The European Alps have been affected by intensification of meteorological droughts in recent years. Due to changing climatic conditions, the region is vulnerable to deviations in water cycling, which can be observed in the context of evapotranspiration (ET) anomalies. Land surface temperature (LST) is a key factor in regulating the exchange of water and energy between land and atmosphere, which directly relates it to ET. Development of two-source energy balance (TSEB) models driven by thermal remote sensing data has made a significant contribution to estimate ET at large scale. However, their coarse spatial resolution and sensitivity of TIR instruments to cloudy-sky conditions make them insufficient for complex ecosystems, such as mountain regions. To overcome these limitations, this thesis served to develop a new clear-sky land surface temperature product at 250 m spatial resolution, as an alternative to 1-km MODIS LST data, for estimating fine-resolution TSEB fluxes. In the first part of the thesis, imbalance between spatial resolution of 1-km MODIS LST data was solved by applying a sharpening procedure to obtain daily LST at 250-m spatial resolution. Due to reduced capabilities of LST–VNIR statistical models in complex ecosystems, multi-source predictors, including normalized difference vegetation index (NDVI) and digital elevation model (DEM) were used. Inspired by superiority of machine learning for non-linear problems, relationships between coarse resolution LSTs and 250-m predictors with random forest (RF) algorithm were exploited. The obtained results indicate an improvement of 20% in the agreement between Landsat and the sharpened LST compared to statistics for the original MODIS dataset. The LST models determined averaged RMSE of 2.3°C and MAE of 1.8°C. In order to reconstruct missing LSTs beneath the clouds, the author proposed a novel approach to predict invalid pixels by exploiting correlation between ground-based LST and air temperature in conjunction with auxiliary variables, e.g., downwelling solar radiation, albedo- and LAI-derived products under long-term cloudy-skies. Considering a high site dependency driven by different land-cover types, LST reconstruction was performed for aggregated stations that represent three vegetation groups: grassland, forest and permanent crops. The gap-filling was performed with two steps: site-based LST modelling from ground-derived variables under cloudy skies, and then applying the fitted models to gridded predictors at subpixel level corresponding to the downscaled output. The reconstruction achieved reliable performance with local data yielding R2 of 0.84 and RMSE of 2.12°C. In the last part of the thesis, the resulting LST maps were incorporated into two-source energy balance model of Priestley-Taylor for estimating energy fluxes at 250-m spatial resolution. First, the performance of the model forced by local temperatures was evaluated with measured fluxes from eddy covariance towers. The benchmark simulations for latent (LE) and sensible heat (H) yielded an averaged RMSE of 57 Wm-2 and mean absolute bias (MB) of 26 Wm-2. Next, the model estimates driven by satellite-based LSTs, i.e., original 1-km MODIS LST product and downscaled maps, were validated against in-situ data. Turbulent fluxes modelled with 250-m LSTs resulted in RMSE of 86 Wm-2 and MB of 55 Wm-2, which translated to 8% and 15% decrease in the respective errors when compared to TSEB estimates combined with original MODIS LST.

