increasing homeowner awareness of natural hazards. The customer interviews were invaluable in improving the prototype’s usability and gaining insights into user behavior. Future plans include extending this methodology to additional natural hazards such as landslides, debris flows, avalanches, and rockfalls to further enhance risk mitigation.

The Swiss Government stresses with its decision to implement cell broadcasting the importance of early-warnings in Switzerland, which is in line with global developments to better prepare for natural hazards. Studies show that effective early warnings are impact-based and personalized. To tailor warnings, personal data is needed which is not openly available, i.e. on caring duties or medical issues, but would have to be shared with the provider of the warnings With a representative survey, we explored i) the interest of the Swiss public for tailored (heat) warnings, ii) the impact of tailored warnings to take protective action and iii) the willingness to share the necessary data. Our results show that the Swiss public would appreciate tailored warnings but is not willing to share the necessary data, which is an example of the privacy paradox. To make early warnings more effective, developers would have to find ways to overcome this paradox. At the RIMMA 2025, we would like to discuss these results and share potential policy impacts of our findings.

As the federal agency responsible, the Federal Office for the Environment (FOEN) coordinates, informs and warns the population about the danger of forest fires. Since 2022, a new system has been in place in collaboration with the cantons. IGNIS (from the Latin ‘fire’) is an information system developed by the FOEN in collaboration with Natural Resources Canada (NRCan). IGNIS is based on the Canadian Fire Weather Index (FWI), developed in Canada in the 1970s, which has been adapted and optimized for the Swiss Alpine region. It consists of six components that account for the effects of fuel moisture and weather conditions on fire behavior. Every day, MeteoSwiss provides the FOEN with measurements from its stations and weather forecasts. These data are subsequently processed and made available to federal and cantonal experts. These two authorities are responsible for the daily assessment of forest fire danger and the publication of danger levels for the different alert regions. In addition to its prevention purpose, IGNIS enables authorities to take preventive measures quickly and speed up decision-making processes for the allocation of fire-fighting resources. Furthermore, the FWI system is widely used in the EU and worldwide, which facilitates the exchange of knowledge with international experts. IGNIS is an integral part of the système d'alerte modulaire (SAM). SAM is a data management system, which ensures the flow and processing of data, the execution of tasks, the smooth running of operational flows and manages the distribution of products to external parties. As IGNIS is part of SAM, it is possible to inform and alert the population about forest fires quickly and safely.

Climate services (CS) provide climate information that supports decision-making, adaptation, and resilience in a changing climate. These services are tailored to meet the specific needs of end-users, including the information itself, frequency, and dissemination channels. CS offer benefits across various sectors such as agriculture, water resources, energy, health, and disaster risk management.

For these services to be effective, they must be co-produced between the relevant actors, National Meteorological and Hydrological Services (NMHS) and end-users throughout the entire process, from the initial acquisition of information to its delivery. This facilitates the establishment of trust by actively engaging end-users. This collaborative approach reduces the use of a top-down approach, where institutions create and deliver services without consideration of the end-user, in addition to promoting ownership and engagement among stakeholders. Despite its increasing prevalence, the concept of co-production undefined in the field. In contrast to conventional, top-down methodologies, knowledge is collectively constructed by stakeholders, thereby ensuring that CS are tailored to the specific requirements and context.

Studies that assess the socioeconomic benefits of CS are of great importance, as they determine their value by quantifying tangible and intangible benefits. Such studies help justify investments in NMHS and deepen the comprehension of how to enhance CS to serve user needs.

The ENANDES “Enhancing Adaptive Capacity of Andean Communities through CS” and ENANDES+ “Building Regional Adaptive Capacity and Resilience to Climate Variability and Change in Vulnerable Sectors in the Andes” projects have the objective to enhance resilience through the provision of CS and entails a collaborative effort between institutions. The participating countries are Argentina, with a NMHS and a Regional Formation Center (CRF), NMHS Bolivia, NMHS Chile, NMHS and Regional Climate Center (CRC) Colombia, NMHS and CRC Ecuador, NMHS and CRF Peru. In addition, there is a Regional Expertise Hub (NUREX), a virtual platform for sharing information created for the project and managed by Peru. Moreover, other institutions are involved, including NMHS Switzerland, Bern University of Applied Sciences (BFH), International Research Centre on El Niño (CIIFEN), and World Meteorological Organization (WMO).

