The prevention of natural hazards, and floods in particular, is one of the key challenges for the coming years, especially considering climate change. In particular, at the EU level, a culture has developed that focuses on mitigating the effects of flooding, rather than simply reducing the risk through structural measures alone. In this context, it is particularly important to study the causes of damage in relation to meteorological forcing, to establish a consistent reference point for general risk management. This data is important, but it is rarely collected in a systematic and rational manner and even more rarely made available to the scientific community. For this reason, it seemed interesting to investigate the scientific potential of a collection of damage data produced by considering the claims completed by citizens affected by flooding. This is undoubtedly a source with numerous limitations and wide margins of subjectivity but, as this work has shown, one with great potential. During the PhD program, various forms of application were investigated, exploring different uses and scales, with the aim of increasing knowledge of the territory regarding its capacity to generate damage as a result of flooding. In particular, the data were first analyzed in terms of their predictive capacity for individual events. The social and territorial context linked to the production of damage was then investigated, with particular attention to the subjective component of the declared damage, which was traced back to differences in perception. Finally, a model for classifying expected damage was implemented, capitalizing on the previous stages of the work in a machine learning model. The products of this phase are maps of the territory in terms of susceptibility to the generation of direct damage to private property following flood events. The results obtained, developed on the basis of events that have occurred in the Tuscany region in recent decades, are undoubtedly encouraging and show strong potential for further development. Among other aspects, this study adopts a broader objective by moving beyond the traditional interpretation of damage—typically based on the interaction of hazard, vulnerability, and exposure—through the introduction of an interpretative model directly grounded in the analysis of damage data collected in the aftermath of national emergencies. The model is derived from a comprehensive survey of observed damage data across the country,with particular focus on Tuscany. The large volume of properly georeferenced data enables a direct analysis of spatial distribution which, when combined with machine learning techniques, provides valuable insights into the propensity of different areas of the territory to experience flood damage.
La prevenzione dei rischi naturali, e delle alluvioni in particolare, è una delle sfide essenziali per i prossimi anni, anche alla luce dei cambiamenti climatici. In particolare, a livello comunitario si è andata sviluppando una cultura mirata al contenimento degli effetti degli eventi alluvionali, rispetto alla mera riduzione della pericolosità incentrata sui soli interventi strutturali. In questo contesto, viene a rivestire particolare importanza lo studio dei danni nella loro genesi a fronte della forzante meteorologica anche in modo da determinare un riferimento coerente per la generale azione di gestione del rischio. Si tratta di dati importanti, ma raramente raccolti in maniera organica e razionale e, ancora più raramente, resi disponibili alla comunità scientifica. Per questo motivo, è parso interessante investigare le potenzialità scientifiche di una raccolta di dati di danno prodotta considerando le schede di ricognizione compilate dai cittadini alluvionati. Si tratta indubbiamente di una fonte con numerosi limiti e ampi margini di soggettività ma, come è risultato da questo lavoro, in grado di dispiegare grandi potenzialità. Nel percorso del dottorato si sono indagate diverse forme di applicazione, esplorando usi e scale diverse, con lo scopo di incrementare la conoscenza del territorio rispetto alla capacità di generare danno a seguito di alluvioni. In particolare, i dati sono dapprima stati analizzati nella loro capacità predittiva sul singolo evento. Si è poi investigato il contesto sociale e territoriale legato alla produzione dei danni, con particolare attenzione alla componente di soggettività delle cifre dichiarate, ricondotta a differenze di percezione. Infine, si è implementato un modello per la classificazione del danno atteso, capitalizzando le fasi precedenti nel lavoro in un modello di machine learning. I prodotti di questa fase sono mappature del territorio in termini di suscettibilità alla generazione del danno diretto ai beni privati a seguito di eventi di alluvione. I risultati ottenuti, sviluppati sugli eventi occorsi nelle ultime decadi nella regione Toscana, sono senz’altro confortanti e mostrano forti possibilità di ulteriore sviluppo. Il lavoro, tra le altre cose, viene a superare la classica interpretazione del danno basata sulla combinazione tra la pericolosità, vulnerabilità ed esposto introducendo un modello interpretativo direttamente basato sulla interpretazione dei danni raccolti a valle delle emergenze nazionali. Questo modello è ottenuto a partire da una profonda ricognizione dei dati osservati di danno sul territorio nazionale, con un focus specifico sulla toscana. La rilevante numerosità del dato, adeguatamente georiferito, consente una elaborazione diretta della distribuzione spaziale che, attraverso tecniche di machine learning, acquisisce particolare significato in termini di propensione delle diverse aree di territorio a produrre danno da alluvione.
Analyzing flood damage beyond hazard: data-driven applications
Zambrini, Federica
2025/2026
Abstract
The prevention of natural hazards, and floods in particular, is one of the key challenges for the coming years, especially considering climate change. In particular, at the EU level, a culture has developed that focuses on mitigating the effects of flooding, rather than simply reducing the risk through structural measures alone. In this context, it is particularly important to study the causes of damage in relation to meteorological forcing, to establish a consistent reference point for general risk management. This data is important, but it is rarely collected in a systematic and rational manner and even more rarely made available to the scientific community. For this reason, it seemed interesting to investigate the scientific potential of a collection of damage data produced by considering the claims completed by citizens affected by flooding. This is undoubtedly a source with numerous limitations and wide margins of subjectivity but, as this work has shown, one with great potential. During the PhD program, various forms of application were investigated, exploring different uses and scales, with the aim of increasing knowledge of the territory regarding its capacity to generate damage as a result of flooding. In particular, the data were first analyzed in terms of their predictive capacity for individual events. The social and territorial context linked to the production of damage was then investigated, with particular attention to the subjective component of the declared damage, which was traced back to differences in perception. Finally, a model for classifying expected damage was implemented, capitalizing on the previous stages of the work in a machine learning model. The products of this phase are maps of the territory in terms of susceptibility to the generation of direct damage to private property following flood events. The results obtained, developed on the basis of events that have occurred in the Tuscany region in recent decades, are undoubtedly encouraging and show strong potential for further development. Among other aspects, this study adopts a broader objective by moving beyond the traditional interpretation of damage—typically based on the interaction of hazard, vulnerability, and exposure—through the introduction of an interpretative model directly grounded in the analysis of damage data collected in the aftermath of national emergencies. The model is derived from a comprehensive survey of observed damage data across the country,with particular focus on Tuscany. The large volume of properly georeferenced data enables a direct analysis of spatial distribution which, when combined with machine learning techniques, provides valuable insights into the propensity of different areas of the territory to experience flood damage.| File | Dimensione | Formato | |
|---|---|---|---|
|
FedericaZambrini_tesiPHD.pdf
non accessibile
Dimensione
12.62 MB
Formato
Adobe PDF
|
12.62 MB | 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/10589/255457