Local Climate Zone (LCZ) classification provides a standardized framework for analyzing urban thermal environments and urban heat island effects. Current LCZ mapping approaches mainly rely on optical remote sensing imagery and urban canopy parameters (UCPs), which effectively describe urban morphology but underexploit the thermal dimension. As a result, surface materials and building types with similar spectral and structural characteristics often remain difficult to distinguish. The forthcoming NASA SBG-TIR (Surface Biology and Geology - Thermal Infrared) mission is expected to provide high-resolution thermal infrared observations specifically designed to capture surface temperature and emissivity with enhanced detail. This thesis aims to simulate the expected contribution of SBG-TIR to LCZ mapping. To achieve this, ECOSTRESS (ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station) is used as a proxy, as it currently offers the most comparable high-resolution thermal infrared products in terms of spatial resolution and multi-variable surface characterization. Sentinel-2 multispectral imagery and UCPs are combined with ECOSTRESS-derived land surface temperature (LST), emissivity, NDVI, and albedo within a Random Forest classification framework to assess whether integrating high-resolution TIR information enhances classification accuracy, temporal robustness, and geographical transferability. The analysis is conducted in three stages. First, a primary case study over Milan (Northern Italy) evaluates the progressive contribution of ECOSTRESS features for a representative summer date. Incorporating all ECOSTRESS products increases overall accuracy from 0.815 to 0.859 (+4.4%), with improvements in both built-type (+4.8%) and land cover (+4.0%) classes. Thermal variables (LST and emissivity) reduce material-related confusion and improve building subtype separation, while ecological variables (NDVI and albedo) strengthen vegetation density and natural land cover discrimination. Second, a multi-temporal extension across additional summer dates in Milan confirms that built-type improvements are temporally stable (OA gains between +0.7 and +4.4%), although natural-class performance varies with phenological conditions. Third, a transferability analysis across cities with contrasting climates and urban structures (Rome, Hanoi, and Ho Chi Minh City) demonstrates consistent gains in land cover accuracy, while built-type improvements depend on urban complexity and acquisition consistency. Overall, the findings demonstrate that high-resolution TIR data, representative of the future SBG-TIR capabilities, can significantly enhance LCZ classification, particularly by resolving material-related ambiguities and strengthening natural class discrimination. This study provides methodological and empirical guidance for integrating next-generation TIR observations into urban climate research and supports more physically informed LCZ mapping for urban planning and climate adaptation.
La classificazione delle Zone Climatiche Locali (Local Climate Zone, LCZ) fornisce un quadro standardizzato per l’analisi degli ambienti termici urbani e degli effetti di isola di calore urbana. Gli attuali approcci di mappatura LCZ si basano principalmente su immagini satellitari ottiche e parametri di canopy urbano (Urban Canopy Parameter, UCP), che descrivono efficacemente la morfologia urbana ma sfruttano in modo limitato la dimensione termica. Di conseguenza, materiali superficiali e tipologie edilizie con caratteristiche spettrali e strutturali simili risultano spesso difficili da distinguere. La futura missione NASA SBG-TIR (Surface Biology and Geology - Thermal Infrared) fornirà osservazioni termiche ad alta risoluzione, progettate per rilevare temperatura superficiale ed emissività con maggiore dettaglio. Questa tesi mira a simulare il contributo atteso della missione SBG-TIR alla mappatura LCZ. A tal fine, viene utilizzato ECOSTRESS (ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station) come proxy, poiché attualmente tale sensore offre i prodotti termici ad alta risoluzione più comparabili in termini di dettaglio spaziale e caratterizzazione multi-variabile delle superfici. Le immagini multispettrali Sentinel-2 e gli UCPs sono integrati con prodotti derivati da ECOSTRESS (temperatura superficiale (Land Surface Temperature (LST), emissività, NDVI e albedo) all’interno di un framework basato su Random Forest, per valutare se l’integrazione di informazioni TIR ad alta risoluzione migliori l’accuratezza, la robustezza temporale e la trasferibilità geografica della classificazione. L’analisi si articola in tre fasi. In primo luogo, un caso studio principale sull’area metropolitana di Milano valuta il contributo progressivo delle variabili ECOSTRESS per una data estiva rappresentativa. L’integrazione di tutti i prodotti ECOSTRESS incrementa l’accuratezza complessiva da 0.815 a 0.859 (+4.4%), con miglioramenti sia per le classi costruite (+4.8%) sia per le