A High Resolution Analysis of Spatio-temporal Variations of Geomagnetic Field using Parallel Computing.
A Thesis Submitted to the School of Postgraduate Studies, University of Lagos.
This work presents a high resolution analysis of spatio-temporal variations of geomagnetic field using parallel computing. Geomagnetic field diurnal and hourly variation have been used extensively to study extant and emerging geophysical phenomena. Although, geomagnetic field variations occur even on nanosecond basis, however, to date, algorithm development and computing hardware architecture limit the resolution of monitoring solar quiet (Sq) daily variation to hourly and regional basis. This is obviously inadequate for detailed analyses of some geophysical phenomena with time scale less than hours. Also, regional analysis of this variation is insufficient for inter-regional phenomena. It therefore follows that any tool that facilitates a more detailed analysis of the earth’s spatio-temporal magnetic field is bound to lead to a better understanding of related phenomena observed on earth. Thus, for the purpose of high resolution spatio-temporal analysis of geomagnetic field variations, a 366-day of 1- minute sampled geomagnetic field dataset was obtained from 64 observatories. These observatories are evenly distributed across the globe on the INTERMAGNET’s network of 1996. The data was pre-processed and corrected for non-cyclic variations leading to a [366 x 276480] partitioned matrix. To obtain solar quiet (Sq) daily variation from the resulting large system, this work developed serial and parallel computing algorithms that were executed on a MatlabR2010a environment. It employed the well-known kriging method for geostatistical data gridding and mapping of solar quiet (Sq) daily variations. The results show that whereas sequential algorithm running on an Intel Xeon E5410 2.33GHz processor generates Sq conforming to international quiet day (IQD) standards in 18.5 minutes, the parallel algorithm running on 8 similar processors produces the Sq in 2.95 minutes. This translates to a speedup of 4.32 for the developed parallel algorithm using the benchmark performance metrics of Amdahl’s. On logarithmic extrapolation, it was deduced that the parallel algorithm would run optimally on 13 Intel Xeon E5410 2.33GHz processors to generate solar quiet in less than a minute. However, the resultant relatively high index of performance called for an extension of the Amdahl’s model to account for parallel overhead. Thus, in addition to enhanced analysis of geomagnetic field variations, the developed computational platform also enabled a high frequency availability of solar quiet daily variations. Consequently, this work provides a basis for early warning system and better understanding of some hitherto unclear geophysical phenomena such as travelling ionospheric disturbances, equatorial anomalies and high geophysical activities in the Polar Regions.