adoption as the tool of choice for subsurface characterisation at and away from
control. The derivation of the impedance data required to perform this characterisation typically utilises a starting background
model (LFM). This LFM is required because seismic data lack the low to ultra-low frequencies required to extract absolute elastic property estimates, which are a pre-requisite for quantitative subsurface characterisation. The LFM is iteratively updated over the seismic bandwidth by matching forward synthetic seismic models with seismic trace data until the misfit between the two is reduced to an acceptable level – a process known as model-based inversion.
The challenge with this approach is that the LFM is not updated outside the seismic frequency band; e.g. the overall compaction trends present in the starting LFM are fundamentally unchanged. In other words, there are many different LFM scenarios that will give the same or similar fits to the seismic data, because the fit is performed only within the seismic bandwidth. However, the outputs from these different LFM scenarios can give rise to dramatically different absolute rock property estimates and a wide spread
in resulting petrophysical rock property estimates - there is much ambiguity. This fact is sometimes overlooked during subsurface seismic characterisation, with many practitioners preferring to utilise a single ‘best’ LFM, or perhaps just a handful of simple input LFM’s utilising different well
calibration points in the construction of each. There is also considerable uncertainty in the inversion result over the seismic bandwidth as seismic is a noisy signal. The uncertainty over the entire frequency band needs to be understood in order to better characterise e.g. reservoir connectivity and pay thicknesses, particularly in marginal reservoirs. After all, depending on the number of (elastic) facies
in the subsurface and their corresponding elastic properties there may be a large number of different plausible combinations of facies interfaces which will give rise to the observed seismic response.
In this paper, we propose a two tiered inversion strategy that aims to address and better explore the solution space during seismic inversion and reservoir characterisation. First a range of plausible geological prior scenarios are defined in terms of layer configurations and horizon/picking uncertainty, number of facies and their corresponding abundances, and rock property trends and relationships (with associated uncertainties). Then, per scenario, stochastic Markov chain Monte Carlo sampling (McMC) is performed to create equiprobable realisations from the posterior distribution.