| Title | Bayesian Emulation Of Geothermal Numerical Models: A Synthetic Reservoir Case-Study |
|---|---|
| Authors | Vidal, Ariel and Archer, Rosalind |
| Year | 2014 |
| Conference | New Zealand Geothermal Workshop |
| Keywords | Bayesian emulators, numerical modelling, uncertainty. |
| Abstract | Complex numerical models are built in almost all fields of science as a means to simulate the behaviour of natural, real-world systems. These models are normally implemented as computer codes, which can take from fractions of seconds to several hours for a single run. The outputs of these models are generally used as a prediction of the real-world phenomena that are simulated by the model, but they will inevitably be faulty in some way. There will be uncertainty about how close the true quantities will be to the outputs of the numerical model and in the input parameters used for setting up the model. A formal sensitivity/uncertainty analysis for a particular numerical model may require thousands of model runs which could be impractical for complex models. In the last years there has been an increasing attention on the use of Bayesian methods to quantifying, analysing and managing uncertainty in the application of complex numerical models. Statistical emulators are being used to understand complex numerical codes and their parameter space in a wide variety of applications. Emulators based on Gaussian processes can inexpensively produce a reasonable representation of outcomes for a numerical model for a large set of potential input parameter settings without running the simulator itself, which may be valuable when the expense to run the simulator is high. In this paper a brief review of how a Gaussian process emulator works is presented together with some preliminary results on its implementation on a synthetic geothermal reservoir. |