On Uncertainty Quantification of History Matched Facies Models

Estimation of facies boundaries adds additional complexity to history matching process. Because facies types are categorical variables, history matching methods that rely on derivatives (gradient based methods) or Gaussian assumption (e.g. ensemble-based methods) cannot be readily used. It has been shown that by appropriate parameterization history matched models with desired geological facies features can be obtained. Little attention, however, has been paid to history matched facies models in terms their ability to quantifying uncertainty after conditioning to dynamic data. In this study, we analyze the quality of uncertainty quantification of history matched TPG and MPS facies models in terms of their representation in the model space and in terms of predictability using several synthetic examples. Although it is generally thought that the predictability of reservoir models with realistic facies distributions will be better than less realistic models, the benefit of model complexity for predictions has not been well established. In this study we show scenarios in which good prediction can be achieved with models that do not have correct geological features. The importance of geology for model predictability likely depends on the type of geological features, available data for model calibration, and the quantity to be predicted.