Dynamic Intercept-gradient Inversion

Seismic data provides bandwidth limited approximations of the Earth reflectivity. Its bandwidth is non-stationary; varying with time, space, angle and azimuth. Many seismic analysis techniques are not directly applicable to non-stationary data. For example, pre-stack amplitude variation with angle/offset (AVA/O) analysis requires data pre-conditioning to eliminate the angle dependence of bandwidth, restricting its application to a limited time window. To avoid these irritations we cast Shuey’s two-term AVA equation as an inverse problem, searching for best fitting intercept (I) and gradient (G) models simultaneously, given a set of dynamic wavelets. The dynamic wavelet set describes the non-stationary nature of input data, and the inversion is damped towards a seismically derived rock physics trend. We show that dynamic intercept-gradient inversion produces broader bandwidth results (without noise amplification) which cover the full time extent of the data. No data pre-conditioning is required and the inversion needs no input from well logs, making the method appropriate in exploration settings. Using the I and G volumes in extended elastic projections accentuates fluid related anomalies, as evidenced through blind comparison to well logs. These anomalies can be automatically identified, extracted as 3D geobodies, and ranked; allowing rapid screening of large seismic volumes.