This Ph.D. thesis highlights the necessity to propose a framework that blends the precision of Geomatics, the global vision of Remote Sensing, and the predictive power of GeoAI. The research activity has been based upon three fundamental pillars. The first pillar explores how to map vulnerability on a large scale using open-source computing platforms like Google Earth Engine to overcome the computing effort due to a lot of data for long time series analysis. By focusing on the Cultural Heritage sites of Hue City, Vietnam, it is proven that open-access data (e.g. Landsat imagery and Copernicus products) can be effective for protecting heritage in areas where local information is limited. To overcome these limitations and to harmonize and validate ground-based data with satellite data, the second pillar moves directly on field. Satellite thermal data of different satellite platforms have been compared with data acquired by sensors installed on the ground. The third pillar represents the GeoAI models application in environmental risk assessment. Through the use of models such as U-Net and YOLO, complex imagery was transformed into instantaneous maps of buildings and roads, serving as effective tools during emergencies or post-disaster phases. Furthermore, the research investigated non-invasive techniques for monitoring soil health and showcased the power of Augmented Reality in visualizing invisible threats, such as air pollution, effectively bringing scientific data directly to the public. This thesis can be seen as a practical contribution according to Destination Earth (DestinE) platform, a dynamic digital ecosystem designed to create high-precision replicas of Earth’s systems (Earth Digital Twins). By integrating multi-source geospatial data and high-performance computing, it enables the transition from historical observation to predictive and real-time urban management.
Questa tesi di dottorato evidenzia la necessità di proporre un framework che integri la precisione della Geomatica, la visione globale del Remote Sensing e il potere predittivo della GeoAI. L’attività di ricerca è stata basata su tre pilastri fondamentali. Il primo pilastro esplora come mappare la vulnerabilità su vasta scala utilizzando piattaforme di calcolo open-source, come Google Earth Engine, per superare lo sforzo computazionale richiesto dall’analisi di grandi serie temporali di dati. Focalizzandosi sui siti del Patrimonio Culturale di Hue City, in Vietnam, viene dimostrato come i dati open-source (come immagini Landsat e prodotti Copernicus) possano rappresentare una valida risorsa per la protezione del patrimonio in aree dove le informazioni locali sono limitate. Per superare i limiti intrinseci dei dati satellitari e per armonizzare e validare i dati misurati a terra, il secondo pilastro si sposta direttamente sul campo. I dati termici di diverse piattaforme satellitari sono stati confrontati con i dati acquisiti mediante sensori a terra. Il terzo pilastro riguarda l’applicazione di modelli di GeoAI per la valutazione del rischio ambientale. Attraverso l’uso di modelli come U-Net e YOLO, immagini complesse sono state trasformate in mappe istantanee di edifici e strade, risultando strumenti efficaci durante le emergenze o nelle fasi post-disastro. È stato inoltre esplorato come “leggere” la salute del suolo in modo non invasivo mediante realtà aumentata, come nel caso dell’inquinamento atmosferico, portando la scienza direttamente nelle mani dei cittadini. Questa tesi può essere considerata un contributo pratico in linea con l’iniziativa Destination Earth (DestinE), un ecosistema digitale dinamico progettato per creare repliche ad alta precisione dei sistemi terrestri (Digital Twins). Integrando dati geospaziali multi-sorgente e calcolo ad alte prestazioni, essa consente il passaggio dall’osservazione storica a una gestione urbana predittiva e in tempo reale.
Multi-Source Remote Sensing and GeoAI Methodologies for Enhanced Environmental Risk Assessment / Sanita', M.. - (2026).
Multi-Source Remote Sensing and GeoAI Methodologies for Enhanced Environmental Risk Assessment
SANITA', MARSIA
2026-01-01
Abstract
This Ph.D. thesis highlights the necessity to propose a framework that blends the precision of Geomatics, the global vision of Remote Sensing, and the predictive power of GeoAI. The research activity has been based upon three fundamental pillars. The first pillar explores how to map vulnerability on a large scale using open-source computing platforms like Google Earth Engine to overcome the computing effort due to a lot of data for long time series analysis. By focusing on the Cultural Heritage sites of Hue City, Vietnam, it is proven that open-access data (e.g. Landsat imagery and Copernicus products) can be effective for protecting heritage in areas where local information is limited. To overcome these limitations and to harmonize and validate ground-based data with satellite data, the second pillar moves directly on field. Satellite thermal data of different satellite platforms have been compared with data acquired by sensors installed on the ground. The third pillar represents the GeoAI models application in environmental risk assessment. Through the use of models such as U-Net and YOLO, complex imagery was transformed into instantaneous maps of buildings and roads, serving as effective tools during emergencies or post-disaster phases. Furthermore, the research investigated non-invasive techniques for monitoring soil health and showcased the power of Augmented Reality in visualizing invisible threats, such as air pollution, effectively bringing scientific data directly to the public. This thesis can be seen as a practical contribution according to Destination Earth (DestinE) platform, a dynamic digital ecosystem designed to create high-precision replicas of Earth’s systems (Earth Digital Twins). By integrating multi-source geospatial data and high-performance computing, it enables the transition from historical observation to predictive and real-time urban management.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


