Observation Sensitivity in Computer-assisted History Matching

This study concerns computer-assisted history matching of reservoir simulation models, i.e. systematic updating of model parameters to minimize the mismatch between observed and simulated production data, with the aim to improve the predictive capacity of the model. The goal of our research is to quantify the information content of the observations. Following a method developed in meteorology, we employ an observation sensitivity matrix to quantify the effect of observed data on predicted data. The use of this matrix is illustrated with an example in which we adjust the permeability field of a two-phase two-dimensional reservoir model by means of a particular history matching technique, the representer method. This method particularly allows for efficient computation of the observation sensitivity matrix. Conceptually, however, the use of an observation sensitivity matrix is equally valid for other history matching techniques. In our example the information content of the updated model comes mostly (96%) from a priori knowledge and to a much lesser extent (4%) from the observations. This finding is in line with the practical experience that in computer-assisted history matching using production data the results are strongly influenced by the prior model.