Enabling Petascale Ensemble-based Data Assimilation for Numerical Analysis and Prediction of High-Impact Weather
Collaborative Research project #0904938 funded by the NSF Office of Cyberinfrastructure under the American Recovery and Reinvestment Act of 2009
- Principal Investigators
- Ming Xue, University of Oklahoma
Subhash Kak, Oklahoma State University
Sergiu Sanielevici, Pittsburgh Supercomputing Center
Xiaolin (Andy) Li, University of Florida
This project focuses on the very difficult problem of predicting thunderstorm initiation, propagation, and evolution. To increase the accuracy of forecasts, high volumes of weather observations must be assimilated. The size and scale of the problems to be addressed are only attainable on petascale computing systems.
This research aims to produce more accurate and timely thunderstorm forecasts, and could have a potentially transformative impact on hazard analysis and management.
Advanced Regional Prediction System Software
The Advanced Regional Prediction System (ARPS) code is being used to support this project. Because the data scales are so large, and the effectiveness of the predictions so time-dependent, it is very important for the ARPS code to run as efficiently as possible.
PSC scientists David O'Neal and Mahin Mahmoodi worked with project researchers to discover ways to improve the ARPS code. They used the Tuning and Analysis Utilities (TAU) tool kit to determine areas of interest in the code. A discussion of the work done profiling the ARPS code can be helpful to anyone applying TAU to a large scientific code. For more information, including insight and tips on how to proceed, see this discussion.