Le Alpi sono state colpite dall'intensificarsi della siccità meteorologica negli ultimi anni. A causa delle mutevoli condizioni climatiche, la regione è vulnerabile alle deviazioni nel ciclo dell'acqua, che possono essere osservate nel contesto delle anomalie dell'evapotraspirazione (ET). La temperatura della superficie terrestre (LST) è un fattore chiave nella regolazione dello scambio di acqua ed energia tra terra e atmosfera, che la mette in relazione direttamente con ET. Lo sviluppo di modelli di bilancio energetico a due fonti (TSEB) guidati da dati di telerilevamento termico ha dato un contributo significativo alla stima dell'ET su larga scala. Tuttavia, la loro risoluzione spaziale grossolana e la sensibilità degli strumenti TIR alle condizioni del cielo nuvoloso li rendono insufficienti per ecosistemi complessi, come le regioni di montagna. Per superare queste limitazioni, questa tesi è servita per sviluppare un nuovo prodotto della temperatura della superficie terrestre in cielo sereno con una risoluzione spaziale di 250 m, in alternativa ai dati MODIS LST di 1 km, per la stima dei flussi TSEB a risoluzione fine. Nella prima parte della tesi, lo squilibrio tra la risoluzione spaziale dei dati MODIS LST di 1 km è stato risolto applicando una procedura di sharpening per ottenere LST giornaliero a una risoluzione spaziale di 250 m. A causa delle ridotte capacità dei modelli statistici LST-VNIR in ecosistemi complessi, sono stati utilizzati predittori multi-sorgente, tra cui l'indice di NDVI e il DEM. Ispirato dalla superiorità dell'apprendimento automatico per problemi non lineari, sono state sfruttate le relazioni tra LST a risoluzione grossolana e predittori di 250 m con algoritmo a foresta casuale (RF). I risultati ottenuti indicano un miglioramento del 20% nell'accordo tra Landsat e l'affilato LST rispetto alle statistiche per il set di dati MODIS originale. I modelli LST hanno determinato un RMSE medio di 2,3°C e un MAE di 1,8°C. Al fine di ricostruire gli LST mancanti sotto le nuvole, l'autore ha proposto un nuovo approccio per prevedere i pixel non validi sfruttando la correlazione tra LST a terra e temperatura dell'aria in combinazione con variabili ausiliarie, ad esempio radiazione solare discendente, albedo e LAI sotto cieli nuvolosi a lungo termine. Considerando un'elevata dipendenza del sito determinata da diversi tipi di copertura del suolo, la ricostruzione LST è stata eseguita per stazioni aggregate che rappresentano tre gruppi di vegetazione: praterie, foreste e colture permanenti. Il gap-filling è stato eseguito con due passaggi: modellazione LST basata sul sito da variabili derivate dal suolo sotto cieli nuvolosi e quindi applicazione dei modelli adattati a predittori a griglia a livello di subpixel corrispondenti all'output ridotto. La ricostruzione ha ottenuto prestazioni affidabili con dati locali che producono R2 di 0,84 e RMSE di 2,12°C. Nell'ultima parte della tesi, le mappe LST risultanti sono state incorporate nel modello di bilancio energetico a due fonti di Priestley-Taylor per la stima dei flussi energetici a una risoluzione spaziale di 250 m. Innanzitutto, le prestazioni del modello forzato dalle temperature locali sono state valutate con flussi misurati da torri di eddy covariance. Le simulazioni di riferimento per il calore latente (LE) e sensibile (H) hanno prodotto un RMSE medio di 57 Wm-2 e un bias assoluto medio (MB) di 26 Wm-2. Successivamente, le stime del modello guidate da LST basati su satellite, ovvero il prodotto LST MODIS originale di 1 km e le mappe ridimensionate, sono state convalidate rispetto ai dati in situ. Flussi turbolenti modellati con LST di 250 m hanno portato a RMSE di 86 Wm-2 e MB di 55 Wm-2, che si sono tradotti in una diminuzione dell'8% e del 15% nei rispettivi errori rispetto alle stime TSEB combinate con l'LST MODIS originale.

(2022). DEVELOPMENT OF A NEW LAND SURFACE TEMPERATURE PRODUCT FOR IMPROVING SATELLITE-BASED EVAPOTRANSPIRATION MODELLING IN THE EUROPEAN ALPS. (Tesi di dottorato, Università degli Studi di Milano-Bicocca, 2022).

