This thesis work aimed to calibrate the FEST-EWB hydrological model developing a Bayesian approach over four grasslands sites across an aridity gradient, basing on the datasets stored in the Fluxnet database. Hence, the final objective of the thesis was to reproduce the dynamic of the water and energy fluxes, especially looking at soil moisture and evapotranspiration. The first step involved the choice of 4 eddy covariance stations within the available ones on FLUXNET 2015 database. The selection criteria were two: the vegetation had to be grassland and aridity indexes should have varied from each other. Another mandatory condition for the selection was a satisfying timespan of data harvesting for the station. By using this method, the most suitable stations were the ones located in: US (station name: Walnut Gulch Kendall Grasslands), Austria (station name: Neustift), Switlzerland (station name: Früebüel), Italy (station name: Torgnon). After the selection, for a simpler narration, the following operations are explained for a single station even if they were performed for each of them. The whole process began with the extrapolation of the desired input data from the half-hourly and daily FULLSET downloaded: data trimming (when possible) and interpolation were performed for the outliers. After the phase of raw extraction of data from the stations, satellite data of leaf area index (LAI) have been extracted from the MODIS sensor on board the Terra and Aqua satellites. Right after calculating the albedo and vegetative fraction (fveg), it was possible to prepare the model inputs by selecting a two-years timespan for the data. Once the inputs were ready, the model runs were performed to apply every possible combination of energy balance parameters and soil parameters. These combinations were performed structuring every variable with a minimum and maximum value, keeping their physical meaning, and a pre-defined step. In this way it was possible to predict the number of considered values for the same variable. Applying this procedure to all the parameters, it was then possible to calculate the number of predicted simulations (one per combination). To perform the calibration process, the post-simulations data were managed and compared to the observed values from the Fluxnet dataset: through the calculations of Root Mean Squared Error (RMSE) and Nash-Sutcliffe Efficiency (NSE) indexes for each variable it was possible to find backwards the most suitable combination of parameters. Once the calibration process was achieved with respect to the timespan selected, final configuration of parameters was ran using as input the entire timespan of data. Then it was re-calculated the NSE for every variable and qualified the effectiveness of the calibration process for each station. The thesis aims to demonstrate that through this statistical method and the input data only coming from Fluxnet and MODIS sensor is possible to calibrate and obtain satisfying results for the FEST-EWB model.
Questo lavoro di tesi ha l'obiettivo di calibrare il modello idrologico FEST-EWB sviluppando un approccio Bayesiano su quattro siti meteorologici accomunati dalla stessa vegetazione (prateria), aventi differenti indici di aridità, basandosi sui dataset archiviati nel database Fluxnet. Pertanto, l'obiettivo finale della tesi è riprodurre la dinamica dei flussi d'acqua ed energia, concentrandosi in particolare su umidità del suolo ed evapotraspirazione. Il primo passo ha coinvolto la scelta di 4 stazioni tra quelle disponibili nel database FLUXNET 2015. I criteri di selezione sono stati due: la vegetazione doveva essere prateria e gli indici di aridità dovevano variare tra loro. Un'altra condizione obbligatoria per la selezione era un periodo di raccolta dati soddisfacente per la stazione. Utilizzando questo metodo, le stazioni più adatte erano quelle situate in: Stati Uniti (nome della stazione: Walnut Gulch Kendall Grasslands), Austria (nome della stazione: Neustift), Svizzera (nome della stazione: Früebüel), Italia (nome della stazione: Torgnon). Dopo le selezioni, per una narrazione più semplice, le seguenti operazioni vengono spiegate per una singola stazione, anche se sono state eseguite per ognuna di esse. L'intero processo è iniziato con l'estrapolazione dei dati di input desiderati dai set completi scaricati a intervalli di mezz'ora e giornalieri: è stata eseguita la rifinitura dei dati (quando possibile) e l'interpolazione per gli outlier. Dopo la fase di estrazione grezza dei dati dalle stazioni, sono stati estratti i dati satellitari dell'indice di area fogliare (LAI) dal sensore MODIS a bordo dei satelliti Terra e Aqua. Subito dopo aver calcolato albedo e frazione vegetativa (fveg), è stato possibile preparare gli input del modello selezionando un intervallo temporale di due anni per i dati. Una volta pronti gli input, sono