Generating automatic warning proposals for forecasters at MeteoSwiss

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. In this contribution, we will focus on the generation of automatic warning proposal for the forecasters.

The automatic warning proposal algorithm is developed in close collaboration with forecasters that provide the spatio-temporal constraints the algorithm uses to summarize noisy grid-point-level severe weather information into smooth large-scale warning proposals. The first warning parameter we work on is long-lasting heavy rainfall. For this parameter, the algorithm first assigns warning levels valid at grid points based on threshold exceedances of percentiles of cumulative precipitation. Subsequently, individual potential events are identified by grouping together lead times in which substantial threshold exceedances occur. For each potential event at every grid point, the maximum precipitation accumulation, its associated warning level, and the time it occurs are determined. On this collapsed event-specific warning-level data set, a series of spatial constraints are applied to make the data smoother. Close-by features of the same warning level are merged, small-scale features are removed, and the resulting features are mapped to MeteoSwiss’ warning regions. Afterwards, each feature is assigned a start and an end time with the time of maximum exceedance determining the end time and the precipitation accumulation period the start time of the feature. Additionally, the features are split into further sub-features if the associated grid-point-level end times vary strongly across the feature and the feature is large enough to warrant an additional split.

We will illustrate our automatic warning proposal algorithm for heavy rainfall with case studies while paying special attention to the constraints the forecaster provided for the algorithm and to which degree they can be met.