Implementation Of The GPU-Based Parallel Algorithm Of High Accuracy Surface Modeling

In terms of the fundamental theorem of surfaces, High Accuracy SurfaceModeling (HASM) method is based on the differential geometry theory, transforms thesurface modeling to a linear system by using the principle of least square method anddescribes the properties of the surface. Compare to the classical surface methods inGIS application, such as inverse distance weight (IDW), triangulated irregular network(TIN), KRIGING and SPLINE, HASM has a much higher accuracy and the errorproblem of the surface modeling is solved in theory by HASM. However, thecomputational cost of HASM is high.With the development of electronic technology, the capability to floating pointcalculation of Graphic Processing Unit (GPU) will surely improve over time. Parallelcomputing based on GPU is playing an increasing important role out of the imageprocessing field. Keeping the high accuracy of HASM, the purpose of the paper is tosolve the high computational cost by using parallel processing capacity of the GPU. Tosolve the linear system in HASM, we select Preconditioned Conjugate Gradient (PCG)method which is widely used for solving the large sparse system. The devicearchitecture of GPU is the Computer Unified Device Architecture (CUDA) fromNUIDIA Company. Considering the PCG method in HASM, we modify the calculationorder and abandon the non-zero elements in the linear system so as to improve thecomputational efficiency and then we paralleled the PCG algorithm of HASM.Finally, combine to the digital elevation model of the global, Long term (1998-2008) averaged monthly data, obtained from sites measuring temperature (3000sites), are used to simulate the global mean temperature in May, June, July, and Augustof the eleven years from1998to2008. Numerical results show that with this CUDAtechnique, the computing efficiency of HASM is increased by20times as comparedwith the existing sequential algorithm.