Mapping eruption affected area using Sentinel-2A imagery and machine learning techniques

Abstract

Volcanic eruptions are natural disasters with significant environmental and societal impacts. Timely detection and monitoring of volcanic eruptions are crucial for effective hazard assessment, mitigation strategies, and emergency response planning. Remote sensing technology has emerged as a valuable tool for detecting and assessing the effects of volcanic eruptions. One of the challenges in remote sensing image processing is handling large data dimensions that are difficult to address using traditional methods. Machine learning approaches offer a suitable solution to tackle these challenges. Machine learning demonstrates increasing computational capabilities, the ability to handle big data and automation. This study aimed to compare different machine learning classification algorithms, including Random Forest (RF), Support Vector Machine (SVM), Gaussian Mixture Model (GMM), and K-Nearest Neighbors (KNN). The data utilized in this study was derived from Sentinel-2A Multi-Spectral Instrument (MSI) imagery, which was tested in areas affected by the eruption of Mount Agung, Bali Province, in 2017. The results indicated that the GMM algorithm performed the best among the machine learning classifiers, achieving an Overall Accuracy (OA) value of 82.04%. It was followed by RF (78.86%) and KNN (77.55%). The areas affected by volcanic eruptions were determined by overlaying disaster-prone regions with areas mapped using the machine learning approach. The total affected area was measured as 29.89 km<sup>2</sup>, with an additional 3.31 km<sup>2</sup> outside the designated zone. The findings of this study serve as a guideline for governmental entities, stakeholders, and communities to implement effective mitigation efforts for disaster risk reduction.