Lesson Learned From Static Model Uncertainty Analysis

This paper documents lessons learned from conducting uncertainty analysis for 3D static modeling. Example cases, evaluation techniques and a proposed guideline are presented to resolve misconceptions in probabilistic studies. The technological advancements nowadays have allowed thorough probabilistic studies to be conducted. In most 3D static modeling software, a recorded 3D modeling workflow can be incorporated with various variables and run as an experimental design with several runs. However, the lack of transferred knowledge and publications regarding probabilistic guidelines has led to many inconsistencies, misconceptions, and confusion about the fundamentals, during designing, running, or post-run analysis. For example, during running analysis, the result is inconsistent between the field-level median model (P50) with zone-level median models. Later, this could raise a history matching issue since it obscures unreliable zone-level recovery factors due to overlapping in-place among tanks. An example from the post-run analysis is a missing P50 from the list due to an even number of experiments. Every tool has different methods of processing and visualizing statistics output. Assuming a similar definition and approach can lead to confusion. A strategy to assign a facies model that represents a geo-body at a specific time or zone is described by considering dependency and hierarchy among variables. The result is calculated, visualized, and ranked at the zone level to avoid percentile misconceptions. Then, the corresponding percentile from each tank is combined to build a consistent field-level model. It is essential to understand which statistical approach a tool uses, and whether it uses a function to determine the percentile inclusive/exclusive of the first and last values in the array. Integration work could identify problem earlier and then resolve them before further effort. This paper aims to guide in building a consistent uncertainty analysis modeling and avoid unnecessary rework due to misconception.