Estimation of near-surface temperature in Suwawa Geothermal Prospect, Gorontalo, Sulawesi, Indonesia, based on magnetotelluric and artificial neural network

A geophysical survey using broadband magnetotelluric (MT) technology was carried out in Suwawa Geothermal Prospect Area, Gorontalo Province, Sulawesi Island, Indonesia. The target of that research is to evaluate the geothermal potential hidden below the surface, based on underground resistivity distribution. However, the information about resistivity alone is not enough to get a proper understanding of the geothermal system in this area. Another important subsurface feature that could be useful for the evaluation is temperature. In this study, an attempt to predict the subsurface temperature using resistivity and limited information from a shallow borehole thermogram was carried out. Employing the dependency between resistivity and temperature an indirect temperature estimator was built, thanks to the applicability of artificial neural network (ANN) to learn the pattern connecting both parameters. Comparing some neural network training data shows that the predictive powers of the calibrated neural network highly influenced by the geological difference between the location of borehole and MT station. The best trained ANN was then used to predict the temperature below the other MT stations. The result shows that a proper ANN architecture is important to improve the deeper temperature estimation. The best ANN estimator was obtained from the BT01 and AMT39 data pair, which has the highest correlation as well. This preliminary study gives useful insight into how resistivity could be an alternative tool to delineate the near-surface temperature profile, in order to get a more comprehensive image of the subsurface geothermal system.