Stochastic Modelling of Multi-Channel Data for Imaging the Clay Cap, Temperature Distribution and Heat Output in the Wairakei-Tauhara

Abstract

In geothermal exploration, subsurface electrical resistivity models derived from magnetotelluric (MT) inversions have played a leading role in defining drilling targets during the early stages of projects. Resistivity models can indicate the depth and extent of a shallow conductor presumed to be reflective of a clay cap, a low permeable formation that lies on top of high-temperature geothermal reservoirs. When modeling MT data, gradient-based deterministic inversion has been the standard. However, for geothermal energy developers interested in knowing the clay cap boundary location, an undesirable aspect is the inherent non-uniqueness and the implied uncertainty that is not properly captured when inferring the conductor geometry from deterministic inversions. Also, when selecting a single model from a set of plausible models, such as the optimal solution derived from deterministic inversions, incorrect inference on conductor thickness and resistivity is likely being made. Consequently, operational decision-making – drilling, resource estimation – that relies on these could be compromised. In this situation, it is preferable to present the set of plausible models derived from a stochastic inversion that explicitly represents the irreducible uncertainty as estimated depth intervals for clay cap boundaries. The rock resistivity signature in geothermal fields is controlled by variations in multiple physical parameters like permeability, temperature, fluid salinity, and conductivity of the rock matrix, which has potentially been hydrothermally altered. So, the correct interpretation of resistivity models from MT inversions in geothermal fields requires comparison with direct observation in wells of key features as temperature, lithology, and clay content. Typically, comparisons between multi datasets are performed qualitatively, but the derived interpretations could be subjective. Joint inversion/modeling schemes of multi datasets offers a different perspective. They could provide new insights on rock properties correlations consistent with multiple sources of information. This thesis first develops a 1D MT Bayesian inversion that incorporates methylene blue data from wells (MeB), a clay content indicator, as a constraint. We use this lithological information to construct structural prior distributions for the MT inversion. The inversion is formulated in the context of geothermal fields to infer under uncertainty the shallow conductor boundary reflective of the clay cap. The inversion is tested on a synthetic example and then applied to the Wairakei-Tauhara geothermal field of New Zealand. The second part of this thesis develops a multi dataset stochastic modeling of the inferred clay cap by MT and MeB data along with well lithology and temperature logs to estimate temperature gradients, clay formation temperatures, and the heat output of Wairakei-Tauhara. Our goal in developing these methods is to provide a means to propagate uncertainty through the inverse problem so that the output is useful to decision-makers. Furthermore, by bringing in information obtained by direct observations in wells, we strengthen the connection between model and reality while reducing uncertainty. The methods could improve conceptual models of shallow geothermal hydrology, provide additional constraints for geothermal reservoir simulations, and support decision-making processes that define drilling targets or field management strategies. Further, by integrating multiple data sets, we generate a holistic image of the upper part of the magmatic-hydrothermal system that is consistent with the constituent parts each data type is informative of. This helps us understand the clay alteration distribution in geothermal fields, how they relate to temperature, and, by quantifying uncertainty, define appropriate limits on those inferences. Finally, we explore the inferred clay cap’s explicit inclusion to a TOUGH2 geothermal flow model, adding new constraints to reservoir simulations. The scheme developed serves as a proof of concept experiment of how prior structural information and geophysical inferences can be incorporated into reservoir simulations. Its application to more refined models will allow to construct better permeability models, improve the understanding of the clay cap and its influence on the geothermal reservoir simulations. Also, we apply our Bayesian inversion scheme to MT data and well salinity logs collected offshore for groundwater exploration. We show how the developed inversion scheme can be applied to other exploration fields where geophysical electromagnetic surveys and drill hole data are available. This constitutes another example of how the non-uniqueness arising through trade-offs between parameters in geophysical inversions can be managed by including informative priors based on well logs.