Co-production enables integration of perspectives and expertise, a crucial aspect for addressing the heterogeneous climate challenges. This facilitates the formation of a collective ownership of the project, enhancing its effectiveness and sustainability beyond its duration. However, organizing co-production across different institutions, interfaces, and levels presents several challenges. It requires establishing transparent communication, and shared objectives. For this, a standardized approach for co-production must be defined.

Co-production should be seen as a continuous process, crucial for building trust among actors and ensuring that the information remains relevant and used by end-users. The ENANDES and ENANDES+ projects serve as a fertile environment to define a framework for co-production in CS. As climate challenges intensify, adopting a co-production approach becomes fundamental to deliver CS benefits of supporting adaptation, resilience, and decision-making, in addition to strengthening regional cooperation.

The Rhineland-Palatinate Flood Warning Service has been issuing flood warnings in Rhineland-Palatinate since 1986. Initially, the warnings were issued based on measured values at the gauging stations of the major rivers. Over time, the Flood Forecasting Centre's products have been expanded to include forecasts at water gauges, including ensemble forecasts and region-specific flood warnings in small catchments. In addition, flood reports are published on a modernised website and via warning apps (Johst & Demuth 2022). The floods on the River Ahr in 2021, in particular, once again demonstrated the importance of flood forecasting and the timely issuing and disseminating of warnings. This contribution aims to present the current status and future developments in flood forecasting and the issuing and disseminating flood warnings for Rhineland-Palatinate.

The Flood Forecasting Centre uses the deterministic-conceptual water balance model LARSIM (LEG 2024) for operational flood forecasting. However, the forecasts produced are always subject to a certain degree of uncertainty. To visualize this uncertainty, ensemble forecasts of the weather forecasts (ICON-D2-EPS of the DWD) are used as model input for a forecast period of 48 hours. A discharge forecast is thus produced for each of the 20 ensemble members. On the Flood Forecasting Service website, the result of the ensemble calculation is displayed as a bandwidth around the median in a hydrograph. The darker the colouring within the band, the more likely it is that the expected values will lie within this range (Johst & Demuth, 2022). In addition, the thresholds of certain flood return periods (2-, 10-, 20-, 50- and 100-year return periods) are also shown in the graph for the respective water level for information purposes. The warning map on the Flood Forecasting Centre's website visually shows the flood risk for the small rivers in individual warning regions in relation to the next 24 hours.

The warning regions are coloured according to the flood return period from low to extreme (Johst & Demuth 2022). The flood reports produced in the event of flooding are sent by e-mail to subordinate authorities that coordinate disaster control on-site, as well as to the radio and press. In addition, the flood reports and flood warnings from the region-specific warnings are sent to the Länderübergreifendes Hochwasserportal (LHP) and the warning apps Meine Pegel, NINA and KATWARN (Johst & Demuth 2022). This also ensures that the public is informed at an early stage.

As part of a ‘co-design’ project realised together with the German Weather Service, the user-oriented communication of weather and flood information is to be improved. The project aims to provide users with a better understanding of the uncertainties associated with the forecasts and the probability information provided, thereby creating an improved basis for decision-making for future flood events.

Extremes of river flow such as fluvial floods and hydrological droughts are expected to be influenced by human-induced climate change. To design effective adaptation measures, knowledge on projected changes in streamflow along river networks is crucial. Global climate models (GCMs) like those contributing to the CMIP6 archive offer future projections, however, they typically only provide runoff at the grid cell level rather than routed river discharge. One way to overcome this limitation is to combine GCMs with global hydrology models (GHMs) which are designed to close the terrestrial water balance. However, this approach can only consider a limited number of GCM projections. As a result, estimating the robustness of projected changes in river flow is challenging as both uncertainties associated with GCMs as well as effects of internal climate variability may be underestimated.

