Bias Aware Data Assimilation in Reservoir Characterization
B. Sebacher, M. Glegola, R.G. Hanea and G. Kaleta
Event name: ECMOR XIV - 14th European Conference on the Mathematics of Oil Recovery
Session: Ensemble Methods
Publication date: 08 September 2014
Info: Extended abstract, PDF ( 2.44Mb )
Price: € 20
In reservoir characterization, modern reservoir modeling and Assisted History Matching aim at delivering integrated models with quantified uncertainty, constrained on all relevant data. Traditionally, the reservoir model is updated using only the dynamic production data from the wells. Recently, more and more efforts are made to use Geophysical Reservoir Monitoring (GRM) data in history matching, as these types of data can provide valuable information about the reservoir characteristics and geological formations over the whole field. Time-lapse (4D) gravimetry is a direct measure of a subsurface mass flow and can provide valuable information in this context. It offers an attractive aerial monitoring technique for reservoirs containing fluids with high density contrasts, e.g., gas and water or oil and steam. The method is especially promising for shallow reservoirs as the 4D signal will be stronger for large and shallow reservoirs, compared to smaller and deeper reservoirs. In reservoir history matching, often an assumption is made that the forward model predictions and the observations are unbiased, i.e., there are no systematic errors. In this study we investigate the added value of gravimetric observations for gas field monitoring and aquifer support estimation, under the assumption that both model and observations are biased. We perform a numerical study with a realistic 3D gas field model which contains a large and complex aquifer system. The aquifer support along with other reservoir parameters, such as porosities, permeabilities, reservoir top and bottom horizons etc., are jointly estimated using the Ensemble Smoother (ES). We show that the influence of the observation bias and/or the model bias on assimilation results can be severe and may lead to large errors in the estimations of the states/parameters. By using bias-aware data assimilation methodology, the bias can be estimated separately from the state, and we show that the deteriorating bias influence on the assimilation results to a large extent can be mitigated.