Author ORCID Identifier
https://orcid.org/0000-0002-5421-9206
Biosketch
I am an interdisciplinary Earth science researcher with expertise in earthquake forecasting, geophysical data analysis, and machine learning. I hold a PhD in Physics from SASTRA University, where my research focused on multi-precursor earthquake modelling in the Himalayan belt. My research experience includes serving as a CSIR Senior Research Fellow and MoES Junior Research Fellow, specialising in spatio-temporal analysis and anomaly detection. I have published in leading journals and conferences and possess technical proficiency in Python, MATLAB, ArcGIS, and advanced laboratory instrumentation. Additionally, I have completed multiple government and academic projects in seismic hazard analysis and contributed to teaching undergraduate and graduate-level courses.
Date of Award
23-7-2025
Document Type
Thesis
School
School of Electrical & Electroncis Engineering
Programme
Ph.D.-Doctoral of Philosophy
First Advisor
Dr.N.Venkatanathan
Keywords
Earthquake Forecasting, Solid Earth Tide, Micro seismicity, Outgoing Longwave Radiation, Machine Learning
Abstract
Understanding and forecasting earthquakes is not just about science it’s about saving lives, even as the dynamic and hidden nature of Earth's tectonic processes makes this an intricate challenge. The Himalayan belt is one of the most seismically active regions globally due to the collision of the Indian and Eurasian tectonic plates. Investigating this region is essential for understanding highly complicated tectonic mechanisms and reducing substantial hazards posed to millions of people by persistent and intense seismic events. This study aims to enhance forecasting methodologies by applying machine learning techniques while dealing with the inherent challenges of geological complexity and data limitations.
Our research explores diverse parameters, such as solid earth tides (SET), outgoing longwave radiation (OLR), and Min-Light quakes, by studying their role as potential precursors to seismic activity. By integrating these precursory signatures, we develop an advanced earthquake forecasting model using machine learning techniques to improve forecasting accuracy and resilience. In this study, we utilize precursor datasets spanning the temporal extent from 1993 to 2024 based on the lunar phase. To analyze solid earth tide (SET) data, singular spectral analysis (SSA) has been applied to identify irregularities in the SET. Rolling window approach is implemented to handle periodic and noisy components in the decomposition to ensure robust analysis and pattern detection.
The study finds that irregularities in SET, particularly variations in the sixth EF component, may serve as reliable indicators for long-term earthquake forecasting. While the stress induced by Solid Earth Tides is significantly lower than that from tectonic plate motion, EF analysis of the SET time series remains a vital indicator of its inherent characteristics. Further, the possible vulnerable seismic nucleation zone and the possible depth and magnitude of triggering in mainshocks are identified through cluster analysis.
The OPTICS clustering algorithm was applied to Min-Light quake data, identifying distinct seismic clusters and vulnerable zones. These clusters align closely with the epicenters of major earthquakes, emphasizing their importance for seismic hazard assessment. The OLR observations give us insight into how atmospheric parameters are affected by seismic activity in an area. There was a correlation between the magnitude of the earthquake and the distinct features extracted from the analysis of Min-Light quakes and OLR.
An ensemble model has been developed by integrating spatio-temporal analysis of solid earth tides and Min-Light quakes with implications for earthquake forecasting. The models demonstrated near-perfect success in forecasting impending earthquake’s latitude (99%) but showed moderate success for “Earthquake_Occurrence” (60%) and longitude (50%). Forecasting the Depth of the impending earthquake is the most challenging one, with a low success rate of 40%, reflecting complexities in modelling subsurface processes. The findings emphasize the performance of ensemble models, especially in elucidating complex linkages and enhancing accuracy in forecasting for geophysical forecasting.
Recommended Citation
M, Senthil Kumar Mr, "Study of Anomalous Variations in Seismic and Non-Seismic Parameters Using Machine Learning Techniques to Develop the Earthquake Forecasting Model in The Himalayan Belt" (2025). Theses and Dissertations. 131.
https://knowledgeconnect.sastra.edu/theses/131