This thesis explores the application of Bayesian methods through the use of Integrated Nested Laplace Approximation (INLA) in the analysis of spatio-temporal data, focusing particularly on geostatistical data and the monitoring of air pollution. In the first chapter, the fundamental concepts and the importance of Bayesian inference in modern statistical contexts are presented. This serves to introduce the notation and the necessary elements to understand the subsequent chapters, providing a comprehensive overview of the theoretical principles underlying Bayesian inference. The choice of prior and posterior distributions, essential for statistical modeling, is also discussed. The second chapter details the Integrated Nested Laplace Approximation (INLA) algorithm, a deterministic method that allows for obtaining results more efficiently compared to traditional methods used in the Bayesian domain, such as Markov Chain Monte Carlo (MCMC). The principles of Laplace approximation, latent Gaussian models, and Gaussian Markov random fields are explained, highlighting the advantages of INLA in terms of speed and accuracy. In the third chapter, the fundamentals and applications of spatio-temporal models for geostatistical data are explored. Particular attention is given to the approach based on stochastic partial differential equations (SPDE), which allows for the coherent and effective modeling of spatial and temporal data. Stochastic Gaussian processes and the integration of hierarchical Bayesian models with SPDE are discussed. The final chapter presents a practical application using real atmospheric pollution data from the Regional Environmental Protection Agencies (ARPA). A detailed analysis of the data is conducted, with the development of spatio-temporal models for pollution monitoring. The obtained results are examined, comparing different models.
Questa tesi esplora l'applicazione dei metodi bayesiani tramite l’utilizzo dell’Integrated Nested Laplace Approximation (INLA) nell'analisi dei dati spazio-temporali, concentrandosi in particolare sui dati geostatistici e sul monitoraggio dell'inquinamento atmosferico. Nel primo capitolo, vengono presentati i concetti fondamentali e l'importanza dell'inferenza bayesiana nei contesti statistici moderni. Questo serve a introdurre la notazione e gli elementi necessari per comprendere i capitoli successivi, offrendo una panoramica completa dei principi teorici alla base dell'inferenza bayesiana. Viene anche discussa la scelta delle distribuzioni a priori e a posteriori, essenziale per la modellazione statistica. Il secondo capitolo espone in dettaglio l’algoritmo Integrated Nested Laplace Approximation (INLA), un metodo deterministico che consente di ottenere risultati in modo più efficiente rispetto ai metodi tradizionali usati nell'ambito bayesiano, come il Markov Chain Monte Carlo (MCMC). Vengono spiegati i principi dell'approssimazione di Laplace, i modelli latenti gaussiani e i campi casuali di Markov gaussiani, evidenziando i vantaggi di INLA in termini di velocità e accuratezza.Nel terzo capitolo, si esplorano i fondamenti e le applicazioni dei modelli spazio-temporali per dati geostatistici. Particolare attenzione è dedicata all'approccio basato sulle equazioni differenziali stocastiche parziali (SPDE), che permette di modellare i dati spaziali e temporali in modo coerente ed efficace. Vengono discussi i processi stocastici gaussiani e l'integrazione dei modelli gerarchici bayesiani con SPDE. L’ultimo capitolo presenta un’applicazione pratica utilizzando dati reali sull'inquinamento atmosferico, provenienti dalle Agenzie Regionali per la Protezione Ambientale (ARPA). Viene effettuata un'analisi dettagliata dei dati, con sviluppo di modelli spazio-temporali per il monitoraggio dell'inquinamento. Si esaminano i risultati ottenuti, confrontando diversi modelli.
L’approccio SPDE in modelli bayesiani spazio-temporali per il monitoraggio dell’inquinamento atmosferico
MONTE, VALENTINA
2023/2024
Abstract
This thesis explores the application of Bayesian methods through the use of Integrated Nested Laplace Approximation (INLA) in the analysis of spatio-temporal data, focusing particularly on geostatistical data and the monitoring of air pollution. In the first chapter, the fundamental concepts and the importance of Bayesian inference in modern statistical contexts are presented. This serves to introduce the notation and the necessary elements to understand the subsequent chapters, providing a comprehensive overview of the theoretical principles underlying Bayesian inference. The choice of prior and posterior distributions, essential for statistical modeling, is also discussed. The second chapter details the Integrated Nested Laplace Approximation (INLA) algorithm, a deterministic method that allows for obtaining results more efficiently compared to traditional methods used in the Bayesian domain, such as Markov Chain Monte Carlo (MCMC). The principles of Laplace approximation, latent Gaussian models, and Gaussian Markov random fields are explained, highlighting the advantages of INLA in terms of speed and accuracy. In the third chapter, the fundamentals and applications of spatio-temporal models for geostatistical data are explored. Particular attention is given to the approach based on stochastic partial differential equations (SPDE), which allows for the coherent and effective modeling of spatial and temporal data. Stochastic Gaussian processes and the integration of hierarchical Bayesian models with SPDE are discussed. The final chapter presents a practical application using real atmospheric pollution data from the Regional Environmental Protection Agencies (ARPA). A detailed analysis of the data is conducted, with the development of spatio-temporal models for pollution monitoring. The obtained results are examined, comparing different models.File | Dimensione | Formato | |
---|---|---|---|
Monte_Valentina.pdf
accesso riservato
Dimensione
22.98 MB
Formato
Adobe PDF
|
22.98 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/71212