TOWARDS PROCESS-BASED GEOLOGICAL RESERVOIR MODELLING: OBTAINING BASIN-SCALE CONSTRAINTS FROM SEISMIC AND WELL DATA

Forward stratigraphic modelling aims at representing the spatial distribution of lithology as a function of physical processes and environmental conditions at the time of deposition so as to integrate geological knowledge into the reservoir modelling workflow, thus increasing predictive capabilities of reservoir models and efficient exploitation of hydrocarbons. Application of process-based models in inverse mode is not yet well-established due to our limited insight into the information content of common subsurface data and the computational overhead involved. In this paper we examine inverse modelling of stratigraphy by using a typical dataset acquired in the hydrocarbon industry, which consists of seismic data and standard logs from a limited number of wells. The approach is based on the use of a forward model called SimClast, developed at Delft University of Technology, to generate facies distribution and architecture at the regional scale. Three different goodness of fit functions were proposed for model inversion, following an inference approach. A synthetic reservoir unit was used to investigate the impact of the uncertainty affecting the input parameters and the information content of seismic and well data. The case study showed that the model was more sensitive to the initial topography and to the location of the sediment entry point than to sea level. The depth of the seismic reflector corresponding to the top-reservoir surface was the most informative data source; the initial and boundary conditions of the simulation were constrained by evaluating the depth of this reflector across the whole basin area. In the reservoir area, where the seismic-to-well tie was established, the depth of the reservoir top does not give enough information for constraining the model parameters. Our results thus indicate that evaluation of basin-scale data permits reduction of uncertainty in (geostatistical) reservoir models relative to the current workflow, in which only local data are used. Effective use of well data to generate reservoir models conditioned to basin-scale scenarios requires post-processing methods to downscale the output of the forward model used in the experiments.