Subsurface temperature estimation in a geothermal field based on audio-frequency magnetotelluric data

In geothermal exploration, the subsurface temperature distribution is an important parameter since it is closely related to the existing geothermal system. However, subsurface temperature information can only be obtained from boreholes, requiring a considerable cost and a limited number. A new geothermometry technique based on the relationship between temperature and resistivity has been proposed to overcome this difficulty. In order to enhance the reliability of this method, the subsurface heat condition in a different geological and geographical location was reconstructed. The temperature up to some depth beneath the surface could be estimated incorporating the back-propagation neural network using resistivity determined by high-resolution audio-magnetotelluric soundings. The temperature profiles resulting from the neural network fit with those observed from nearby boreholes, providing a satisfying evaluation result of the model’s predictive powers. Joint interpretation was then carried out following the one-dimensional resistivity inversion result and temperature cross-section. Both resistivity and temperature anomalies show an excellent agreement, so that subsurface anomaly such as altered layers, reservoir zone, and some possible faults was detected. With an appropriate training strategy, this technique can help to significantly reduce the costs at estimating the subsurface temperature.