To bridge this gap, we route 250 years of daily runoff from 21 CMIP6 models along the global river network at a horizontal resolution of 0.1 degrees. Specifically, runoff from the historical CMIP6 experiment as well as future Shared Socioeconomic Pathways (SSP) is used to construct a new dataset of global daily routed streamflow. For the routing step, the multiscale Routing Model (mRM) is used which can flexibly be adapted to a wide range of spatial scales. In mRM, gridded daily runoff data is routed through an upscaled river network derived from high-resolution morphological data based on a kinematic wave equation. The fidelity of the proposed modelling chain is carefully evaluated with special focus on the underlying assumptions and the scale mismatch between spatial resolution of the GCMs and routing model. To this end, simulated river discharge climatologies are compared with observations. Leveraging the new global streamflow projections, we study how anthropogenic climate change has affected mean and extreme river flows. In addition, we harness the comprehensiveness of the newly created streamflow projections and compare routed runoff across all available CMIP6 models to espread ad agreement between models is evaluatexplore the range of future streamflow projections and their robustness.

Geological Hazards have always caused and still creates a threat to the important part of the population, also causes damage/destroy of existing infrastructure facilities. In the last decades, protection of the population from debrisflow hazards and safe operation of infrastructure objects became significant social-economic and geoecological problem for the most countries in the word. These problems are more in countries with the complicated geology, relief, climate, seismicity, human activities (Gaprindashvili M. et. al 2021; Fourth National Communication of Georgia 2021).

Georgia belongs to the most complicated region among the world’s mountainous countries with development scale of debrisflow hazard, recurrence of these processes, and with negative impacts to the population, infrastructural objects, agricultural lands and environment, and the most tragically, it causes human casualties. Hundreds of settlements, infrastructure objects are periodically affected by debrisflow disaster (Tsereteli E. et. al 2022). Annually the cases of debrisflow processes increase significantly. During the period of 2011-2023 the activity of debrisflows has been recorded in 2973 river gorges/ravines throughout the country.

Activation of debrisflow hazards and their risk depends on the geological environment, such as lithological composition of the rocks of the territory, their physical and mechanical properties and energy potential of terrain, as it is known that development of hazards at high hypsometric levels is faster and more intense. Accordingly, the magnitudes of the slope of the relief surface and the erosive intrusion depths of the rivers/gorges, the coefficients of horizontal divisions and, most importantly, the tensions of the gravity fields increase, together with process-driven factors such as meteorological events, seismicity and anthropogenic pressure (Gaprindashvili M. 2022).

Different methods are used for debrisflow hazard asssessment, which is based on the data availability: Qualitative, Quantitave, Rapid Mass Movement Simulation/Modelling, Spatial Multi Criteria decision-making (SMCE) et al. In Georgia different researches were conducted for the purpose of debrsiflow Hazard Assessment (Gaprindashvili G., et al. 2018, Gaprindashvili M. 2022).

References

Fourth National Communication of Georgia Under the United Nations Framework Convention on Climate Change. Chapter 4.9 Geological Hazards in Georgia, Tbilisi, 2021, pp. 278-286

Gaprindashvili G., Tsereteli E., Gaprindashvili M. Landslide hazard assessment methodology in Georgia. // Special Issue: XVI DECGE Proceedings of the 16th Danube ‐ European Conference on Geotechnical Engineering, 2018, vol. 2, N 2-3, pp. 217-222.

Gaprindashvili M., Kurtsikidze O., Gaprindashvili G., Rikadze Z., et. al, Informational Bulletin - The results of the development of natural geological processes in Georgia in 2023 and the forecast for 2024, Department of Geology, National Environmental Agency, Tbilisi, Georgia, 2024, 488 pages

Gaprindashvili M., Assessment of geo-ecological hazards caused by geodynamic processes on the territory of Tbilisi and their prevention. PhD dissertation, 2022, 168 pages.

