Designing an Infrasound-based Regional Volcano Monitoring, Classification and Forecasting for Anak Krakatau

Adi, Muhamad Haykal Hanif Gifari (2024) Designing an Infrasound-based Regional Volcano Monitoring, Classification and Forecasting for Anak Krakatau. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 02311940000154-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2026. Download (5MB) | Request a copy

Abstract

This study investigates the application of machine learning to the detection and clustering of seismo-acoustic signals for volcanic activity monitoring, focusing on the Anak Krakatau eruption. Utilizing the Adaptive F-Detector and Bartlett beamforming, we analyzed infrasound signals from the I06AU station to identify characteristics corresponding to different volcanic states. A Convolutional Autoencoder was employed for feature extraction, proving effective in encoding crucial seismic signatures. Deep Embedded Clustering (DEC) was applied to the reduced-dimensionality data, revealing distinct patterns significant for understanding volcanic activities. Our findings demonstrated that, upon iterative fine-tuning, the machine learning model adeptly captured both temporal and frequential features. These features are then utilized to classify different types of volcanic activity, showcasing the model's capability to discern between various states of volcanic unrest. Additionally, the incorporation of Time-to-event forecasting using Gradient Boosted Trees further enhanced our ability to predict eruptive activities, underscoring its potential as a cornerstone for early detection systems. Despite the inherent unpredictability of volcanic systems, the model achieved an appreciable level of accuracy, with a mean squared error (MSE) value of 22.167 for unsupervised pattern discovery, and in eruption forecasting, it delivered a mean absolute error (MAE) of 73.41 minutes and a root mean square error (RMSE) of 106.44, alongside an R2 of 75.4%. Although these results indicate the model's efficacy, they also point to the necessity for ongoing refinement to enhance its predictive precision.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	infrasound, seismo-acoustic, volcanic activity
Subjects:	G Geography. Anthropology. Recreation > GB Physical geography
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis
Depositing User:	Adi Muhamad Haykal Hanif Gifari
Date Deposited:	22 Feb 2024 05:46
Last Modified:	22 Feb 2024 05:46
URI:	http://repository.its.ac.id/id/eprint/107627

Actions (login required)

View Item