Seminar: Parallelizing the Geophysical Inversion Program ArjunAir: A Hybrid Distributed/Shared Memo
Department of Earth Sciences
Memorial University of Newfoundland
Parallelizing the Geophysical Inversion Program ArjunAir: A Hybrid Distributed/Shared Memory Approach
Department of Computer Science
Thursday, February 20, 2014, 1:00 p.m., Room EN-2022
Airborne electromagnetic (AEM) induction surveying is frequently used in mineral and groundwater exploration to locate and characterize electrically conductive structures beneath the surface of the earth. Geophysical AEM inversion is the quantitative process of estimating the conductivity structure of the earth from AEM survey data. This presentation will describe efforts to parallelize the open-source AEM inversion software package ArjunAir and reduce its memory footprint. The program was developed in the 1990s and early 2000s, written entirely in Fortran 90. It poses the two dimensional AEM inverse problem as an underdetermined, regularized non-linear least squares optimization problem. Use of the program has been limited by its computational cost and large memory requirements. Replacing the most time consuming routines in ArjunAir with parallel versions has greatly reduced runtimes and allowed the code to take advantage of cluster computing resources at MUN. Two new parallel versions of the ArjunAir forward modelling algorithm will be discussed in this talk. The first uses a distributed memory programming model, implemented using MPI and the second a hybrid distributed/shared memory model implemented using MPI and OpenMP. Modifications of the ArjunAir inversion algorithm designed to reduce its memory use will also be discussed.
Bio: Patrick Belliveau is an M.Sc. candidate in geophysics in the MUN Earth Sciences department. His thesis research focuses on developing parallel algorithms for airborne electromagnetic geophysical inversion. He has a B.Sc. degree in physics and mathematics from Simon Fraser University. Before coming to MUN, Mr. Belliveau worked in the mineral exploration industry, processing, visualizing and interpreting geophysical data.