state eseguite le simulazioni del modello per applicare ogni possibile combinazione dei parametri del bilancio energetico e dei parametri del suolo. Queste combinazioni sono state eseguite strutturando ogni variabile con un valore minimo e massimo, mantenendo il loro significato fisico, e uno step di incremento predefinito. In questo modo è stato possibile prevedere il numero di valori considerati per la stessa variabile. Applicando questa procedura a tutti i parametri, è stato quindi possibile calcolare il numero di simulazioni previste (una per ogni combinazione). Per eseguire il processo di calibrazione, i dati delle post-simulazioni sono stati gestiti e confrontati con i valori osservati dal dataset Fluxnet: attraverso il calcolo degli indici di Errore Quadratico Medio (RMSE) e Efficienza di Nash-Sutcliffe (NSE) per ogni variabile, è stato possibile ricondursi alla combinazione di parametri più adatta. Una volta ottenuto il processo di calibrazione rispetto all'intervallo temporale selezionato, è stata eseguita la configurazione finale dei parametri utilizzando l'intero intervallo di dati come input. Successivamente è stato ricalcolato l'NSE per ogni variabile e valutata l'efficacia del processo di calibrazione per ciascuna stazione. La tesi mira a dimostrare che attraverso questo metodo statistico e i dati di input provenienti esclusivamente da Fluxnet e dal sensore MODIS è possibile calibrare e ottenere risultati soddisfacenti per il modello FEST-EWB.
Calibration of the FEST-EWB hydrological model through a Bayesian approach applied to four different aridity indexes grasslands
PREVIDERÈ, FEDERICO
2022/2023
Abstract
This thesis work aimed to calibrate the FEST-EWB hydrological model developing a Bayesian approach over four grasslands sites across an aridity gradient, basing on the datasets stored in the Fluxnet database. Hence, the final objective of the thesis was to reproduce the dynamic of the water and energy fluxes, especially looking at soil moisture and evapotranspiration. The first step involved the choice of 4 eddy covariance stations within the available ones on FLUXNET 2015 database. The selection criteria were two: the vegetation had to be grassland and aridity indexes should have varied from each other. Another mandatory condition for the selection was a satisfying timespan of data harvesting for the station. By using this method, the most suitable stations were the ones located in: US (station name: Walnut Gulch Kendall Grasslands), Austria (station name: Neustift), Switlzerland (station name: Früebüel), Italy (station name: Torgnon). After the selection, for a simpler narration, the following operations are explained for a single station even if they were performed for each of them. The whole process began with the extrapolation of the desired input data from the half-hourly and daily FULLSET downloaded: data trimming (when possible) and interpolation were performed for the outliers. After the phase of raw extraction of data from the stations, satellite data of leaf area index (LAI) have been extracted from the MODIS sensor on board the Terra and Aqua satellites. Right after calculating the albedo and vegetative fraction (fveg), it was possible to prepare the model inputs by selecting a two-years timespan for the data. Once the inputs were ready, the model runs were performed to apply every possible combination of energy balance parameters and soil parameters. These combinations were performed structuring every variable with a minimum and maximum value, keeping their physical meaning, and a pre-defined step. In this way it was possible to predict the number of considered values for the same variable. Applying this procedure to all the parameters, it was then possible to calculate the number of predicted simulations (one per combination). To perform the calibration process, the post-simulations data were managed and compared to the observed values from the Fluxnet dataset: through the calculations of Root Mean Squared Error (RMSE) and Nash-Sutcliffe Efficiency (NSE) indexes for each variable it was possible to find backwards the most suitable combination of parameters. Once the calibration process was achieved with respect to the timespan selected, final configuration of parameters was ran using as input the entire timespan of data. Then it was re-calculated the NSE for every variable and qualified the effectiveness of the calibration process for each station. The thesis aims to demonstrate that through this statistical method and the input data only coming from Fluxnet and MODIS sensor is possible to calibrate and obtain satisfying results for the FEST-EWB model.File | Dimensione | Formato | |
---|---|---|---|
2024_04_Previdere.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
14.22 MB
Formato
Adobe PDF
|
14.22 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/10589/218443