Thunderstorms or Severe Convective Storms (SCSs), respectively, pose a significant threat to property, agriculture and economy as well as to human life. Associated phenomena such as large hail, heavy precipitation, strong wind gusts or even tornadoes may cause substantial damage to buildings, cars, arable land and infrastructure, leading to billions of dollars disclosed by insurance companies every year. Possible short-term safety precautions can often not be conducted adequately because of too unsafe predictions and warnings issued by National Weather Services. Thus, detailed real-time knowledge about the spatiotemporal occurrence of potentially destructive convective cells (CCs) is desirable, but needs further intensive research in order to exploit the information available from both observational and model data in an optimal way.
Meteorological observations and Now-casting
Strong CCs over Germany and parts of its neighboring countries can be detected most comprehensively and with high quality by the radar network of DWD. Moreover, satellite remote sensing as well as an increasingly efficient lightning detection network provide additional observational data of CCs. Meteorological parameters characterizing favorable atmospheric environmental conditions for the onset and development of strong CCs, on the other hand, are computed operationally with the Numerical Weather Prediction (NWP) model cycles of the Weather Services. The role of these parameters and their interplay as an indicator for possible future convective development is still part of the current research, but several insights have been gained already in the recent decade. Convection-favoring ambient conditions can be estimated from NWP with a lead time of one to (maximum) three days in advance on the regional scale. The actual local occurrence of CCs or SCS, however, has to be estimated using nowcasting (NC) techniques; that is, a cell is localized not until it is detected by a weather radar, e.g., and subsequently its possible life-cycle and development is predicted by means of elaborated physical-mathematical procedures.
Many of these NC procedures extrapolate potential tracks of CCs using displacement vectors from the previous time steps – radar images are available in a 5 min interval. Meanwhile, different approaches and designs exist upgrading these schemes by integrating information from multiple observation networks (multi-sensor analysis) or by making use of simulated atmospheric environmental parameters. The ultimate goal of NC procedures is a precise cell track forecast including growth, intensification, attenuation, splitting and merging tendencies of CCs in real-time.
The current project veers toward an innovative NC scheme for DWD. The life cycle of potentially destructive SCSs (that is, the evolution of a cell from its birth to its dissipation) is statistically analyzed based on a sufficient sample of representative events. Historical radar detections of CCs over the last ten years are available. Subsequently, this life cycle information will be combined with simulated (predicted or analyzed) environment conditions from COSMO or ICON model of DWD. At the end, an on-line procedure will be developed performing a life-cycle analysis for every radar-detected and tracked CC. In doing so, the ‘state’ of a cell at a certain time instance is identified with respect to its whole life cycle. The life cycle analysis is then used to predict the possible evolution of the cell in consideration of both the statistically determined evolution and the observed life cycle history. Furthermore, the states of CCs could also be assimilated in NWP model cycles. Difficulties arising in the scheme described above will be the handling with interactions between spatially adjacent CCs (key words back-building, splitting and merging). Deficiencies of radar data and of the radar-based cell detection and tracking algorithm (KONRAD) have to be identified and accounted for in the NC scheme.
The mathematical methods, which shall be applied to this problem in the future, reach from Principal Component Analysis over Multivariate Statistical Procedures (e.g., regression methods) to Inverse Modeling and Machine Learning (Neural Networks). Software will be drafted and developed modularly in order to integrate it into existing environments at DWD. Several verification studies are planned, once the scheme has reached a sophisticated status. Ultimately, the warning management process at DWD could take benefit by integrating the NC scheme into its operational cycle.