| Title | Stochastic Joint Inversion of a Geothermal Prospect |
|---|---|
| Authors | R. MELLORS, A. RAMIREZ, A. TOMPSON, M. CHEN, X. YANG, K. DYER, J. WAGONER, W. FOXALL, and W. TRAINOR-GUITTON |
| Year | 2013 |
| Conference | Stanford Geothermal Workshop |
| Keywords | joint inversion, exploration, flow modeling, geophysics |
| Abstract | We are developing a stochastic inverse algorithm to jointly analyze multiple geophysical and hydrological datasets for a geothermal prospect. The purpose is to improve prospect evaluation and estimate the likelihood of useful temperature and flow fields at depth. We combine Bayesian inference with a Markov Chain Monte Carlo (MCMC) global search to conduct a staged inversion of the different data sets. The results consist of a detailed description of the uncertainty in the solution as well as a suite of alternative geothermal reservoir models. The method is highly flexible and capable of accommodating multiple and diverse datasets. Currently, we are seeking to match observed temperature profiles at specific wells, geophysical observations such as electrical resistivity, and particular elements of the geologic structure. An a priori structural model is used as a starting point but is allowed to vary during the inversion process. Material properties such as permeability, porosity, and heat capacity are also allowed to vary in the process, although far-field pressure (P) and temperature (T) boundary conditions are held fixed. The algorithm searches a solution space defined by possible models and parametric values using the MCMC approach. The first stage runs a hydrothermal flow model using the NUFT (Nonisothermal, Unsaturated Flow and Transport) code in which temperatures and fluid and heat flows are predicted and compared to observations. In the second stage, electrical resistivity is predicted and compared to observations. The algorithm has been tested on a synthetic model with good results. The model is based on the Superstition Mountain area in Southern California. Temperature profiles from three exploratory wells along with detailed geologic information are available. The provisional simulation domain was discretized by a 3D mesh using fixed P, T boundary conditions. Steady state MCMC reservoir simulations are generated by populating the structural and parametric characteristics (e.g., fault dimensions, permeability, porosity, and heat capacity) in the domain and running the model to equilibrium. Typically, thousand of runs are required, presenting a significant computational challenge. In the future, we plan to integrate additional geophysical data such as magnetotellurics (MT) and gravity. The flexibility of the approach allows the potential inclusion of other data types such as geochemical signatures and geostatistical-based models of geologic structure. We envision the potential use of the algorithm as a method to generate a range of possible models and corresponding likelihoods to estimate uncertainty associated with a prospect. The initial mesh and model generation is developed to be compatible with commercially available geological modeling packages. This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. |