Petascale Supercell Simulations

Leigh G. Orf, University of Wisconsin-Madison

Usage Details

Tornadoes continue to kill people every year across the world, partly due to our inability to forecast thunderstorm behavior accurately. A major challenge to thunderstorm modeling work is the amount of computing resources required to run at very high resolution (doubling resolution requires 16 times more resources assuming perfect scaling). Such high resolution is required in order to resolve features thought to be important in tornado genesis, maintenance, and decay, and is required to adequately resolve the multiple vortex tornado.

Our work on Blue Waters is among the first to explore, at extremely high spatial and temporal resolution, the nature of strongly tornadic supercell thunderstorms. The development of LOFS on Blue Waters has enabled the saving of tremendous amounts of useful, intelligently compressed data that will be analyzed for years. 

We have a stable model, a robust data/visualization/analysis pipeline, strongly tornadic environments to run our simulations in, and have the demonstrated ability to save, organize, and visualize these storms at their full resolution at very high temporal resolution. Our PBS scripts are set up to recover from the inevitable node failures that will occur when running on a large portion of the machine. A handful of extra nodes is requested up front, and the PBS script will manage restarting the model from the most recent set of checkpoint files. To our knowledge, the proposed simulations would be the highest resolution simulations of tornadic supercells conducted to date.

These simulations would represent an important step towards creating a full supercell simulation at resolutions that have only been applied towards idealized chamber models of tornadoes where only the tornado and its immediate surroundings, lacking physically realistic temperature and wind variations found in real supercells, is simulated. Beyond simply conducting "hero" simulations, an abundance of manageable, scientifically useful data will be produced (and archived at the University of Wisconsin). This data, saved in LOFS format, will be analyzed and visualized both at the storm scale and at the tornado scale where we will focus our analysis on processes involving tornado genesis and maintenance, processes that remain the target of intense research by atmospheric scientists.