The cinnabar (±stibnite) deposits of the Mt. Amiata geothermal system and the associated hot springs and gas vents, occur along a N-S directed, narrow longitude region. In this study, we combine a geological and geophysical dataset gathered from the early stages of geothermal exploration of the district with a multivariate statistical analysis carried out by Machine Learning (ML) algorithms to highlight possible correlations between the distribution of the geothermal expressions of Mt. Amiata and its geological/structural features. We used 5 distinct ML supervised models (Ordinary Least Squares Linear Regressor, Multilayer Perceptron Regressor, Support Vector Regressor, CatBoost, and Random Forest) to determine which set of geological or geochemical features of the dataset reproduces the distribution of the geothermal expressions of the area with sufficient accuracy. The regressors CatBoost and Random Forest, which use decision trees for probability calculations, are the most efficient in predicting the narrow-longitude distribution of the geothermal expressions of Mt. Amiata. Also, the only combination of predictors generating probability maps that accurately reproduce the distribution of the geothermal expressions is the one considering permeability, Hg solubility, T, and distances from faults and folds. This shows that only a combination of geological/geochemical factors can explain the peculiar regional distribution.
Large-scale structural controls on hot spring mineral deposits of geothermal systems (Mt. Amiata, Italy) highlighted by machine learning algorithms?
Farella, Elisa Mariarosaria;Rigon, Simone;Remondino, Fabio;
2023-01-01
Abstract
The cinnabar (±stibnite) deposits of the Mt. Amiata geothermal system and the associated hot springs and gas vents, occur along a N-S directed, narrow longitude region. In this study, we combine a geological and geophysical dataset gathered from the early stages of geothermal exploration of the district with a multivariate statistical analysis carried out by Machine Learning (ML) algorithms to highlight possible correlations between the distribution of the geothermal expressions of Mt. Amiata and its geological/structural features. We used 5 distinct ML supervised models (Ordinary Least Squares Linear Regressor, Multilayer Perceptron Regressor, Support Vector Regressor, CatBoost, and Random Forest) to determine which set of geological or geochemical features of the dataset reproduces the distribution of the geothermal expressions of the area with sufficient accuracy. The regressors CatBoost and Random Forest, which use decision trees for probability calculations, are the most efficient in predicting the narrow-longitude distribution of the geothermal expressions of Mt. Amiata. Also, the only combination of predictors generating probability maps that accurately reproduce the distribution of the geothermal expressions is the one considering permeability, Hg solubility, T, and distances from faults and folds. This shows that only a combination of geological/geochemical factors can explain the peculiar regional distribution.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.