Efficient Compression Methods Of Seismic Forward Wavefield

As the scales of seismic exploration and wavefield simulations continue to ex-pand,massive seismic data are generated and they need to be stored for post-processing and interpretation.Although this issue seems to be addressed properly by using large-capacity storage devices,the subsequent processing will face an inherent problem of the current supercomputers,which is a growing disparity between supercomputer compu-tation speeds and I/O rates.Due to the limited performance of the latter one,processing fully-stored seismic data will inevitably spend considerable time on the I/O operations of disks.To this end,many seismic data compression algorithms have been proposed,such as Huffman coding,arithmetic coding,LZW coding,etc.in lossless compression algorithms,and Discrete Cosine Transform(DCT)compression,Dreamlet transform compression,wavelet transform compression,etc.in lossy algorithms.In the compres-sion of massive seismic data,due to the low compression ratios of lossless algorithms,the lossy compression algorithms mentioned above are widely used.However,they can not control the reconstruction errors before compression,and the specific implementa-tion is relatively complex.Based on the detailed research on the above traditional lossless and lossy com-pression algorithms,we have introduced a novel data compression algorithm-Squeeze algorithm,which is error-bounded and easy to implement and use.Before our research,there was no related work of Squeeze algorithm in seismic data compression,so this work is the first one to combine Squeeze algorithm and seismic compression issue as well as the seismic wave forward modeling.Relying on the excellent compres-sion implementations of Squeeze algorithm itself:multi-dimensional and multi-layer prediction,adaptive error-bounded quantization and entropy coding operations,this al-gorithm can obtain a good compression result in the actual compression tests.In our work,Squeeze algorithm is applied to three test data:1,data of 2D seismogram;2,single-section and single-component wavefield data;3,full-space and full-components wavefield data.The first two tests have compared the compression results of DCT and Dreamelt algorithms,which shows among these three,Squeeze algorithm has the best compression effect,followed by DCT.The third test is mainly to explore the differences of compression performance of Squeeze algorithm under different compression settings by analyzing reconstruction data errors and time overheads.For the problem of "ZeroShift",which is caused by Squeeze algorithm when the error bound mode is set to the relative error bound(REL)and the value of error bound is relatively large,the gradient correction method and the bandpass filter correction method are proposed to reduce some negative effect on the later data processing.The changes in errors of seismogram,snapshots and sensitivity kernels obtained from the corrected data illustrate the effectiveness of the correction strategies.We also have proposed a hybrid parameters error-bound method—the HL compression mode,to further reduce the impact of the "ZeroShift",on calculated sensitivity kernels.By applying high-precision compression setting in the initial stage of the wavefield modeling to avoid "ZeroShift" of data as much as possible,and applying low-precision compression setting to ensure compression ratio in subsequent stage,this method can obtain satisfactory compression results with a little loss of compression ratio.