Towards Accounting for Uncertainty in Geological Model Description and Geological Realism of in Reservoir Prediction

We propose a novel approach to evolve the model through the update process based on the ensemble of possible model realisations that are fused together in a data driven way rather than assimilated under certain assumptions. Multiple Kernel Learning (MKL) is a learning-based technique, which provides a way to blend together multiple pattern information and select the principle spatial features that are more relevant to data. Solving the feature selection problem with MKL allows to combine spatial patterns that represent geological characteristics at different scales.