Operations and safety of increasingly automated and connected socio-technical systems are significantly influenced by weather and space-weather events. High-precision short-term forecasts (nowcasting) of system-critical events are essential to prevent operational disruptions.
In IN2ACTION (Insight, Information, Action), the physical understanding (insight) of significant weather and space weather will be processed in such a way that usable and reliable short-term forecasts (information) of small-scale phenomena are made possible, leading to direct recommendations for action. DLR has various methodological chains for nowcasting significant weather and space-weather events in the minute to hour range. A new generation of weather models, satellite data and developments in artificial intelligence promise great potential for improving nowcasting systems and allow information to be derived with unprecedented resolution and quality. In IN2ACTION, modern nowcasting products are being developed and deployed in the three methodological branches of space-weather, cloud and thunderstorm/wind nowcasting in mission-critical situations. To this end, five demonstrators are being set up in the programmatic of aviation, space, security, energy and transport, which highlight the added value of the information chains developed at DLR for application.
The Institute of Data Science is involved in two thematic focal points.
Firstly, a software component for data processing from heterogeneous data sources into a uniform, efficient data format is developed, which offers hierarchical organisation, comprehensive metadata, optimised storage and fast query functions. Furthermore, the processed data will be used in a visualisation component to create interactive representations of space weather phenomena that enable real-time mapping of solar-terrestrial interactions and support operational decision-making in the fields of aviation and shipping.
The second focus is on researching methods for recognising, understanding and reducing uncertainties in existing nowcasting models. The findings from the combined application of methods for uncertainty estimation and causal methods should complete the physical process understanding and show potentials for process-oriented model improvement.
