Mesoscale features, such as eddies and meandering currents, contain most of the kinetic energy of the ocean and mediate the transfer of energy between the large-scale circulation and the small, often turbulent processes below the kilometre range. Mesoscale dynamics is ubiquitous, from global basins to the enclosed Mediterranean Sea, and down to more local and coastal environments. This thesis applies a modelling approach to investigate the mesoscale properties of the northern Adriatic Sea, one of the most dynamic and densely monitored regions of the Mediterranean. Its shallow bathymetry, intense air–sea interactions, and complex coastal morphology make it a challenging environment for numerical studies. The work explores how numerical models, combined with multi-platform observations and data-driven methods, can improve understanding, reconstruction, and prediction of mesoscale processes in the region, with the long-term goal of developing a \textit{Digital Twin} (DT) of the northern Adriatic. The study begins with an evaluation of the sensitivity of the Massachusetts Institute of Technology general circulation model (MITgcm) to different vertical resolutions and turbulence closures. A turbulence kinetic energy (TKE)-based scheme (GGL) provides the most realistic representation of thermohaline stratification and surface temperature, reducing bias and bringing free-running simulations close to the skill of reanalyses that assimilate observations. High-resolution simulations reveal rich mesoscale activity, with eddies and fronts (2–20 km) corresponding to the baroclinic Rossby radius of deformation. These structures drive cross-basin transport, redistributing freshwater from coastal to offshore areas through advection and entrainment. Model skill is further confirmed by the reconstruction of two extreme events in the Gulf of Trieste: a bottom marine heatwave and a flooding episode, both reproduced accurately in timing and intensity. Building on these studies, two complementary approaches are developed to enhance model–data integration. The first is a four-dimensional variational (4DVar) data assimilation scheme, merging simulations and observations into dynamically consistent reconstructions. The second adopts convolutional neural networks for super-resolution, capable of emulating high-resolution fields from coarse inputs and other data such as river discharge. Together, these methods demonstrate how physics-based models and data-driven algorithms can be integrated within a Digital Twin framework. The thesis concludes by outlining a conceptual architecture for a Digital Twin of the northern Adriatic Sea, where the numerical model forms the dynamic core, continuously informed by data assimilation and augmented by machine learning components. This integrated system would enable near-real-time simulation, reconstruction, and forecasting of the basin’s state, supporting scientific research, operational applications, climate adaptation, and coastal management. The thesis therefore provides both a methodological and conceptual contribution for the next generation of coastal ocean models, data-integrated, adaptive, and predictive, marking a further step toward a fully operational Digital Twin of the northern Adriatic Sea.
Le strutture di mesoscala, come vortici e correnti meandriformi, contengono la maggior parte dell’energia cinetica dell’oceano e mediano il trasferimento di energia tra la circolazione su larga scala e i processi di piccola scala, spesso turbolenti, che avvengono al di sotto del chilometro. La dinamica di mesoscala è ubiquitaria, dai grandi bacini oceanici fino al Mediterraneo e ai suoi ambienti costieri più locali. Questa tesi adotta un approccio di modellistica numerica per indagare le proprietà alla mesoscala dell’Adriatico settentrionale, una delle regioni più dinamiche e densamente monitorate del Mediterraneo. La batimetria poco profonda, le intense interazioni aria–mare e la complessa morfologia costiera rendono quest’area un ambiente particolarmente impegnativo per gli studi numerici. Il lavoro esplora come modelli numerici, combinati con osservazioni multi-piattaforma e metodi data-driven, possano migliorare la comprensione, la ricostruzione e la previsione dei processi di mesoscala nella regione, con l’obiettivo a lungo termine di sviluppare un Digital Twin (DT) dell’Adriatico settentrionale. Lo studio inizia con una valutazione della sensibilità del modello di circolazione generale del Massachusetts Institute of Technology (MITgcm) a diverse risoluzioni verticali e schemi di chiusura turbolenta. Lo schema GGL basato sull’energia cinetica turbolenta (TKE) fornisce la rappresentazione più realistica della stratificazione termoalina e della temperatura superficiale, riducendo le discrepanze e avvicinando le simulazioni libere alle prestazioni delle rianalisi che assimilano dati osservativi. Le simulazioni ad alta risoluzione rivelano un’intensa attività di mesoscala, caratterizzata da vortici e fronti di scala compresa tra 2 e 20 km, corrispondenti al raggio di deformazione di Rossby baroclino. Tali strutture regolano i processi di trasporto trasversale del bacino, ridistribuendo l’acqua dolce dalle zone costiere verso il mare aprto attraverso meccanismi di avvezione ed entrainment. L’abilità del modello è ulteriormente confermata dalla ricostruzione di due eventi estremi verificatisi nel Golfo di Trieste: un’ondata di calore marina al fondo e un episodio di inondazione costiera, entrambi