Transdimensional inversion of borehole temperature profiles from the Waiwera geothermal reservoir with discrete fracture network models
Abstract
The shallow and low temperature hydrothermal system of Waiwera (New Zealand) is situated in a complex geological location in proximity to the sea. Geothermal fluid of approximately 50°C is feeding into the well fractured reservoir from below with a strong vertical flow component. This inflow prevents seawater from entering the aquifer. Flow and mixing of fresh, geothermal and marine water form a steady-state temperature anomaly with an extent of 500 m in diameter from the surface down to approximately 400 m. The flow system is governed by stratified sandstone which has been tilted, folded, faulted and fractured providing the pathways. The site has been the subject of multiple research studies, and the large amount of measurements makes it an ideal candidate to test and compare different modelling and inversion methodologies. Existing models mainly used the continuity approach to represent the aquifer, neglecting the effect of local discrete features, faults and discontinuities. In this study we pursue to model the geothermal system using discrete fracture network models (DFN).
Advantage of using DFNs to interpret geological observations is the direct incorporation of geostatistical information. DFNs could be the basis for deterministic simulations, and can be used to build efficient forward models - an essential condition of stochastic inversion. However, disadvantage of DFNs is, that they require a predefined number of parameters, which is a strong prior assumption when limited information is available. A solution to overcome this challenge is to use transdimensional inversion, a data-driven approach that varies the number of parameters during interpretation. We use the reversible-jump Markov Chain Monte Carlo (rjMCMC), which is well-established and widely used. rjMCMC is an iterative algorithm, that allows to introduce or to remove model parameters during the inversion process. The algorithm provides an ensemble as result, a set of model realizations that represent the posterior distribution of the inverse problem. The ensemble can be further processed to extract statistical properties, or to provide a visualization in the form of a fracture probability map. We use the rjMCMC DFN inversion algorithm to provide a better fit to the measured borehole temperature profiles than previous models were able to.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFM.H53M1760S
- Keywords:
-
- 1829 Groundwater hydrology;
- HYDROLOGYDE: 1835 Hydrogeophysics;
- HYDROLOGYDE: 1873 Uncertainty assessment;
- HYDROLOGYDE: 3260 Inverse theory;
- MATHEMATICAL GEOPHYSICS