A probabilistic approach to jointly integrate 3D/4D seismic, production data and geological information for building reservoir models

Reservoir modeling aims at understanding static and dynamic components of the reservoir in order to make decisions about future surface operations. The practice of reservoir modeling calls for the integration of expertise from different disciplines, as well as the in tegration of a wide variety of data: geological data, (core data, well-logs, etc.), production data (fluid rates or volumes, pressure data, etc.), and geophysical data (3D seismic data). Although a single 3D seismic survey is the most common geophysical data available for most reservoirs, a suite of several 3D seismic surveys (4D seismic data) acquired for monitoring production can be available for mature reservoirs. The main contribution of this dissertation is to incorporate 4D seismic data within the reservoir modeling workflow while honoring all other available data.; This dissertation proposes two general approaches to include 4D seismic data into the reservoir modeling workflow. The Probabilistic Data Integration approach (PDI), which consists of modeling the information content of 4D seismic through a spatial probability of facies occurrence; and the Forward Modeling (FM) approach, which consists of matching 4D seismic along with production data.; The FM approach requires forward modeling the 4D seismic response, which requires to downscale the flow simulation response. This dissertation introduces a novel dynamic downscaling method that takes into account both static information (high-resolution per meability field) and dynamic information in the form of coarsened fluxes and saturations (flow simulation on the coarsened grid).; The two proposed approaches (PDI and FM approaches) are applied to a prominent field in the North Sea, to model the channel facies of a fluvial reservoir. The PDI approach constrained the reservoir model to the spatial probability of facies occurrence (obtained from a calibration between well-log and 4D seismic data) as well as other static data while satisfactorily history matching only production data; however, a probabilistic type of match is achieved rather than an deterministic physics-based match of the 4D seismic data. The FM approach achieved a partially good quantitative match on both production and 4D seismic data; however, history matching two data of very different support was considerably more challenging than the PDI approach.