Benchmarking Prestack AVO Inversion

In 2005, AVO inversion of prestack seismic data from a BP gas field was performed by three different companies. The results were highly variable and inconclusive at identifying the known zones of fizz or residual gas saturation within the field. In 2006, BP decided to perform a benchmark of the available AVO inversion algorithms using a very simple synthetic AVO dataset, generated from a known model containing variations in gas saturation at all depths. Six companies took part and delivered predictions of variations in acoustic impedance (AI), shear impedance (SI) and density away from a calibration well. Results were analysed by cross-plotting attributes from band-limited inversions versus attributes from the true band-limited model. Comparative results are presented here and show that three companies were able to recover variations in AI, SI and density with a high degree of confidence. The remaining three companies struggled with either excessive scatter or non-linearity in the inversion results. It is suggested that AVO inversion codes should be tested using appropriate synthetic AVO gathers before commencing AVO inversion of real prestack data.