Modeling the thermal behavior of geothermal systems at Mount Meager, Southwestern Canada, using artificial neural networks and audio-magnetotellurics
Fateme Hormozzade Ghalati, Dariush Motazedian, James A. Craven, Stephen Edward Grasby, Victoria Tschirhart · Geophysics · 2025
ABSTRACT Precise modeling of subsurface temperatures is crucial for a comprehensive understanding and the exploitation of geothermal reservoirs. An artificial neural network method is used to estimate the subsurface temperature by analyzing 3D resistivity models derived from audio-magnetotelluric data and temperature logs from the Mount Meager Volcanic Complex (MMVC), southwestern British Columbia, Canada. A multilayer perceptron algorithm is used to capture the complexity of the data and estimate the subsurface temperature to a depth of 3 km. The model is trained on 70% of the 1160 data points, validated using the remaining 30%, and fine tuned based on data and error analysis. Subsequently, it is tested on three temperature logs that are not part of the training process, to ensure the robustness and reliability of the model predictions. The final model achieved a root mean square (rms) of 13.1°C (5% error) and an R2 value of 0.97 when estimating the subsurface temperature using the training data set, which is much more promising than using conventional analytical models that indicate an rms of 61%. The 3D temperature model of the MMVC is correlated with the available geologic data. This methodology offers a cost effective and noninvasive alternative for the thermal characterization of potential geothermal reserves, providing a powerful tool for resource development.