Parallel partitioned inverse method for least-squares adjustment

Parallel computing is undoubtedly the trend in numerical applications of highly intensive computation. There has been much related research and development on parallel computer architecture, algorithm design, and supplementary packages. However, computational technology has seen little interest in the surveying area since the North American Datum of 1983 adjustment. In this research, a parallel partitioned inverse algorithm is implemented and applied to a least-squares adjustment of horizontal survey networks to present the potential of parallel computing methods for surveying data. Two observation data sets with 2,412 and 1,902 unknowns were used for the test. To improve performance of the algorithm, two different partitioning schemes also were investigated with the data sets. The computational experiment presents the good scalability of the algorithm and batter partitioning approach with the improved speed. However, it is noted that parallel factorization of sparse matrices is required to fully utilize the proposed approach.