Gaprindashvili, M., Tsereteli, E., Gaprindashvili, G., Kurtsikidze, O. (2021). Landslide and Mudflow Hazard Assessment in Georgia. In: Bonali, F.L., Pasquaré Mariotto, F., Tsereteli, N. (eds) Building Knowledge for Geohazard Assessment and Management in the Caucasus and other Orogenic Regions. NATO Science for Peace and Security Series C: Environmental Security. Springer, Dordrecht. https://doi.org/10.1007/978-94-024-2046-3_14

A sophisticated geoportal has been developed for EG.D, a company focused on the distribution of electricity and gas in the southern region of the Czech Republic. This portal provides hourly meteorological information and various warnings, covering not only the present but also the past (several days back) and the future (a few days ahead). Users can access map layers with a resolution of 0.5 km. The system also automatically sends warnings via SMS and email based on user-defined criteria.

The forecasts include essential meteorological elements crucial for the energy sector, such as air temperature and humidity, wind speed and gusts, global radiation, and snow water equivalent. Additionally, the system provides warnings for weather events that could disrupt the distribution and transmission network, such as strong winds, thunderstorms, frost, or heavy snow. These warnings are categorized by severity in relation to the needs of energy distribution.

The geoportal also offers centralized access to information from radars, lightning detection systems (for storm identification), and satellite images. For frost prediction, data from EG.D’s measuring stations are integrated into the system. Artificial intelligence is used daily for calibration and optimization.

Data is updated hourly based on the latest measurements from the Czech Hydrometeorological Institute (ČHMÚ). The forecasts are generated by combining several models suitable for the area, such as ICON, IFS, ARPEGE, and WRF, tuned to our specific conditions. The transition between station measurements and model outputs is managed hourly through nowcasting, refining the forecasts for the upcoming hours. All calculations are performed for every pixel of the background maps, ensuring precise spatial nowcasting.

Based on user requirements, a data warehouse has also been created on the MySQL platform. This warehouse enables data analysis according to specific criteria, such as individual points, area aggregation, or weighted averages considering population density.

The use of process-based numerical models for the simulation of mass movements is an important component of modern integral natural hazard management. When informed by high quality data and expert interpretation, numerical simulations aid process understanding, scenarios building as well as the design of mitigation measures. Despite a relatively long history and tradition in model development and application, problems still exist both in the expert’s community and at the interface between policy makers and the public. Understanding the domain in which the model and its results can be validated, the sources and magnitude of errors and uncertainty or the physical and numerical approximations needed are only some examples of problematic topics.

A spectrum where people swing between systematic scepticism versus over-trust in the models and in their potential exist both amongst experts and the public. This paradigm has been exacerbated in the past few decades, with the fast development of computing power and numerical methods, which has led to an imbalance in what is technically possible and what the community can handle. A large gap appears in front of us with regards to the application of three-dimensional multi-physics models to engineering and geological problems. In this abstract, the authors propose to discuss one aspect of this impending challenge: results visualization and communication with stakeholders.

In particular, we use the case of the Material Point Method (MPM), which enables the simulation of elasto-plastic constitutive relations integrated within a hybrid Eulerian/Lagrangian numerical scheme. The method, initially developed for civil engineering problems involving large deformations, later gained widespread recognition in the movie industry, in particular after its use in Disney movie Frozen. However, its application to real-world physical problems soon revealed its significant research and engineering potential. In the years that followed, numerous highlighted its effectiveness in modelling snow, ice, rock avalanches and multi-phase process cascades at various scales (Cicoira et al., 2022).

The application of MPM to alpine mass movements is emblematic in many ways, highlighting the complexities at both ends of the spectrum. The scepticism and the overconfidence in the model are both possibly driven by the animated movies and complex renderings that can be created from the results. On the one hand, the association with the movie industry misleadingly leads some to view the results as mere illustrations, lacking in scientific validity. On the other hand, the realistic representation of the results enhances the understanding of the process and builds trust in the model. Interestingly, in both scenarios, the visual graphics overshadow the model itself and its results.

