| Title | Distributed Bayesian Geophysical Inversions |
|---|---|
| Authors | Lachlan McCALMAN, Simon O'CALLAGHAN, Alistair REID, Darren SHEN, Simon CARTER, Lars KRIEGER, Graeme BEARDSMORE, Edwin BONILLA, , Fabio RAMOS |
| Year | 2014 |
| Conference | Stanford Geothermal Workshop |
| Keywords | exploration, Bayesian inference, inversion, software, drilling risk reduction, magnetotellurics |
| Abstract | The quantification of uncertainty in exploration for geothermal targets is central to determining the best locations for drilling. We develop algorithms to perform data fusion and joint inversions from a broad range of independent geophysics surveys to infer rock properties such as density, magnetic susceptibility, seismic velocity, resistivity, thermal conductivity and temperature. Our methods seamlessly integrate surveys including geology, gravity, magnetics, seismic reflection, magnetotellurics and borehole geophysics to predict the most relevant properties for predicting geothermal drilling success. As part of a probabilistic inference procedure, we quantify the uncertainty for each quantity of interest, providing principled strategies for risk minimisation in drilling projects. While developed initially for geothermal exploration in Australia, the methodology is readily adaptable for other geological settings and geothermal drilling targets. In this work we develop an inference procedure based on parallel tempering Monte Carlo techniques that is well suited to large-scale computation infrastructures such as cloud computing. This enables our probabilistic approach to the inversion problem not only to quantify the risk associated with drilling, but also guide strategies for further pre-drilling exploration, increasing drill target confidence. The Bayesian formulation allows expert knowledge to be added in the form of prior distributions, making explicit the contribution of human judgment to the final result. We demonstrate our approach in three experiments: 1) a joint magnetics and gravity inversion to infer locations of buried granite outcrops, 2) a joint magnetotelluric and gravity inversion to illustrate the benefits of data fusion and; 3) a performance test comparing our cluster-based parallel inference algorithm to traditional methods. |