Comparison of Uncertainty Estimates from Deterministic and Geostatistical Inversion

Estimating uncertainty is an important aspect of seismic reservoir characterisation. One reason, amongst others, given for carrying out geostatistical inversion is to quantify uncertainty. Recently a number of authors have shown that uncertainties can be approximately captured from deterministic inversion. Given that deterministic inversion is much quicker than geostatistical inversion, it is worth understanding the differences that can be expected in terms of uncertainty estimates from the two processes. In this paper, we compare the inversion of multi-angle stack data using a simultaneous constrained sparse-spike algorithm for deterministic inversion with the inversion obtained using a Markov Chain Monte Carlo algorithm for geostatistical inversion. The deterministic inversion yields a single model of elastic properties that can be subsequently manipulated to provide estimates of lithology probability and net pay. The geostatistical inversion produces multiple models of elastic properties, reservoir properties (net pay) and lithology so that uncertainty estimation is straightforward without ad hoc procedures. The inversion results are compared in terms of lithology probability estimates and net pay maps.