Accelerating a Geostatistical Approach to Groundwater Pollution Source Identification with GPU Computing

Pollution source identification (PSI) is a very important step of groundwater contaminant treatment strategy. This paper aims at improving the computing efficiency of the geostatistical approach to the PSI problem by the GPU parallel technique. Firstly, we introduce the geostatistical approach. Then, we analyze the time consuming of the geostatistical approach. As the main steps to solve the geostatistical model, the Levenberg-Marquart method involves a lot of iteration operations and matrix computations costs 79% computing time. The estimation procedure of geostatistical approach could be divided into an out loop and an inner loop. The GPU parallel technique is used to accelerate the inner loop, while the out loop is implemented by the serial scheme. Finally, a numerical case is used to analyze the performance of the parallel method. The evaluated result of the parallel method has good agreement with the real contaminant release history in the numerical case. The GPU parallel technique improve the computing efficiency of geostatistical approach obviously

Keywords:Geostatistical; Groundwater; Pollution; Identification; Parallel; GPU