Predicting Electrical Resistivity in Hydrothermal and Natural Degassing Geological Systems through petrophysical and thermodynamic data: a machine learning approach

Hydrothermal and natural degassing geological systems present various hazards. Monitoring them is crucial to understanding their behavior, assessing risks comprehensively, and mitigating potential impacts on both the environment and human safety. Electrical resistivity, which is closely related to water content, gas content, and fluid temperature, is a key parameter for studying these systems. However, existing mathematical relationships, such as Archie's law, have limitations, particularly in their applicability to a wide range of petrophysical and thermodynamic properties. Linking the observed variations in measured resistivity to variations in the dynamics of the hydrothermal or natural degassing system under investigation is not straightforward.

 

The aim of this study is to establish a numerical relationship between petrophysical and thermodynamic input variables and resistivity data obtained from geoelectrical field surveys. This numerical relationship could predict changes in the electrical resistivity distribution based on variations in simulated petrophysical and thermodynamic values over time. Comparison between predicted and field resistivity data would ultimately validate the current dynamic state of the system, providing a powerful monitoring tool.

 

To this end, two 3D petrophysical and thermodynamic numerical models for two natural degassing systems were constructed by 3D electrical resistivity tomography surveys using constraints derived from different types of data (e.g., geological, geochemical and/or hydrogeological data). The models were validated through the comparison of predicted temperature, pressure, and gas flow distributions with field survey data. We then trained a Random Forest algorithm to predict the resistivity values for each cell of the models using the petrophysical and thermodynamic parameters of each cell as input and the field resistivity values as the target variable.

 

The results obtained for both models on the test data demonstrate the effectiveness of the Random Forest algorithm in successfully predicting resistivity values. This predictive capability, which allows adjustments to the system’s petrophysical and thermodynamic parameters until the predicted resistivity aligns with newly observed values, could shed light on the ongoing dynamics within the system, thereby enhancing its understanding through geophysical monitoring. The developed methodology could be a powerful addition to resistivity monitoring in active geological systems.