Conference Agenda

At MeteoSwiss, most hazard warnings are currently verified manually by a team of forecasters. While this approach has proven to be very successful and makes it possible to incorporate the extensive knowledge of the forecasters into the warning verification, it also has its limitations. Manual verification is labor-intensive and somewhat subjective, which can lead to inconsistencies in the assessment of warnings. Furthermore, manual verification only allows for limited granularity in the verification process and long-term statistics, making it difficult to identify systematic biases or potential areas of improvement in the warning system.

To address these challenges, we are developing an objective and automated verification system to improve the efficiency, consistency, and detail of the warning evaluation. Such an approach will lead to a more standardized verification of warnings, reduce human bias, capture a richer set of observation data, and be able to generate a more extensive set of key figures and scores. Automated verification also enables a more straightforward evaluation of past warnings and, therefore, the possibility to identify trends, patterns, and potential biases that might have been overlooked with case-by-case manual verification. In this regard, an automatic verification system will also be beneficial for developing and testing automated warning proposals.

This presentation will share preliminary results from our ongoing work, including verification results from case studies and a comparison to the traditional manual approach. Moreover, we will present a statistical analysis based on the objective verification of past warnings and some first-identified shortcomings of the warning system at MeteoSwiss. We will also highlight some of the key challenges we have encountered so far.

Our goal is to demonstrate the potential benefits of an objective warning verification system, particularly in terms of its ability to gain deeper insights into the quality of issued hazard warnings and to drive the continuous improvement of the warning system at MeteoSwiss.

The Snow Avalanche Forecast Editor SAFE was introduced in the Swiss avalanche warning system in December 2023. This is based on European standards, but for the first time also permits the integration of machine-learning models by including the model data as an additional virtual forecaster in the decision-making process. It is available as open source to all interested parties.

Severe weather poses significant threats to society, necessitating the development of effective forecasting and warning systems to mitigate their impacts. In the framework of the renewal of the warning system at MeteoSwiss, we use the possibility to redesign the software as well as the scientific approach of generating warnings. The production chain is designed such that the data is guided and refined along the pathway. Here, we present the operational production process currently in development for the next generation warning system. The production chain can be divided into four steps. The first step is the combination of all available data into one data stream (seamless weather). The second step identifies extreme weather events in this data stream and prepares warning proposals for the forecasters. The third step consists of the forecasters evaluating the proposals, changing them if necessary, and issuing the official warnings. In this step, the only human interaction with the otherwise automatic warning system takes place. In the fourth step, the warning products are customized and distributed to our customers. In the fifth and last step the warnings are verified automatically to monitor the quality of the system.

MeteoSwiss furthermore aims to integrate impact-based forecasts into warnings and the warning chain introduced above. The premise of reducing people’s harm by introducing possible impacts in extreme weather warnings is a paradigm change for many national weather services (Geiger et al. 2024). Being at the intersection of probabilistic forecasts, uncertainty analysis and state-of-the art risk modelling, this new paradigm also entails significant challenges. Through collaboration with stakeholders including first responders and local authorities, MeteoSwiss draws on pilot regions to find specific use cases from professionals impacted by severe weather to learn about the specific needs and enhancement potentials for early warning. Co-developing these use cases, conducting impact studies, and eventually providing stable products serves as a learning for further development and scaling up of impact-based warnings.

This requires an operational impact model for Switzerland, which flows back into the warning production chain, posing various scientific and technical challenges on the way; we will discuss a solution developed at MeteoSwiss, and highlight the need for seamless conceptual and technical integration within the current warning production processes, with the ultimate aim to better cater to the diverse needs of society in the face of severe weather events while maintaining a reliable warning chain.