Surface seismic imaging with multi-frequency full-waveform amplitude inversion

Abstract

Numerical experiments have been performed to test three inversion schemes for the imaging of surface seismic data. Three shallow subsurface models were considered: buried karst topography, dipping blocks, and isolated waste ponds. The synthetic experiments involved only a limited number of ?surveys? with just 8, 3, or 1 shots into variable length geophone arrays. In all cases, conventional seismic data processing hardly recovers the structure. The inversion algorithm works in the frequency domain and relies on a finite element method to do the 2D/2.5D acoustic wave modelling. Our original inversion scheme used the known source signature (KSS), whereas the other two schemes either estimate it from the various shot gathers (ESI), or perform a normalised data inversion (NDI). The spectral inversions typically worked over 50-400 Hz and involved about 3500 data points. The features in the models typically have dimension of 5-20 m, which can be compared with the geophone spread length of 40 m (8 shots) to 100 m (3 or 1 shots). The 8 shot inversion performed best for all the models. In fact, it was not possible to recover the karst structure with just 1 or 3 shots, but the other models were recovered, albeit with less accuracy, using such a small number of shots. ESI and NDI perform remarkably well on the dipping block and waste pond models. KSS yielded superior results to the other two schemes. The estimated source spectra were close to but did not exactly match the known wavelets.