From basin to reservoir models: an integrated workflow

Reservoir geological modeling encompasses all the aspects related to the definition of the structural, stratigraphic, lithological and petrophysical features of the mineralized rock volume, leading to the estimation of the spatial distribution and of the total amount of stored hydrocarbons. Reservoir models typically result from the quantitative integration of well and seismic data through geostatistical tools. Based on such models, equiprobable realizations of structural settings and property distributions can be generated by appropriate stochastic simulation techniques. On the contrary, the integration of regional (or basin) scale information is commonly performed in a qualitative or semi-quantitative way, for example through the definition of regional property trends. This qualitative or semi-quantitative approach can strongly limit the assessment of the impact of the uncertainties associated with the regional knowledge on the overall uncertainty affecting the reservoir model. In order to overcome the limits of the traditional methodologies a different approach is here proposed, which leads to the quantitative integration of the typical dataset for a reservoir geological model (including well and seismic data) with the parameters estimated by a quantitative dynamic sequence stratigraphic model. The proposed quantitative approach could significantly improve the capability to predict the 3D facies distribution and architecture and the lithological fraction of the hydrocarbon-bearing rocks, i.e., sand fraction in a shaley/clayey environment. These features are well known to be crucial during the appraisal phase of the reservoir when relevant decisions have to be taken but few wells are drilled and volumetric estimates are performed with a limited amount of available data. Furthermore, a proper prediction of the 3D facies architecture might be very effective when planning the location of new wells or infilling wells. The application of a deterministic model (SimClast, by Delft University of Technology) is suggested to generated a 3D facies distribution and architecture at the regional scale. A reasonable range of uncertainty affecting the input parameters should be assumed to account for actual interpretation uncertainties. A geostatistical approach is then conceived to transfer the facies architecture at the reservoir scale through the integration of the available geological and geophysical data, such as well logs and seismic surfaces. Key properties at specified locations surrounding the reservoir volume which serve as boundary conditions for the reservoir models can be defined with the aid of 3-D process-based stratigraphic modeling. In this way, reservoir models can be constrained to maintain quantitative coherence with the large-scale geological setting defined by the basin-scale model and the uncertainty associated with each key basin property can be propagated all the way to reserve estimation. This approach provides a rigorous assessment of the information content of all data sources which may be used to guide further data-acquisition campaigns. The impact of the quantitative integration of basin-scale derived boundary conditions on reservoir models has been evaluated through the application of the new workflow to a synthetic case study. In particular, the workflow was applied to a fluvio-deltaic environment so as to evaluate the uncertainty reduction in the description of a facies distribution at the reservoir scale by constraining stochastic simulations to basin-derived boundary conditions. The impact of the uncertainty affecting the model input parameters (sediment entry point, sea level and initial topography) on the stratigraphic setting and channels distribution at the basin scale was investigated; furthermore, a Bayesian approach for uncertainty reduction at the basin scale was introduced by application of a likelihood function comparing each simulated scenario with the available dataset. This approach was conceived for application to real cases where the typical dataset consists of well and seismic data. Eventually, the uncertainty at the reservoir scale was evaluated by constructing sand probability curves at specified reservoir locations. The comparison with the classical procedure highlighted that the main advantage arising from the integration of basin data in reservoir modeling is an improved predictability of channel occurrence. In the basin unconstrained case the predictability of the facies architecture is null. In the constrained cases a significant vertical variability of the sand probability curve is observed, thus the channel location can actually be predicted. Furthermore, in the proposed methodology channels and floodplain occurrence are statistically preserved both as global fractions and local position. The reduction of the uncertainty of the environmental input parameters at the basin scale by application of a likelihood function significantly improves the predictability of the facies distribution at the reservoir scale. The analysis of the facies sequence at the monitoring points indicated that more precise and accurate estimates are obtained if the uncertainty is reduced by the application of the likelihood function.