Real-Time Estimation of Volcanic Tremor Depth Using Histogram Gradient Boosting Regressor

Authors

1 Department of Physics, Na.C., Islamic Azad University, Najafabad, Iran

2 Instituto Nazionale di Geofisica e Vulcanologia, Osservatorio Etneo, Catania, Italy

Abstract

Volcanic tremors are crucial indicators of volcanic activity and hazard assessment. This paper presents a new approach to tremor depth calculation, emphasizing automation, precision, and real-time capabilities. The method utilizes Root Mean Squared (RMS) values of the tremor signals to train the Histogram Gradient Boosting Regressor (HGBR). Generalizing the trained model allows for real-time estimation of tremor depth, leading to quicker and more reliable hazard analysis. In this research we used 39,790 tremor RMS samples reported from 17 stations located around Etna volcano, Catania ,Italy. The importance of selecting the real data from this location is that Etna is the biggest active volcano between all European countries and also it is a well monitored volcano among all active volcanos around the world. There are several monitoring stations around this active volcano and there are various conditions of its activities including quiet, Stromboli and paroxysm. The investigated tremor data was recorded in the time span between May of 2011 and November of 2014. Model performance was assessed with various metrics, achieving a Mean Absolute Error (MAE) of 139.33 meters for the unseen test dataset. The results of the model showed a robust correlation with R-squared values of 0.850, affirming its regression capability during various volcanic phases. Statistical evaluation revealed that for the expected error, there's a 68% chance the error lies within 207 m, a 95% chance it is bounded to 414 m, and a 99% chance it is less than 621 m.

Keywords