DATA-SPACE INVERSION FOR EFFICIENT GEOTHERMAL RESERVOIR MODEL PREDICTIONS AND UNCERTAINTY QUANTIFICATION

Abstract

The ability to make accurate predictions with quantified uncertainty provides a crucial foundation for successfully managing a geothermal reservoir. The standard, state-of-the-art approach to delivering accurate predictions for a geothermal model requires two steps. First, the underlying uncertain parameters (such as subsurface permeability, strength, and location of deep upflows, etc.) are estimated (i.e., calibrated), as well as the associated posterior uncertainty based on measured field data. Second, this uncertainty is propagated from the parameters to the predictions using linear uncertainty analysis or full nonlinear model runs. In most cases, calibrating the unknown parameters accounts for most of the computational costs and time. However, these parameters, and their associated uncertainty, are not always of direct interest. Instead, we are often only interested in the predictions themselves.
The so-called data-space inversion (DSI) methodology provides a solution to this problem. It effectively bypasses the need to calibrate the unknown parameters and directly provides approximate posterior predictions, i.e., model predictions conditioned on the measured field data. Here we review the DSI framework and demonstrate its effectiveness and robustness on several geothermal problems. We find that the DSI approach can lead to significant savings in both computational resources and time, providing an attractive alternative to the more conventional calibration-based approach.