Prediction and predictability of hurricanes with high-performance computers and cloud-resolving ensembles

This talk will be primarily devoted to the use of high-performance computing facilities to perform ensemble-based state estimation of hurricanes with cloud-resolving numerical weather prediction models. I will be sharing our recent experience in using approximately 30,000 cluster cores simultaneously at the Texas Advanced Computing Center which successfully assimilates high-resolution airborne Doppler radar observations in realtime and subsequently delivers 2 deterministic and 60-member ensemble forecasts running at 4.5/1.5-km effective horizontal grid spacings in a timely fashion. Since the predictability of hurricanes may be fundamentally limited by chaotic moist convection and subsequent upscale error growth, I would advocate that besides the need of~continuously~improving the hurricane forecast models and~ingesting high-resolution observations into the models to better initialize the storm, the hurricane state estimation is fundamentally probabilistic that demands cloud-solving ensemble-based data assimilation and forecasting. Improvements of forecast models may come from ever increasing computer power to better resolve the storms numerically and from improved~fundamental understanding of the dynamics and impact of subgrid-scale turbulence in hurricanes. Improvements of better state estimation may also come from development of new theories that are applicable for~high-dimensional, non-linear, non-Gaussian dynamic systems such as in hurricanes.