Improved convergence in seismic history matching combining experimental design with a divide and conquer approach

History matching is performed by changing parameters in reservoir models to reduce the difference between observed dynamic data and predictions. Time-lapse (4D) seismic offers spatial information in the inter-well region and acts as additional constraints to more conventional production data from wells. Finding appropriate models can be a slow process even with additional seismic constraints. In this study we speed up convergence of history matching using experimental design to firstly set up a proxy regression model and then analyse this to identify sub-domains of the parameter space. These can then be searched very efficiently to find models that predict production and seismic data. We apply the approach to the Schiehallion field and show a factor ten speed up in convergence with no loss of match to observations of 4D seismic and production data.