| Title | INVESTIGATION OF PARAMETER UNCERTAINTY FOR AN IDEALIZED GEOTHERMAL MODEL USING LINEAR ANALYSIS |
|---|---|
| Authors | J. Omagbon, M. OSullivan, J. OSullivan, C. Walker and E. Bjarkason |
| Year | 2017 |
| Conference | New Zealand Geothermal Workshop |
| Keywords | Sensitivity analysis, parameter estimation, inverse modelling, uncertainty analysis, reservoir modelling, PEST, linear uncertainty analysis |
| Abstract | The parameters of a geothermal model are inherently uncertain. Model parameters are usually estimated by calibrating the model against observations. However, the information contained in the data is typically insufficient for determining the parameters uniquely. Therefore, parameter uncertainty still remains after calibrating a model. The present study, considers linear analysis as a way of quantifying the uncertainty of common geothermal parameters. A synthetic 2D slice model was constructed as test a case for this study. We also considered observation types that are typically used during calibration. Linear analysis can be used to estimate the posterior covariance matrix that contains the uncertainty information for the model parameters in a linear model. The main disadvantage of linear analysis is the fact that geothermal models are highly nonlinear. This was considered somewhat indirectly by looking at the variation in the model outputs with varying the parameters. The slopes of the relevant plots give the elements of the Jacobian matrix. This has enabled us determine the extent at which the model nonlinearity compromises the results obtained from linear analysis. We used a quantity called “relative parameter uncertainty reduction” and the correlation coefficient between pairs of parameters as a measure of how well the parameters were constrained by the calibration dataset. Both of these are easily obtained from the posterior covariance matrix. The results obtained from this study suggest that there were parameters which were highly correlated with other parameters despite the simple rock-type distribution used in the model. By mapping out the values of the relative parameter uncertainty reduction, we were also able to infer areas of the model where the observations constrain the permeability and porosity values. |