Consistent stochastic depth conversion using UDOMORE Depth
Performing depth conversion using mainstream deterministic/stochastic frameworks can lead to bias and result in flawed depth models. With the UDOMORE Depth Ocean plug-in to Petrel, Seisquare offers a consistent stochastic framework to address bias at its core, which ultimately maximizes the accuracy of depth models and better supports E&P decision-making.
Under mainstream depth conversion frameworks, there are two main sources of bias: first, the “base case” velocity model; second, the “layer cake” framework.
- The “base case” velocity model: usually computed as the sum of a velocity “trend” model built from interval velocity functions and well tying velocity “residuals”. The velocity “trend” model is usually deterministic and taken for granted (i.e. there is an assumption that the velocity “trend” model has no uncertainty). At best, only local well tying depth or velocity “residual” uncertainty is simulated around the trend (using kriging or other interpolators). This is a strong and risky assumption: no matter how good a velocity “trend” model may be, it never fully reflects the true unknown velocity model; it always contains inherent uncertainty that needs to be addressed.
- The “layer cake” approach: Ztop (top of a layer in depth) interpreted at time T = Ttop, controls the conversion of the layer immediately below. This is also a strong and risky assumption: in a “layer cake” depth conversion case, Ztop results from depth converting the previous layer. As mentioned above, Ztop may contain bias which propagates through successive layers in depth. Uncertainty cannot be quantified correctly with this framework, as exposed by P. Abrahamsen (1993, Bayesian Kriging for Seismic Depth conversion of Multi-layer Reservoir, in A. Soares (ed.)) in his example on deviated wells.
Seisquare advocates that the most consistent stochastic framework for depth conversion combines a “multilayer” approach with “stochastic” velocity model building. It relies on Bayesian Co-Kriging [Abrahamsen 1993, Omre 1987, Omre & al 1989, Sandjivy & al 2009] of the well depth markers using appropriate time derived external drifts. The main argument in favor of using Bayesian Co-Kriging is consistent integration of all sources of uncertainty throughout all layers within a unique probability model.
A benchmark of consistent stochastic vs. mainstream depth conversion approaches, using exactly the same prior inputs, shows considerable differences:
The consistent stochastic framework described above has produced excellent results in E&P on-shore and off-shore operations in Continental Europe, Middle-East, North Africa, Norway South Atlantic and UK. The output depth and velocity models and associated volume predictions are recognized for accuracy/precision, better consistency through time, and are easy to update as new data comes in.
This framework is made seamlessly accessible to users thanks to Seisquare’s UDOMORE Depth Ocean plug-in to Petrel. For additional information on our approach, download our technical paper now!