About

The extreme precipitation events of autumn 2024 across Europe have once more underscored the urgent need to improve proactive measures and early warning to prevent and mitigate the impacts of precipitation-related natural hazards. In Austria, hazards such as landslides, flash floods, and hailstorms frequently cause harm to people, their belongings and infrastructure. The ongoing effects of climate and environmental change further emphasize the necessity of advancing prevention and adaptation strategies to enhance safety and community resilience. However, understanding and predicting the impacts of precipitation-related hazards at the national scale remains complex due to the complex relationships between meteorological drivers, associated biophysical and geomorphological processes, and socioeconomic factors. European national meteorological services, including GeoSphere Austria, are aligning with the United Nations Early Warnings for All initiative and the UN 2030 Agenda for Sustainable Development by undertaking a paradigm shift in their warning strategies. This shift moves from traditional weather warnings to impact-based warnings (IbW), which focus on the consequences of weather events (“what the weather will do”) rather than merely forecasting the weather itself (“what the weather will be”). However, beyond methodological and data challenges, skepticism among potential end-users regarding the implementation of IbW is also recognized, as they often perceive it as overly complex and opaque, which limits its adoption. From a technical perspective, machine learning (ML) has already demonstrated significant potential for improving predictions of complex phenomena over large areas by enabling the integration of diverse datasets across space and time. Yet, the lack of transparency in many ML applications, or their missing physical plausibility, often hinders their uptake in real-world decision-making, as end-users require interpretable, plausible, and trustworthy tools.

PRE4IMPACT-AT adopts a risk-oriented perspective and conceptualizes the impacts of landslides, flash floods and hail as a function of the interplay between atmospheric conditions, biophysical factors, and socioeconomic drivers. The project aims to create explainable and physically plausible spatiotemporal predictive models at the national scale for Austria, while also adopting a multi-hazard perspective. These impact-based predictive models will be evaluated for short-term applications, such as IbW, as well as for assessing trends and long-term patterns to better understand the evolution of critical conditions and the effects of climate change.

The methodological framework of PRE4IMPACT-AT comprises an initial formalization and prioritization of the manifold impact drivers through conceptual models, titled impact chains. Based on this, impact drivers are parameterized and harmonized using meteorological, biophysical/geomorphological and socioeconomic data. Numerous in-house as well as external resources, including meteorological data, nowcasting systems, reanalysis data, national damage databases, hail insurance data, and additional sources representing environmental conditions and exposure/vulnerability, will be utilized. Then, explainable ML will be applied to create spatiotemporal predictive rules for the respective impacts. Supervised classifiers and statistical learning will establish relationships between impacts and key static and dynamic drivers to predict, for example, impact probabilities or magnitudes under specific conditions. These models will undergo iterative improvements through quantitative procedures and enduser feedback obtained during repeated model evaluation workshops. The developed models will be tested in hindcast and nowcast contexts, while model-based trend analyses will reveal evolving spatiotemporal patterns in critical conditions. Finally, individual impact models will be integrated into a multi-hazard framework to advance multi-hazard impact assessment. Throughout all project phases, strong emphasis is placed on user engagement through co-design activities that also involve potential end-users holding warning mandates. This participatory approach ensures the developed models are iteratively improved to address skepticism and align with real-world needs, thus building a shared understanding between scientists and end-users.

Ultimately, the project will deliver scientifically robust and transparent models that provide a foundation for the transition toward IbW in Austria. It will also adopt a long-term perspective, offering valuable insights into long-term spatio-temporal trends in critical conditions. This effort is underpinned by an interdisciplinary collaboration involving GeoSphere Austria, the Technical University of Vienna, ETH Zürich, and the University of Graz. The collaboration of universities with GeoSphere Austria in its multifaceted role as a research institution, data provider, and national authority with a warning mandate ensures the research is firmly grounded in scientific and operational expertise.