Geothermal reservoir temperature prediction using hydrogeochemical data of Northern Morocco: A machine learning approach

Abstract

Geothermal energy exploration depends on accurate estimation of reservoir temperatures. However, conventional methods are complex, costly and uncertain, especially those based on indirect measurements and assumptions. A dataset of 99 sets of hydrogeochemical and reservoir temperature data was created and five machine learning (ML) algorithms including decision tree regression (DTR), extreme gradient boosting (XGBoost), extremely randomised trees (XRT), natural gradient boosting (NGB) and deep neural network (DNN) were applied to address the issue. The models' predictive accuracy and generalisation potential in northern Morocco were evaluated by essential performance metrics including mean absolute error (MAE), root mean square error (RMSE) and coefficient of determination (R²). The XGBoost model proved superior with the highest R² of 0.9967 and the lowest MAE and RMSE of 0.7046 and 0.9992 respectively. Further, this study utilises Shapley additive explanation (SHAP) as an explainable technique to evaluate XGBoost predictive decisions. SHAP interpreted that $\text{Si}{O}_2$ is the most important variable in predicting reservoir temperature. This study highlights the potential of ML for accurate reservoir temperature prediction, offering a reliable tool for model selection and advancing understanding of geothermal resources.