classi naturali (+4.0%). Le variabili termiche (LST ed emissività) riducono le confusioni legate ai materiali e migliorano la separazione tra sottotipi edilizi, mentre le variabili ecologiche (NDVI e albedo) migliorano la distinzione della densità di vegetazione e delle coperture naturali. In secondo luogo, un’estensione multi-temporale su ulteriori date estive a Milano conferma che i miglioramenti per le classi costruite sono stabili nel tempo (incrementi di OA compresi tra +0.7% e +4.4%), mentre le prestazioni per le classi naturali variano in funzione delle condizioni fenologiche. In terzo luogo, un’analisi di trasferibilità condotta su città con climi e strutture urbane differenti (Roma, Hanoi e Ho Chi Minh City) evidenzia miglioramenti consistenti per le classi di copertura del suolo, mentre i benefici per le classi costruite dipendono dalla complessità urbana e dalla coerenza temporale delle acquisizioni. Nel complesso, i risultati dimostrano che dati TIR ad alta risoluzione, rappresentativi delle future capacità della missione SBG-TIR, possono migliorare significativamente la classificazione LCZ, in particolare riducendo le ambiguità legate ai materiali e rafforzando la discriminazione delle classi naturali. Lo studio fornisce indicazioni metodologiche ed evidenze empiriche per l’integrazione delle osservazioni TIR di nuova generazione nella ricerca sul clima urbano e contribuisce allo sviluppo di una mappatura LCZ più fisicamente basata a supporto della pianificazione urbana e delle strategie di adattamento climatico.
Simulating SBG-TIR data for urban applications:local climate zone classification using ECOSTRESS
Tan, Xiao
2025/2026
Abstract
Local Climate Zone (LCZ) classification provides a standardized framework for analyzing urban thermal environments and urban heat island effects. Current LCZ mapping approaches mainly rely on optical remote sensing imagery and urban canopy parameters (UCPs), which effectively describe urban morphology but underexploit the thermal dimension. As a result, surface materials and building types with similar spectral and structural characteristics often remain difficult to distinguish. The forthcoming NASA SBG-TIR (Surface Biology and Geology - Thermal Infrared) mission is expected to provide high-resolution thermal infrared observations specifically designed to capture surface temperature and emissivity with enhanced detail. This thesis aims to simulate the expected contribution of SBG-TIR to LCZ mapping. To achieve this, ECOSTRESS (ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station) is used as a proxy, as it currently offers the most comparable high-resolution thermal infrared products in terms of spatial resolution and multi-variable surface characterization. Sentinel-2 multispectral imagery and UCPs are combined with ECOSTRESS-derived land surface temperature (LST), emissivity, NDVI, and albedo within a Random Forest classification framework to assess whether integrating high-resolution TIR information enhances classification accuracy, temporal robustness, and geographical transferability. The analysis is conducted in three stages. First, a primary case study over Milan (Northern Italy) evaluates the progressive contribution of ECOSTRESS features for a representative summer date. Incorporating all ECOSTRESS products increases overall accuracy from 0.815 to 0.859 (+4.4%), with improvements in both built-type (+4.8%) and land cover (+4.0%) classes. Thermal variables (LST and emissivity) reduce material-related confusion and improve building subtype separation, while ecological variables (NDVI and albedo) strengthen vegetation density and natural land cover discrimination. Second, a multi-temporal extension across additional summer dates in Milan confirms that built-type improvements are temporally stable (OA gains between +0.7 and +4.4%), although natural-class performance varies with phenological conditions. Third, a transferability analysis across cities with contrasting climates and urban structures (Rome, Hanoi, and Ho Chi Minh City) demonstrates consistent gains in land cover accuracy, while built-type improvements depend on urban complexity and acquisition consistency. Overall, the findings demonstrate that high-resolution TIR data, representative of the future SBG-TIR capabilities, can significantly enhance LCZ classification, particularly by resolving material-related ambiguities and strengthening natural class discrimination. This study provides methodological and empirical guidance for integrating next-generation TIR observations into urban climate research and supports more physically informed LCZ mapping for urban planning and climate adaptation.| File | Dimensione | Formato | |
|---|---|---|---|
|
Simulating SBG-TIR Data for Urban Applications - Local Climate Zone Classification Using ECOSTRESS.pdf
accessibile in internet per tutti a partire dal 27/02/2027
Dimensione
108.36 MB
Formato
Adobe PDF
|
108.36 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/10589/252533