DEVELOPMENT OF A NEW LAND SURFACE TEMPERATURE PRODUCT FOR IMPROVING SATELLITE-BASED EVAPOTRANSPIRATION MODELLING IN THE EUROPEAN ALPS

BARTKOWIAK, PAULINA
2022

Abstract

The European Alps have been affected by intensification of meteorological droughts in recent years. Due to changing climatic conditions, the region is vulnerable to deviations in water cycling, which can be observed in the context of evapotranspiration (ET) anomalies. Land surface temperature (LST) is a key factor in regulating the exchange of water and energy between land and atmosphere, which directly relates it to ET. Development of two-source energy balance (TSEB) models driven by thermal remote sensing data has made a significant contribution to estimate ET at large scale. However, their coarse spatial resolution and sensitivity of TIR instruments to cloudy-sky conditions make them insufficient for complex ecosystems, such as mountain regions. To overcome these limitations, this thesis served to develop a new clear-sky land surface temperature product at 250 m spatial resolution, as an alternative to 1-km MODIS LST data, for estimating fine-resolution TSEB fluxes. In the first part of the thesis, imbalance between spatial resolution of 1-km MODIS LST data was solved by applying a sharpening procedure to obtain daily LST at 250-m spatial resolution. Due to reduced capabilities of LST–VNIR statistical models in complex ecosystems, multi-source predictors, including normalized difference vegetation index (NDVI) and digital elevation model (DEM) were used. Inspired by superiority of machine learning for non-linear problems, relationships between coarse resolution LSTs and 250-m predictors with random forest (RF) algorithm were exploited. The obtained results indicate an improvement of 20% in the agreement between Landsat and the sharpened LST compared to statistics for the original MODIS dataset. The LST models determined averaged RMSE of 2.3°C and MAE of 1.8°C. In order to reconstruct missing LSTs beneath the clouds, the author proposed a novel approach to predict invalid pixels by exploiting correlation between ground-based LST and air temperature in conjunction with auxiliary variables, e.g., downwelling solar radiation, albedo- and LAI-derived products under long-term cloudy-skies. Considering a high site dependency driven by different land-cover types, LST reconstruction was performed for aggregated stations that represent three vegetation groups: grassland, forest and permanent crops. The gap-filling was performed with two steps: site-based LST modelling from ground-derived variables under cloudy skies, and then applying the fitted models to gridded predictors at subpixel level corresponding to the downscaled output. The reconstruction achieved reliable performance with local data yielding R2 of 0.84 and RMSE of 2.12°C. In the last part of the thesis, the resulting LST maps were incorporated into two-source energy balance model of Priestley-Taylor for estimating energy fluxes at 250-m spatial resolution. First, the performance of the model forced by local temperatures was evaluated with measured fluxes from eddy covariance towers. The benchmark simulations for latent (LE) and sensible heat (H) yielded an averaged RMSE of 57 Wm-2 and mean absolute bias (MB) of 26 Wm-2. Next, the model estimates driven by satellite-based LSTs, i.e., original 1-km MODIS LST product and downscaled maps, were validated against in-situ data. Turbulent fluxes modelled with 250-m LSTs resulted in RMSE of 86 Wm-2 and MB of 55 Wm-2, which translated to 8% and 15% decrease in the respective errors when compared to TSEB estimates combined with original MODIS LST.
COLOMBO, ROBERTO
CITTERIO, SANDRA
CASTELLI, MARIAPINA
downscaling; random forest; cloudy-sky condition; evapotranspiration; TSEB
downscaling; random forest; cloudy-sky condition; evapotranspiration; TSEB
GEO/11 - GEOFISICA APPLICATA
English
21-apr-2022
SCIENZE CHIMICHE, GEOLOGICHE E AMBIENTALI
34
2020/2021
open
(2022). DEVELOPMENT OF A NEW LAND SURFACE TEMPERATURE PRODUCT FOR IMPROVING SATELLITE-BASED EVAPOTRANSPIRATION MODELLING IN THE EUROPEAN ALPS. (Tesi di dottorato, Università degli Studi di Milano-Bicocca, 2022).
File in questo prodotto:
File Dimensione Formato  
phd_unimib_848463.pdf

accesso aperto

Descrizione: Bartkowiak Paulina Matricola: 848463
Tipologia di allegato: Doctoral thesis
Dimensione 15.42 MB
Formato Adobe PDF
15.42 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/10281/374721
Citazioni
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
Social impact