Application of Principal Component Analysis for Frequency-Domain Full Waveform Inversion

In recent years, the research and application of the full waveform inversion (FWI) gradually tend to be diversified. Whereas how to improve the computational efficiency have always been a hot research topic and an important prerequisite allowing for full waveform inversion a mature seismic imaging tool. Building supershots with random source encoding technique could significantly reduce the computational costs at the expense of introducing crosstalk noise during the inversion. Alternatively, this paper presents a new data-reduction FWI approach based on a classical statistical technique called principal component analysis (PCA). Model experiment and numerical analysis prove its ability to compress the data dimension and obtain a high speedup within the FWI procedure, especially at low frequencies. Moreover, a hybrid method combining random source encoding and PCA is presented and embedded in the frequency-domain FWI. The aim of this method is to provide a tradeoff between the speedup and the robustness of FWI.