Abstract
Typical seismic inversion methods require a source wavelet estimated from the data. Inaccurate wavelet estimation may severely undermine the seismic inversion results. We propose a pseudo-deconvolution method that mitigates the requirement for accurate wavelet estimation in early arrival waveform inversion as follows. For a pair of observed and predicted data-sets, we first extract the respective source wavelets from the near-offset part of the two data-sets. Next, we convolve the former data-set with the latter wavelet, and the latter data-set with the former wavelet. The mismatch between these two is then minimized as in standard seismic inversion. Conveniently implemented in the frequency domain, this method is both fast and robust. This is verified by numerical tests on both synthetic and field data.