GNGTS 2018 - 37° Convegno Nazionale

GNGTS 2018 S essione 3.3 749 spatial correlation and zero-mean Gaussian distribution of velocity pertubations. Under these assumptions, if the seismic image approximates a zero-offset primary reflectivity section then the autocorrelation function of the image should approximate the autocorrelation of the subsurface velocity structure convolved with the autocorrelation of the source wavelet in depth, with a correction for the lateral resolution of the migrated section (Irving et al. , 2010). This method seeks to characterise the medium by inverting the chaotic zone in the seismic image for vertical and lateral correlation lengths (Fig. 2). These parameters can characterise the degree and type of internal deformation from sliding. As found by Irving et al. (2009), the inversion is poorly constrained for vertical correlation lengths due to the “vertical derivative” effect of the seismic wavelet in surface reflection experiments. However, the aspect ratio (lateral correlation length divided by vertical correlation length) is relatively well constrained. If vertical correlation length can be estimated from downhole data (eg Cheragi et al. , 2013) then absolute lateral correlation length can be calculated. We demonstrate this technique using a synthetic benchmark. We build a sigmoidal submarine slope model (5000 m x 400 m) containing a buried submarine landslide. The synthetic seismic image is produced by 2-D visco-elastic forward modelling with a marine multi-channel Fig. 2 - Flowchart showing i) generation of candidate models for the stochastic zone from the 2-D autocorrelation function (ACF), here parameterised by the lateral and vertical characteristic scale lengths (a x and a z ); ii) integration of deterministic and stochastic components of the model onto a spatial grid; iii) seismic forward modelling, processing and imaging; iv) extraction of relevant chaotic zone from modelled seismic image and v) comparison of autocorrelation functions of modelled and observed chaotic zones. Models which meet the acceptance criteria are considered “plausible” for the observed data and their parameters are added to the output ensemble.