Environmental policy making increasingly demands rapid, precise, and scalable analytical tools capable of addressing the accelerating pressures of climate change and urbanization. Traditional process-based models often fall short in capturing the complexity and heterogeneity of modern environmental Big Data streams, limiting their capacity for accurate prediction and adaptive decision-making. This dissertation responds to this challenge by developing and applying integrated Big Data mining and machine-learning frameworks to strengthen predictive modelling and evidence-based decision support in the urban water sector. The research is centered on two complex, flood-prone basins in Northern Italy. First, a 70-year spatiotemporal analysis of groundwater recharge in Milan Province demonstrates the dominant influence of increasing urbanization associated with reduced recharge and enhanced surface runoff. Second, hydrodynamic modelling of the Seveso River Basin using HEC-HMS/HEC-RAS explores the impacts of land-use change and evaluates the mitigation potential of Sustainable Urban Drainage Systems (SUDS), highlighting their relevance for urban flood risk reduction. To extend this analysis to emerging modelling paradigms, the thesis conducts a systematic review of Big Data and machine-learning applications, including remote sensing, IoT, and crowdsourced data, in urban flood management. Informed by these insights, the final component of the research develops and compares multi-paradigm streamflow forecasting models comprising standalone conceptual models (HEC-HMS), standalone machine-learning algorithms (Random Forest and Support Vector Regression), and hybrid frameworks that merge physical and data-driven representations. The hybrid HEC-HMS + RF approach consistently achieves superior performance, demonstrating the value of combining physically based hydrological structure with the pattern-recognition strengths of advanced ML techniques and achieving a validation performance. Overall, this dissertation shows that effective Big Data mining, particularly through physics-aware ML models, can bridge the persistent divide between scientific modelling and policy implementation. The integrated frameworks developed here provide a robust basis for advancing groundwater sustainability, flood mitigation, and adaptive urban water governance in data-rich and rapidly evolving environmental contexts.
La formulazione delle politiche ambientali richiede sempre più strumenti analitici rapidi, precisi e scalabili, in grado di affrontare le crescenti pressioni dovute ai cambiamenti climatici e all’urbanizzazione. I modelli basati su processi tradizionali spesso non riescono a cogliere la complessità e l’eterogeneità dei moderni flussi di Big Data ambientali, limitando la loro capacità predittiva e il supporto decisionale adattivo. Questa tesi risponde a tale sfida sviluppando e applicando quadri integrati di Big Data mining e tecniche di machine learning per rafforzare la modellazione predittiva e il supporto alle decisioni basate su evidenze nel settore idrico urbano. La ricerca è incentrata su due bacini complessi e soggetti a inondazioni dell’Italia settentrionale. In primo luogo, un’analisi spaziotemporale di 70 anni della ricarica delle falde acquifere nella Provincia di Milano evidenzia l’influenza dominante dell’urbanizzazione crescente, associata a una riduzione della ricarica e a un aumento del deflusso superficiale. In secondo luogo, la modellazione idrodinamica del bacino del fiume Seveso mediante HEC-HMS/HEC-RAS esplora gli impatti dei cambiamenti d’uso del suolo e valuta il potenziale mitigativo dei Sistemi di Drenaggio Urbano Sostenibile (SUDS), evidenziandone la rilevanza per la riduzione del rischio di alluvioni urbane. Per estendere questa analisi ai paradigmi modellistici emergenti, la tesi conduce una revisione sistematica delle applicazioni di Big Data e machine learning, includendo dati da telerilevamento, IoT e fonti partecipative nella gestione delle inondazioni urbane. Informata da tali approfondimenti, l’ultima parte della ricerca sviluppa e confronta modelli previsionali di portata fluviale basati su approcci concettuali autonomi (HEC-HMS), algoritmi di machine learning puri (Random Forest e Support Vector Regression) e framework ibridi che uniscono rappresentazioni fisiche e data-driven. L’approccio ibrido HEC-HMS + RF mostra costantemente le migliori performance, dimostrando il valore della combinazione tra strutture idrologiche fisicamente basate e le capacità di riconoscimento dei pattern proprie delle tecniche avanzate di ML. In definitiva, questa tesi dimostra come un efficace Big Data mining, in particolare mediante modelli di ML consapevoli dei processi fisici, possa colmare il divario persistente tra modellazione scientifica e implementazione delle politiche. I quadri integrati sviluppati rappresentano una base solida per il progresso della sostenibilità delle risorse idriche sotterranee, della mitigazione delle alluvioni e della gestione adattiva delle risorse idriche urbane in contesti ambientali data-rich e in rapida evoluzione.
Mining big data to support environmental policymaking
MIREMAD, SEYEDMOEIN
2025/2026
Abstract
Environmental policy making increasingly demands rapid, precise, and scalable analytical tools capable of addressing the accelerating pressures of climate change and urbanization. Traditional process-based models often fall short in capturing the complexity and heterogeneity of modern environmental Big Data streams, limiting their capacity for accurate prediction and adaptive decision-making. This dissertation responds to this challenge by developing and applying integrated Big Data mining and machine-learning frameworks to strengthen predictive modelling and evidence-based decision support in the urban water sector. The research is centered on two complex, flood-prone basins in Northern Italy. First, a 70-year spatiotemporal analysis of groundwater recharge in Milan Province demonstrates the dominant influence of increasing urbanization associated with reduced recharge and enhanced surface runoff. Second, hydrodynamic modelling of the Seveso River Basin using HEC-HMS/HEC-RAS explores the impacts of land-use change and evaluates the mitigation potential of Sustainable Urban Drainage Systems (SUDS), highlighting their relevance for urban flood risk reduction. To extend this analysis to emerging modelling paradigms, the thesis conducts a systematic review of Big Data and machine-learning applications, including remote sensing, IoT, and crowdsourced data, in urban flood management. Informed by these insights, the final component of the research develops and compares multi-paradigm streamflow forecasting models comprising standalone conceptual models (HEC-HMS), standalone machine-learning algorithms (Random Forest and Support Vector Regression), and hybrid frameworks that merge physical and data-driven representations. The hybrid HEC-HMS + RF approach consistently achieves superior performance, demonstrating the value of combining physically based hydrological structure with the pattern-recognition strengths of advanced ML techniques and achieving a validation performance. Overall, this dissertation shows that effective Big Data mining, particularly through physics-aware ML models, can bridge the persistent divide between scientific modelling and policy implementation. The integrated frameworks developed here provide a robust basis for advancing groundwater sustainability, flood mitigation, and adaptive urban water governance in data-rich and rapidly evolving environmental contexts.| File | Dimensione | Formato | |
|---|---|---|---|
|
PhD_Thesis_Miremad.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
5.31 MB
Formato
Adobe PDF
|
5.31 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/10589/257478