With this abstract, we want to discuss some strategies and address open challenges in the communication of complex numerical simulations by means of cartography, three dimensional renderings and animations. With the anticipated rapid growth of such visual tools in the coming years, it is crucial to anchor these advancements in strong fundamental principles of research, engineering, and science communication.

The landslide of Cadegliano Viconago in the province of Varese (IT) has been for over twenty years a danger to the population and infrastructures located along the course of the Tresa river, at the border between Italy and Switzerland. The landslide leads to the accumulation of debris which has caused several events, involving the road passing at the foot of the landslide. The slope even shows traces of deep damage which manifests with the exposure of scarps and traction fractures, with even centimeter displacements during intense and continuous rainfall events. A catastrophic collapse would lead to the partial or total interruption of the road system, a barrage of the Tresa river with the possible formation of a basin and the destruction of wells field for drinking water use in Switzerland.

Even though in recent years of monitoring it was observed a trend towards smaller displacements, intense and prolonged rains could drastically influence the dynamic behavior of the landslide and cause bigger displacements that could even lead to a catastrophic failure.

For this reason, thanks to the financing of an Interreg project with the involvement of the Interregional Agency for the Po River and the Canton Ticino Land Department, two DMS (Differential Monitoring of Stability) columns (Lovisolo,et al. 2003) were installed in March 2021 to investigate the landslide activity and to define alert thresholds. DMS multiparametrical columns provide continuous data with detailed displacement, accelerometric, and piezometric measurements. The deep monitoring data joined by spatially distributed geodetic measurements since 2006, provide information on the landslide’s dynamics. The monitoring, the cores from the drilling, and the in-situ surveys provide a basis to set a numerical stability model to understand the slope kinematics and the response to the water table changes.

The displacement data ranged between 1 cm/year in the first three years, up to 1 cm/week in an intense event in late March 2024 connected to a remarkable lifting of the water table. Similarly, intense rain events recorded in autumn 2002 and 2014, led to small-scale debris flows, partial destruction of the road and development of decimeters-width tensile fractures, making a catastrophic collapse plausible.

The definition of the humidity status identified by the Standard Precipitation Index (SPI), showed a very high value in correspondence of the landslide maximum activity, which makes it a possible factor for setting alert status.

The observation of recorded displacement by DMS and the stratigraphic data, suggests a slip surface at a depth of about thirty meters, which can plausibly lead to the detachment of a volume mass ranging from 70’000 to over 2’000’000 m³.

The numerical stability model was followed by expansion models of the unstable mass based on the observation of previous collapses. The modelled scenarios remarked the hazard for the infrastructures in both Italian and Swiss territories.

The World Meteorological Organization strongly encourages the development and use of impact-based early warning systems (IBEWS), along with recent advancements in research, e.g., Merz et al. (2020). As Potter et al. (2021) note, IBEWS improve public understanding of warnings, enhance interagency communication, and reduce "false alarms". This
the study aims to quantify the added value of an IBEWS specifically designed for flash flood anticipation by focusing on detecting impacted sectors.
In France, local flood forecasting services assess flood risk on major rivers, coordinated by the Central Service of Hydrometeorology and Flood Forecasting (SCHAPI), which provides a national risk map with four levels (green, yellow, orange, red) that incorporate both hazard intensity and potential impacts. However, small ungauged rivers are not covered by this service. An automated warning system based solely on the flood return period exceedance has been implemented for municipalities along these rivers.

We extended this hazard-based system to an impact-based approach for ≃20,000 km of river network in the French Mediterranean area, which is particularly vulnerable to flash floods (Gaume et al., 2016). The model structure aligns with existing European methods (Dottori et al., 2017), using hydrological and hydraulics approaches developed in France
(Piotte et al., 2020, Davy et al., 2017). The real added value of this work lies in the quality and diversity of observed impact data sources for evaluating the benefits of impact-based versus hazard-based warnings and the validation methods employed. [...]

