khalili, Mohammad Amin (2025) Integrated Satellite System for Earth Observation and Mitigating Natural and Anthropogenic Phenomena Using Artificial Intelligence Applications. [Tesi di dottorato]

Documento PDF Final_Thesis_Phd_Khalili.pdf Visibile a [TBR] Amministratori dell'archivio Download (26MB) | Richiedi una copia

Abstract

Monitoring and predicting natural and anthropogenic geohazards are crucial for safeguarding infrastructure, enhancing disaster resilience, and supporting sustainable development. This research integrates advanced Synthetic Aperture Radar (SAR) technologies with Artificial Intelligence (AI) methodologies to address the complexities of geohazard monitoring and prediction. The study focuses on monostatic, bistatic, and multistatic SAR configurations, evaluating their potential to capture deformation across diverse geological and environmental contexts. Additionally, this work explores AI-based models to enhance interpretability, prediction accuracy, and computational efficiency of SAR data analysis, addressing limitations in traditional geohazard monitoring and prediction systems. Monostatic SAR systems, including Sentinel-1A/B and COSMO-SkyMed, are analyzed for their effectiveness in urban and rural environments using Persistent Scatterer Interferometry (PSI) and Small Baseline Subset (SBAS) techniques. The study highlights the advantages of high coherence and stability in monostatic configurations for capturing subtle surface deformations. These systems excel in providing reliable deformation data for urban areas with consistent reflectors, such as buildings and infrastructure. Bistatic and multistatic SAR systems, still under development and explored through simulation and the RODiO mission as part of the ALCOR program under the Italian Space Agency, are evaluated for their capability to address limitations in spatial coverage and multi-angle observations. These configurations are particularly beneficial in complex terrains such as mountainous and rural regions, where traditional monostatic systems face limitations due to temporal decorrelation and sparse stable reflectors. AI methodologies play a pivotal role in this research, enabling advanced analysis of SAR-derived data. Convolutional Neural Networks (CNNs) are employed for spatial pattern recognition, identifying hotspots of deformation with high precision. Hybrid models such as CNN-LSTM (Long-Short Term Memory), Graph Convolutional Networks (GCN)-LSTM, and stack algorithms are developed to integrate spatial and temporal dimensions, allowing for accurate predictions of displacement dynamics over time. GCNs provide a novel approach to analyzing spatial dependencies in deformation data, enabling a better understanding of complex geophysical phenomena, particularly in urban and environmentally sensitive regions. Case studies across diverse terrains—from landslides in Moio Della Civitella caused by excessive groundwater extraction to landslide-prone areas such as Randazzo in Italy—validate the efficacy of these AI-driven approaches. The results demonstrate the adaptability, robustness, and high precision of the models developed. The integration of environmental and anthropogenic factors is a key contribution to this thesis. Variables such as precipitation, vegetation indices (e.g., NDVI), soil moisture, predisposing factors, and geological characteristics are incorporated into SAR data analysis to enhance model accuracy and provide a multi-dimensional perspective on geohazard susceptibility. This holistic approach ensures that the models account for the interplay of natural and human-induced factors in deformation processes. For instance, the analysis of rainfall intensity and its correlation with landslides, or the impact of urban expansion and groundwater depletion on subsidence patterns, underscores the importance of integrating environmental data into SAR-AI frameworks. Such comprehensive assessments contribute to both scientific understanding and practical applications in disaster risk management and infrastructure resilience. Despite significant advancements, the research acknowledges several limitations. Real-world datasets for bistatic and multistatic SAR configurations were unavailable, restricting the analysis to simulations and conceptual studies. Limited access to high-resolution COSMO-SkyMed images, except for study areas in Italy, and the reliance on Sentinel-1 imagery posed challenges in achieving finer spatial resolutions. Additionally, the availability of ground truth data, such as GNSS station measurements and inclinometer readings, was limited, making validation of certain models challenging. The findings of this research offer a robust framework for integrating SAR technologies with AI models, advancing the geohazard monitoring and prediction field. The study demonstrates the practical implications of these technologies for infrastructure management, early warning systems, and sustainable urban planning. By bridging theoretical advancements with real-world applications, this work contributes to developing resilient systems capable of addressing global challenges posed by natural and anthropogenic phenomena. For example, SAR-AI models developed in this research can inform policymakers about high-risk areas, enabling proactive mitigation strategies and resource allocation for disaster preparedness. In conclusion, this thesis establishes a comprehensive approach to geohazard monitoring, leveraging the unique capabilities of SAR configurations and the transformative potential of AI. The research highlights the importance of interdisciplinary collaboration in addressing complex environmental challenges and provides a roadmap for future Earth observation and disaster management innovations. By addressing current limitations and exploring emerging SAR technologies such as polarimetric and hyperspectral SAR, alongside advancements in AI models, the methodologies developed in this thesis have the potential to revolutionize geohazard monitoring and prediction, ultimately contributing to infrastructure resilience, community safety, and sustainable development.

Downloads

Actions (login required)

Modifica documento