Speaker
Description
The nowcasting community at DWD aims to optimally combine all available observational data for a better understanding of convection and severe weather, with a focus on remote sensing data. Traditionally, forecasters use radar and satellite imagery and overlays. Data fusion and plausibility checks thus solely happen in the human brain, based on experience. The human brain is expert in pattern recognition, but can only process a limited amount of information in parallel.
More recently, nowcasting data fusion algorithms at DWD combine multiple weather-related data sets numerically. For instance, satellite data, weather station data, lightning data and NWP model data are used to automatically categorize weather warnings based on thresholds of wind speed, precipitation rate, and other variables. New input variables are carefully checked before inclusion into the system. The selection of potential new variables is based on experience and meteorological knowledge.
So far, the potential of AI in data fusion has not been explored in our working group “Remote Sensing Application Development” at DWD. The general idea is to test if AI can detect so far unknown correlation patterns between the broad range of available input data sources (predictors) and higher-level weather information such as convective storm initiation or residual lifetimes of convective cells. Of special interest is to find out if current Meteosat Second Generation (MSG) data from SEVIRI are able to add benefits to the well-established radar data products at DWD.