riprodotti con accuratezza sia nei tempi sia in intensità. Sulla base di questi risultati, sono stati sviluppati due approcci complementari per migliorare l’integrazione tra modello e dati. Il primo consiste in uno schema di assimilazione variazionale quadridimensionale (4DVar), che fonde simulazioni e osservazioni in ricostruzioni dinamicamente coerenti. Il secondo impiega reti neurali convoluzionali per la super-risoluzione, in grado di emulare campi ad alta risoluzione a partire da input a più bassa risoluzione e da dati aggiuntivi, come la portata fluviale. Insieme, questi metodi dimostrano come modelli numerici basati sulla fisica e algoritmi data-driven possano essere integrati in un quadro coerente all’interno di un Digital Twin. La tesi si conclude delineando un’architettura concettuale per un Digital Twin dell’Adriatico settentrionale, in cui il modello numerico costituisce il nucleo dinamico, costantemente aggiornato attraverso l’assimilazione dei dati e potenziato da componenti di machine learning. Un sistema integrato di questo tipo consentirebbe la simulazione, la ricostruzione e la previsione quasi in tempo reale dello stato del bacino, a supporto della ricerca scientifica, delle applicazioni operative, dell’adattamento ai cambiamenti climatici e della gestione costiera. La tesi fornisce quindi un contributo sia metodologico che concettuale allo sviluppo della prossima generazione di modelli oceanici costieri, integrati con i dati, adattivi e predittivi, segnando un ulteriore passo verso la realizzazione completa di un Digital Twin operativo dell’Adriatico settentrionale.
Dinamica di mesoscala nell’Adriatico settentrionale: approccio modellistico e integrazione di dati multi-piattaforma in un’ottica di Digital Twin
GIORDANO, FABIO
2026
Abstract
Mesoscale features, such as eddies and meandering currents, contain most of the kinetic energy of the ocean and mediate the transfer of energy between the large-scale circulation and the small, often turbulent processes below the kilometre range. Mesoscale dynamics is ubiquitous, from global basins to the enclosed Mediterranean Sea, and down to more local and coastal environments. This thesis applies a modelling approach to investigate the mesoscale properties of the northern Adriatic Sea, one of the most dynamic and densely monitored regions of the Mediterranean. Its shallow bathymetry, intense air–sea interactions, and complex coastal morphology make it a challenging environment for numerical studies. The work explores how numerical models, combined with multi-platform observations and data-driven methods, can improve understanding, reconstruction, and prediction of mesoscale processes in the region, with the long-term goal of developing a \textit{Digital Twin} (DT) of the northern Adriatic. The study begins with an evaluation of the sensitivity of the Massachusetts Institute of Technology general circulation model (MITgcm) to different vertical resolutions and turbulence closures. A turbulence kinetic energy (TKE)-based scheme (GGL) provides the most realistic representation of thermohaline stratification and surface temperature, reducing bias and bringing free-running simulations close to the skill of reanalyses that assimilate observations. High-resolution simulations reveal rich mesoscale activity, with eddies and fronts (2–20 km) corresponding to the baroclinic Rossby radius of deformation. These structures drive cross-basin transport, redistributing freshwater from coastal to offshore areas through advection and entrainment. Model skill is further confirmed by the reconstruction of two extreme events in the Gulf of Trieste: a bottom marine heatwave and a flooding episode, both reproduced accurately in timing and intensity. Building on these studies, two complementary approaches are developed to enhance model–data integration. The first is a four-dimensional variational (4DVar) data assimilation scheme, merging simulations and observations into dynamically consistent reconstructions. The second adopts convolutional neural networks for super-resolution, capable of emulating high-resolution fields from coarse inputs and other data such as river discharge. Together, these methods demonstrate how physics-based models and data-driven algorithms can be integrated within a Digital Twin framework. The thesis concludes by outlining a conceptual architecture for a Digital Twin of the northern Adriatic Sea, where the numerical model forms the dynamic core, continuously informed by data assimilation and augmented by machine learning components. This integrated system would enable near-real-time simulation, reconstruction, and forecasting of the basin’s state, supporting scientific research, operational applications, climate adaptation, and coastal management. The thesis therefore provides both a methodological and conceptual contribution for the next generation of coastal ocean models, data-integrated, adaptive, and predictive, marking a further step toward a fully operational Digital Twin of the northern Adriatic Sea.| File | Dimensione | Formato | |
|---|---|---|---|
|
PhD_thesis_Giordano_pdfA_v2.pdf
embargo fino al 27/03/2027
Licenza:
Tutti i diritti riservati
Dimensione
10.36 MB
Formato
Adobe PDF
|
10.36 MB | Adobe PDF | |
|
PhD_thesis_Giordano_pdfA_v2_1.pdf
embargo fino al 27/03/2027
Licenza:
Tutti i diritti riservati
Dimensione
10.36 MB
Formato
Adobe PDF
|
10.36 MB | Adobe PDF |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/363715
URN:NBN:IT:UNITS-363715