SEISMOLOGICAL MONITORING OF LANDSLIDES AND ASSESSMENT IN THE NORTH-WEST HIMALAYAS

Abstract

Predicting seismic signals has always provided significant interest to engineers and seismologists globally. This thesis presents a comprehensive investigation into the seismic monitoring of landslides and related geohazards in the North-West Himalayas, focusing on identifying seismic sources such as landslides, rockslides, and earthquakes through advanced signal analysis techniques. The research highlights the region's vulnerability, characterised by steep slopes and tectonic activity, and addresses the need for improved monitoring systems. The study introduces a multi-stage methodology, starting with the collection and analysis of seismic signals using Time-Frequency Analysis (TFA) techniques, including the Wavelet Synchrosqueezed Transform (WSST), Short-Time Fourier Transform (STFT), and Ensemble Empirical Mode Decomposition (EEMD). The WSST technique provides superior localization and accuracy in identifying dominant frequencies, which is critical for differentiating seismic events. Following noise filtering, statistical and spectral features are extracted and used to classify seismic sources through machine learning models, specifically Random Forest (RF) classifiers. The classifier is trained and validated using a dataset of 175 landslide signals, in combination with earthquake data from the SNATR, demonstrating high accuracy in distinguishing between landslide and earthquake signals. The research also investigates the potential for extending early warning systems based on the findings, suggesting that seismic networks can play a crucial role in real-time hazard detection and disaster risk reduction. Although the dataset for landslide events is limited, the study provides a valuable framework for implementing real-time monitoring and early warning systems in seismically active and topographically challenging regions such as the Himalayas. The thesis highlights the importance of sophisticated seismic monitoring and machine learning methodologies in improving early warning systems for landslides. The study highlights the necessity for enhancing the established approaches, especially by incorporating real-time monitoring data from various seismic networks and diverse geographical areas. Furthermore, the study underscores the need to augment the dataset to encompass a broader range of seismic occurrences, enhancing forecast accuracy. The deployment of these advanced models in operational early warning systems could markedly improve catastrophe preparedness in seismically active areas, with possible applications extending beyond the Himalayas. The study advocates for the investigation of novel features and methods, including AI-driven predictive modelling, to enhance the system's reliability. These developments offer significant potential for creating a resilient, real-time early warning system for landslides and other geohazards, delivering prompt, actionable insights for disaster risk mitigation.