History Matching with Time-lapse Seismic - Will Any Starting Model Do?

In general our knowledge of the reservoir is insufficient for accurate representation. We generate models that honour well logs, core, 3D seismic etc. but there is inevitably some uncertainty which we try to include in simulatioms via geostatistical modelling. History matching is then based on updating a model to improve the fit to the observed dynamic data. The choice of starting model in the history matching process is therefore critical and we need to know that our parameterisation and updating scheme can convert it into a better representation resembling reality. In this paper we describe how we set up some synthetic test cases to validate the history matching and parameterisation scheme. We find that if we use a pilot point separation that is equivalent to the range of the variogram used in generation of permeability distributions, we can get a good representation of the model.