f Rapid forecasting of uncertain reservoir responses using functional data analysis

Uncertainty in future predictions of oil reservoirs is traditionally obtained via a history matching procedure which generates a set of reservoir models that match the available data to within a user defined tolerance. The process of creating history-matched models can be a very expensive and difficult task, and often needs to be repeated when new data is obtained. In this presentation, we propose a diagnostic tool which indicates rapidly to what extent the posterior uncertainty on the forecast will be reduced, if at all, by the data. To do so, instead of generating models that match the data, we propose to estimate the relationship between the historical and forecast variables. This relationship is then used when new historical observations are available to update statistically the uncertainty on the forecast. Through this procedure, an estimate of Bayesian posterior forecast uncertainty can be obtained without requiring history matching, indicating if the data is indeed informative on the prediction variables.