Addressing structural uncertainty through seismic forward modelling.

Seismic reflection imaging provides one of the most widespread datasets for interpreting subsurface geometry. Subsequent interpretations are known to be uncertain and much remains to be done to understand and quantify that uncertainty. Seismic forward modelling offers the ability to: investigate the accuracy of contemporary workflows through comparison with known truth-case models; understand imaging constraint in different settings by modelling wave propagation, and; compare potential interpretations directly to the original seismic image. Combining contemporary approaches of reservoir modelling, structural restoration and seismic modelling, this thesis sets out to demonstrate a workflow to quickly generate detailed synthetic images. Generating example cases at various scales, I test contemporary analytical approaches to demonstrate the uncertainty introduced by seismic interpretation of structural geometries. In doing so, I demonstrate that; in the case of an isolated normal fault, seismic attribute analysis may both over- and under-estimate fault length. Smoothing of structural geometries during seismic interpretation can increase the apparent cross-sectional areas available for fluid flow, increasing production forecasts by up to 40 %. Using synthetic data to support a multi-stochastic exploration assessment results in a 16 % difference in estimated resource compared a single case deterministic estimation. In all of these cases, seismic forward modelling has assisted quantifying uncertainty and assisting interpretation. However, uncertainty in seismic images is spatially variable, and areas returning little energy to the surface will always challenge interpreters. To this end, mechanical modelling is trialled as a method to integrate geometric data from more certain aspects of an interpretation to produce a geologically feasible geometry. These methods show great potential for testing at present and are likely to become more broadly applicable with increasing computational efficiency.