Seamless Probabilistic Nowcasting and Short-Range Forecasting for High-Impact Weather Events

This abstract presents a new forecasting system aimed at providing seamless and calibrated predictions of high-impact weather events, tailored to the needs of hydrological services, forecasters, and civil protection agencies for effective early-warnings. Focused on thunderstorms, flash floods,  extreme temperatures and gusts, the system aims to create a probabilistic nowcasting and short-range forecasting. In order to capture local events such as thunderstorms or wind gusts, the gridded forecasts have a spatial resolution of 500m and a temporal resolution up to 5 min and tailored (scaled) to weather warnings using all the available information at Geosphere Austria. The first step is creating a probabilistic analysis using data-driven approaches and introducing uncertainty due to data density and external factors such as altitude, orographic factors for temperature or radar beam for spatial distribution of precipitation in high resolution. Afterwards, a radar-based nowcasting technique is employed, leveraging data-driven approaches to account for uncertainty due to external factors such growth and decay for precipitation. Machine learning allows to merge probabilistic nowcasting with a collection of NWP models, ensuring a robust prediction framework by using the probabilistic analysis.  This merging of various sources of information facilitates seamless transitions between nowcasting and NWP-based forecasts with the goal to improve the reliability of early warnings. By providing probabilistic forecasts tailored to specific weather warnings, the system is flexible enough to provide specific weather information to different end-users. The information tailored to specific tasks (such as probability of exceedance) of the final probabilistic method is verified against the individual components of the model chain to guarantee the added value of the system in helping the decision making in high-impact weather situations.