Accelerate large-scale seismic data Kirchhoff time migration in spark

Seismic exploration are creating massive amount of data by using advanced scientific instruments and computer simulations. Pre-stack Kirchhoff Time Migration (PKTM), as one of the most commonly-used imaging method in oil and gas industry, is facing with severe performance defects due to its computationally expensive feature and huge amount of input. This requires new technologies for improving computing performance and processing scalability. In this paper, we propose a lightweight module to support large-scale seismic data processing and parallel implementation of PKTM algorithm atop Apache Spark framework. Some optimization methods such as an auto- chunk method and columnar storage re-arrangement are added to improve the performance of data processing and PKTM migration. Experimental results show an improvement of 2.4 times over direct RDDs-based Spark processing and 3.2 times over Hadoop implementation, and our method shows simplicity in seismic data processing and scales linearly with the input data size.