Stochastic and deterministic seismic inversion methods for thin-bed resolution

Abstract

Results from a typical post-stack or pre-stack seismic inversion are band-limited primarily due to missing low and high frequencies in the wavelet. Thin beds that are of primary interest in reservoir characterization are generally poorly resolved. Inversion methods are designed to incorporate a priori information in several different ways. In this paper, we examine and compare performances of two such methods: a stochastic method that draws a starting broad-band model based on some statistics, derived from well logs, and a deterministic method that incorporates constraints using a model norm, based on the basis pursuit decomposition. We apply a stochastic method employing very fast simulated annealing for optimization, using fractal-based starting models, and a basis pursuit inversion (deterministic method) to a single post-stack dataset to assess the performance of the two algorithms. We find that both methods are able to derive a broad-band acoustic impedance from the band-limited post-stack seismic data. The results show a similar structure and therefore result in possible increasing confidence in the interpretation. The essence of our study is that the calibration of results from two different methods can provide an interpreter with more information in interpreting broad-band inversion results hitherto not derived from standard deterministic algorithms.