Waveform inversion from a poor starting model – using a residual ‘drip-feed’ strategy

We present a new waveform inversion scheme designed to avert the need for an accurate starting<br>model and low frequency content in the data – a necessary key step in making the technique work on<br>a much wider range of exploration datasets and targets than it currently can. The scheme operates by<br>preceding the inversion of the field data by inversion of intermediate target datasets – synthesised out<br>of the curl-free (irrotational) part of the phase mismatch at the lowest useable frequency. We<br>demonstrate its effectiveness over the corresponding conventional approach by inverting data from the<br>Marmousi model with a 1-D starting model and minimum frequency of 5Hz.