This thesis addresses the assessment of outdoor thermal comfort in urban environments, a topic of growing relevance in the context of climate change, which is leading to an increase in both the frequency and intensity of heatwaves. One of the main critical issues highlighted in the state of the art concerns the estimation of the Mean Radiant Temperature (MRT), a key variable for calculating the Universal Thermal Climate Index (UTCI), yet difficult to measure on a large scale due to the limitations of physical models and in situ observations. The objective of this work is to develop a scalable and replicable methodology for estimating MRT using machine learning techniques, integrating open-source meteorological and radiative datasets (ERA5, ERA5-Land, PVlib), and enabling real-time UTCI calculation. The proposed pipeline consists of five main modules: processing and harmonization of climate datasets, ML-based estimation of MRT through a multilayer neural network, UTCI calculation, integration with a socio-demographic vulnerability indicator (percentage of population over 65), and finally the generation of urban heat-risk maps. The methodology was applied to the metropolitan area of Barcelona, selected as a case study due to its high urban density and well-documented vulnerability to heatwaves. The results demonstrate strong model performance in predicting both MRT (R² = 0.993; RMSE = 1.357 °C) and UTCI (R² = 0.975; RMSE = 1.636 °C), with good generalization capability even for periods not included in the training set. The integration of thermal comfort and social vulnerability made it possible to identify areas characterized by elevated urban risk, revealing the co-occurrence of high thermal stress (UTCI > 38 °C) and a significant presence of elderly population. The discussion highlights how the ML-based approach offers an efficient alternative to traditional physical models, improving spatial scalability and reducing computational costs. The resulting maps align well with the urban microclimatic structure and provide valuable support for climate adaptation strategies and early warning systems. In conclusion, the developed methodology proves to be innovative, replicable, and transferable to other urban contexts, contributing to the assessment and management of extreme heat risk in urban environments.
La presente tesi affronta il tema della valutazione del comfort termico outdoor in ambiente urbano, un aspetto sempre più rilevante in un contesto di cambiamento climati-co, che sta determinando un aumento sia della frequenza sia dell’intensità delle ondate di calore. Una delle principali criticità emerse dallo stato dell’arte riguarda la stima della temperatura media radiante (MRT), variabile fondamentale per il calcolo dell’Universal Thermal Climate Index (UTCI), ma difficile da misurare su larga scala a causa dei limiti dei modelli fisici e delle misure in situ. L’obiettivo del lavoro è svi-luppare una metodologia scalabile e replicabile per la stima della MRT tramite tecni-che di machine learning, integrando dati meteorologici e radiativi open-source (ERA5, ERA5-Land, PVlib) e rendendo possibile il calcolo dell’UTCI anche in tempo reale. La pipeline proposta comprende cinque moduli principali: processamento e armoniz-zazione dei dataset climatici, stima ML-based della MRT mediante una rete neurale multistrato, calcolo dell’UTCI, integrazione con un indicatore di vulnerabilità socio-demografica (percentuale di popolazione over 65) e, infine, la generazione di mappe di rischio termico urbano. La metodologia è stata applicata all’area metropolitana di Bar-cellona, scelta come caso studio per l’elevata densità urbana e la riconosciuta vulnera-bilità alle ondate di calore. I risultati mostrano prestazioni elevate del modello nel predire sia la MRT (R² = 0.993; RMSE = 1.357 °C) sia l’UTCI (R² = 0.975; RMSE = 1.636 °C), con buona capacità di generalizzazione anche su periodi non inclusi nel training. L’integrazione fra comfort termico e vulnerabilità sociale ha consentito di identificare aree caratterizzate da ele-vato rischio urbano, evidenziando la co-occorrenza di alti livelli di stress termico (UTCI > 38 °C) e significativa presenza di popolazione anziana. La discussione dei risultati evidenzia come l’approccio ML-based rappresenti un’alternativa efficiente ai modelli fisici tradizionali, migliorando la scalabilità spa-ziale e riducendo il costo computazionale. Le mappe prodotte risultano coerenti con la struttura microclimatica urbana e costituiscono un utile supporto per strategie di adat-tamento climatico e sistemi di allerta precoce. In conclusione, la metodologia svilup-pata si dimostra innovativa, replicabile e trasferibile ad altri contesti urbani, contri-buendo alla valutazione e gestione del rischio da caldo estremo in ambiente urbano.
MISURA E PREVISIONE DEL COMFORT TERMICO NELL'ADATTAMENTO ALLE ONDATE DI CALORE URBANO IN UN CONTESTO DI CAMBIAMENTO CLIMATICO
CECCONI, NICOLAS
2024/2025
Abstract
This thesis addresses the assessment of outdoor thermal comfort in urban environments, a topic of growing relevance in the context of climate change, which is leading to an increase in both the frequency and intensity of heatwaves. One of the main critical issues highlighted in the state of the art concerns the estimation of the Mean Radiant Temperature (MRT), a key variable for calculating the Universal Thermal Climate Index (UTCI), yet difficult to measure on a large scale due to the limitations of physical models and in situ observations. The objective of this work is to develop a scalable and replicable methodology for estimating MRT using machine learning techniques, integrating open-source meteorological and radiative datasets (ERA5, ERA5-Land, PVlib), and enabling real-time UTCI calculation. The proposed pipeline consists of five main modules: processing and harmonization of climate datasets, ML-based estimation of MRT through a multilayer neural network, UTCI calculation, integration with a socio-demographic vulnerability indicator (percentage of population over 65), and finally the generation of urban heat-risk maps. The methodology was applied to the metropolitan area of Barcelona, selected as a case study due to its high urban density and well-documented vulnerability to heatwaves. The results demonstrate strong model performance in predicting both MRT (R² = 0.993; RMSE = 1.357 °C) and UTCI (R² = 0.975; RMSE = 1.636 °C), with good generalization capability even for periods not included in the training set. The integration of thermal comfort and social vulnerability made it possible to identify areas characterized by elevated urban risk, revealing the co-occurrence of high thermal stress (UTCI > 38 °C) and a significant presence of elderly population. The discussion highlights how the ML-based approach offers an efficient alternative to traditional physical models, improving spatial scalability and reducing computational costs. The resulting maps align well with the urban microclimatic structure and provide valuable support for climate adaptation strategies and early warning systems. In conclusion, the developed methodology proves to be innovative, replicable, and transferable to other urban contexts, contributing to the assessment and management of extreme heat risk in urban environments.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tesi Cecconi Nicolas .pdf
embargo fino al 13/06/2027
Dimensione
7.67 MB
Formato
Adobe PDF
|
7.67 MB | Adobe PDF |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.12075/24548