Quantifying uncertainty in production forecasts is critical to making good reservoir management decisions, particularly for many current investment opportunities that require intensive technology and large investments, and that may have marginal profitability indicators. Reservoir studies are conducted to support decision making, but reservoir management decisions must often be made before completion of these studies. This paper presents a new approach to reservoir studies that combines production forecasting with history matching. The approach provides preliminary production forecasts much earlier in reservoir studies. More importantly, the approach provides estimates of uncertainty associated with the forecasts. This is accomplished by using the mismatch of history match runs to weight corresponding forecast runs.We illustrate application of the method to the 8-Sand reservoir in the Green Canyon 18 field, Gulf of Mexico. We observed that, as the accuracy of the model increased during the history match, the uncertainty of forecasted reserves decreased and the distribution of reserves stabilized.Early forecasts and associated estimates of uncertainty provided by our new method can be quite valuable to management in making investment decisions. 446Alvarado et al.

Despite recent advances in uncertainty quantification, the petroleum industry continues to underestimate the uncertainties associated with reservoir production forecasts. This paper describes a calibration process that can improve quantification of uncertainties associated with reservoir performance prediction. Existing methods underestimate uncertainty because they fail to account for all, and particularly unknown, factors affecting reservoir performance and because they do not investigate all combinations of reservoir parameter values. However, the primary limitation of existing methods is that their reliability cannot be verified because the testing of an estimate of uncertainty from existing methods yields only one sample for what is inherently a statistical result. Verification and improvement of uncertainty estimates can be achieved with calibration – comparison of actual performance with previous uncertainty estimates and then using the results to scale subsequent uncertainty estimates. Calibration of uncertainty estimates can be achieved with a more frequent, if not continuous, process of data acquisition, model calibration, model prediction and uncertainty assessment, similar to the process employed in weather forecasting. Improved ability to quantify production forecast uncertainty should result in better investment decision making and, ultimately, increased profitability.