The value of Physics-Guided Deep Machine Learning for Precipitation Nowcasting relative to the state-of-the-art is examined through a hypothesis-driven study. A Physics-Data-Hybrid modeling framework is presented for bringing together extrapolation of radar echoes and multimodel superensembles from numerical weather prediction. Preliminary results from the hypothesis examination and modeling studies are evaluated based on statistical skill scores commonly used in machine learning and in weather forecasting. The value of the forecast improvements in the real world are examined by their end use in transportation network-of-networks. The latter is characterized based on tools from network science and engineering, following which the advance in systemic robustness and recovery, owing to possible improvements in nowcasting, is critically examined. Preliminary results suggest way forward and directions for further research.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner