This thesis develops a functional geostatistical framework to produce earthquake ground-motion predictions that go beyond a mean model by explicitly modelling the residual field as a spatially correlated function of period. Within this setting, physics-based simulations and strong-motion recordings are integrated to leverage the complementary strengths of dense yet less reliable at short periods synthetic data and sparse but trustworthy observations. The approach decou- ples a deterministic mean effect from a residual component, and models the latter within the Object-Oriented Spatial Statistics framework. A central contribution is the accuracy-weighted estimation of the covariance operator, where period- and source-dependent weights modulate the influence of simulations and recordings during covariance learning and subsequent spatial prediction. In parallel, a source- separated formulation treats simulations and recordings as correlated yet distinct functional fields, allowing information to transfer through cross-covariances with- out explicit weighting. Together, these strategies deliver non-ergodic, continuous- spectrum predictions while controlling the impact of less reliable spectral regions. The framework is scalable and transferable, offering a general solution for integrat- ing heterogeneous-accuracy functional datasets in ground-motion modeling.
La tesi sviluppa un quadro geostatistico funzionale per la previsione del moto sismico del suolo che va oltre la stima del solo modello di media, modellando esplicitamente il campo dei residui come un campo spazialmente correlato di funzioni del periodo. In questo contesto vengono integrate simulazioni numeriche basate sulla fisica e registrazioni accelerometriche, sfruttando la natura complementare di dati sintetici densi ma meno affidabili ai periodi brevi e osservazioni reali più rade ma più robuste. L’approccio separa un effetto deterministico di media da una componente residuale, modellata all’interno del framework delle Object-Oriented Spatial Statistics. Un contributo centrale consiste nella stima pesata in accuratezza dell’operatore di covarianza, in cui pesi dipendenti da periodo e sorgente modulano l’influenza di simulazioni e registrazioni sia in fase di apprendimento della covarianza sia in previsione spaziale. In parallelo, una formulazione “source-separated” tratta simulazioni e dati empirici come campi funzionali distinti ma correlati, consentendo il trasferimento di informazione tramite le cross-covarianze senza introdurre pesi espliciti. Congiuntamente, queste strategie forniscono previsioni non ergodiche e a spettro continuo, controllando al contempo l’impatto delle regioni spettrali meno affidabili. Il framework risulta scalabile e trasferibile, e propone una soluzione generale per integrare dataset funzionali a eterogenea accuratezza nella modellazione del moto del suolo.
Weighted functional geostatistics for data-driven and physics-informed modelling of ground-motion
Koka, Darvin
2024/2025
Abstract
This thesis develops a functional geostatistical framework to produce earthquake ground-motion predictions that go beyond a mean model by explicitly modelling the residual field as a spatially correlated function of period. Within this setting, physics-based simulations and strong-motion recordings are integrated to leverage the complementary strengths of dense yet less reliable at short periods synthetic data and sparse but trustworthy observations. The approach decou- ples a deterministic mean effect from a residual component, and models the latter within the Object-Oriented Spatial Statistics framework. A central contribution is the accuracy-weighted estimation of the covariance operator, where period- and source-dependent weights modulate the influence of simulations and recordings during covariance learning and subsequent spatial prediction. In parallel, a source- separated formulation treats simulations and recordings as correlated yet distinct functional fields, allowing information to transfer through cross-covariances with- out explicit weighting. Together, these strategies deliver non-ergodic, continuous- spectrum predictions while controlling the impact of less reliable spectral regions. The framework is scalable and transferable, offering a general solution for integrat- ing heterogeneous-accuracy functional datasets in ground-motion modeling.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_12_KOKA_TESI.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
1.48 MB
Formato
Adobe PDF
|
1.48 MB | Adobe PDF | Visualizza/Apri |
|
2025_12_Koka_Executive Summary.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
694.34 kB
Formato
Adobe PDF
|
694.34 kB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/247414