Improved Storm Data Processing Through Parallel Computing Approaches

Presenter Information

Shauna Smith, Minnesota State University, Mankato

Location

CSU Ballroom

Start Date

21-4-2008 1:00 PM

End Date

21-4-2008 3:00 PM

Student's Major

Computer Information Science

Student's College

Science, Engineering and Technology

Mentor's Name

Rebecca Bates

Mentor's Department

Computer Information Science

Mentor's College

Science, Engineering and Technology

Second Mentor's Name

Deborah Nykanen

Second Mentor's Department

Mechanical and Civil Engineering

Second Mentor's College

Science, Engineering and Technology

Description

A previous research study conducted at Michigan Technological University by Dr. Deborah Nykanen and her colleague Dr. Daniel Harris analyzed storm data in order to develop algorithms that will allow coarse resolution rainfall forecasted by weather models to be optimally used in high resolution hydrology models with the goal of improving stream flow predictions and early detection algorithms that can be used to warn communities about potential flash floods. This research was performed by analyzing a series of independent radar images derived from Weather Surveillance Radar-1988 Doppler (WSR-88D) data obtained from Dr. James A. Smith at Princeton University using a series of computer programs written by the original researcher and her colleagues. The program was run using a .sequential algorithm that takes up to 16 hours to execute. Because of the structure of the problem, there was an opportunity for applying parallel programming techniques to the program code. In order to speed up the program execution time, several different parallel programming approaches have been applied to the code including data and task parallelism. Speedup analysis has been conducted for each different type of parallel programming approach. The parallel programming assessment results show how different parallel approaches affect the speedup of code. The faster code will aid in analyzing future storm data, allowing more data to be analyzed in a shorter amount of time, and will eventually be used in improving lead time on high resolution stream flow predictions and flash flood warnings. The speedup provided by the different parallel programming approaches has been verified on previously analyzed data and applied to new data.