In the impact model literature, predictive models often focus on a country or a region. This is the case of the 510 model (Teklesadik et al., 2023) or the grid-based model by Kooshki Forooshani et al. (2024), both statistical models for predicting impact on buildings in the Philippines or the Hybrid model (Hou et al., 2020) which also predicts impact on buildings in China. However, resources exist at the global level, including the Advanced Disaster Analysis and Mapping (ADAM) system which provides estimates of exposed population to different windspeed and accumulated rainfall ranges (World Food Programme, 2016) at the administrative 2 level or the DisasterAWARE platform (Sharma et al., 2020) by the Pacific Disaster Center which estimates not only population exposed but also infrastructure (Hospitals, schools, etc) and capital exposed at a grid level (30m×30m tiles). In this work, we propose a statistical model (Extreme Gradient Boosting or XGBoost) that predicts the affected population in cases of Tropical Cyclones (TC) at a grid level (1 km2 tile) in 60 countries. This model not only considers weather-related variables like wind speed and accumulated rainfall but also considers country-specific features such as topographical measures and poverty indexes. Furthermore, by employing XGBoost, we can handle non-linear relationships between variables.

Addressing data scarcity in the Global South is increasingly critical as climate change is expected to result in more frequent and severe flood events in highly vulnerable regions. A novel risk assessment framework tailored to these challenges is demonstrated on a use case based in Angwan Iku, a flood prone community in Gwagwalada, Nigeria. The community experienced a high magnitude flood event in 2020, which sustained adverse consequences for built structures, local residents, and their livelihoods. Key inputs include local context provided through community engagement, high resolution data acquired conducting household-level field surveys, and freely available medium resolution (30 m) topography data.

The study first identifies key local flood risk drivers, including flood mechanisms and building vulnerability classes. A multivariate analysis is performed on field data to identify building features that contribute most to flood damage. Flood risk zones are further delineated by performing contour analysis on identified flood threshold levels on local river banks. These zones are then validated against a 2020 flood extent map derived from field surveys, which served as a point of reference or ground truth. Finally, adaptation strategies are proposed for at-risk buildings, informed by the risk zones and building vulnerability classes.

The approaches within this framework can generate custom, baseline flood risk profiles for data-scarce communities, which can also be further refined and validated over time. Integration of community insights ensures that local risk factors are accurately captured and that proposed adaptation measures directly address community defined priorities and needs and are sustainable. Moreover, the methods make use of data that can be acquired for any data-scarce region, and are therefore transferrable to comparable communities and scalable with available resources. Most notably, we demonstrate how the framework extends beyond working with existing data gaps, towards the translation of evidence-based insights into actionable solutions specific to the needs of each community, so that flood risk is effectively minimised on the ground.

Groundwater is among the most important sources of freshwater for drinking water production and irrigation purposes across Switzerland. Recent drought events have shown that these groundwater resources, traditionally considered a safe source of water supply, are affected by prolonged periods with limited or no rainfall and high evaporation. Early warning systems may help inform groundwater resource managers about the possible future state of groundwater levels in different aquifers across Switzerland. Such a system could provide crucial support by enabling timely responses to mitigate the potential consequences of droughts on groundwater resources.

This study aims to investigate the feasibility of forecasting groundwater levels in Switzerland up to 32 days into the future using sub-seasonal meteorological forecasts provided by MeteoSwiss. This study employs lumped-parameter models from the Pastas software (Collenteur et al., 2019) to simulate groundwater levels at seven strategically selected monitoring wells across Switzerland. These models are driven by ensembles of daily meteorological forecasts to generate ensemble daily groundwater level predictions. A noise model is applied to enhance forecast accuracy, a step that was found critical to improving forecast quality. The forecast quality is verified using weekly re-forecasts from 2012-2023 and compared against two naive benchmarks: persistence and climatology-based forecasts. The results indicate that skilful forecasts of groundwater levels can be made for the entire forecast period. Whether the ensemble groundwater predictions outperform, the naive forecasts are related to the response time of the groundwater system.