Ravellino, Fabiana (2025) Hyperspectral satellite remote sensing in the context of climate change: potential land applications and experimental verification. [Tesi di dottorato]

	Documento PDF Ravellino_Fabiana_38_COMPLETA.pdf Visibile a [TBR] Amministratori dell'archivio Download (69MB) | Richiedi una copia
	Documento PDF Ravellino_Fabiana_38_PARZIALE.pdf Visibile a [TBR] Amministratori dell'archivio Download (44MB) | Richiedi una copia

Abstract

Over the past decade, spaceborne hyperspectral technology has experienced rapid development, offering unprecedented opportunities for Earth observation. Hyperspectral sensors capture data in hundreds of narrow, contiguous spectral bands, enabling the detection of subtle physical and chemical properties that are not accessible through traditional multispectral imagery. Despite these advantages, the high dimensionality and complexity of hyperspectral data present significant challenges for preprocessing, analysis, and modeling. This thesis addresses these challenges by investigating preprocessing strategies tailored to hyperspectral data, including cloud and shadow masking, coregistration, pansharpening, spectral correction, and dimensionality reduction. Various band selection techniques, including supervised and unsupervised methods, are evaluated to handle high dimensionality and improve model performance more effectively. The developed methodologies are applied to two problems relevant to climate change. The first focuses on soil organic carbon (SOC), combining laboratory measurements and multispectral and hyperspectral remote sensing data to assess SOC content at regional and national scales. Results demonstrate the importance of accounting for soil moisture, texture, and dataset size in predictive modeling, highlighting the potential of hyperspectral data for large-scale carbon monitoring. The second case study addresses wildfire risk assessment through fuel load mapping, demonstrating the feasibility of generating automated, high-resolution fuel maps using hyperspectral data and machine learning. Limitations related to data acquisition frequency are discussed, along with future prospects enabled by upcoming missions such as CHIME. Overall, this work demonstrates that hyperspectral remote sensing, combined with machine learning and advanced preprocessing techniques, can provide valuable insights for environmental monitoring, carbon management, and wildfire risk assessment. The development of accessible data processing libraries and dimensionality reduction tools will be essential to enable broader operational use of hyperspectral imagery in Earth observation and land management applications.

Downloads

Actions (login required)

Modifica documento