The idea of this thesis is implementing a Burned Area (BA) mapping algorithm into a QGIS workflow. The approach for burned area identification is based on soft computing techniques and on the integration of features (spectral bands and indices) derived from Sentinel-2 multispectral satellite images with ordered weighted average operators (OWAs). The algorithm was previously implemented only in R environment and the core objective of this thesis was to exploit the QGIS user-friendly environment for displaying and visualisation of the results. Further, the workflow created and implemented in QGIS is first tested on the previously analysed fire site Pedrógão Grande, Portugal (2017), and then run over a new Mediterranean fire site to assess the exportability of the algorithm output. The new test region is chosen among as one of the most recent fire events in Antalya, Turkey (2021). In both cases, BA maps are compared to fire reference perimeters derived from very high-resolution Planet images for accuracy assessment. In Portugal, for all applied ordered weighted averages (OWAs), the omission error (oe), ranging in [8,3-48,5%], is greater than the commission error (ce), [0.3-6.4%]. On the other hand, in Turkey, while the omission errors are in a similar range [3,6-60%] as Portugal, commission errors are significantly greater especially for OWA_almost_or (19,5%) and OWA_or (23,2%). This increase leads to a decrease of the Dice coefficient with the lowest value of 0.57 for the OWA_and operator. Despite that, the OWA_average operator, which results with the best values in Turkey, still provides an acceptable oe (8.9%) and ce (11.1) with a high Dice coefficient (90%). Finally, results of the previous study on Mediterranean regions with the same algorithm (Sali et al., 2021) and our results together show that the algorithm is successfully implemented in QGIS and it is suitable to detect burned areas in Mediterranean region even without further calibration.
L’obiettivo principale di questo lavoro di tesi è di implementare un algoritmo di mappatura delle aree bruciate in QGIS. L’approccio per il rilevamento delle aree percorse dal fuoco, che si basa sulla teoria dei sistemi fuzzy e sulle tecniche di soft computing, integra bande e indici spettrali derivati da immagini satellitari Sentinel-2, attraverso l’utilizzo di operatori OWA (ordered weighted average operators). L’algoritmo, già disponibile e codificato in R, è stato implementato come flusso di lavoro in QGIS per poter sfruttare le avanzate funzionalità di visualizzazione e processing di dati geospaziali che offre un ambiente GIS rispetto ad un linguaggio di programmazione. L’implementazione è stata verificata su un evento di incendio già precedentemente analizzato (Pedrógão Grande, Portogallo, 2017) e successivamente è stata testata su un nuovo caso di evento di incendio in Turchia (Antalya, 2021). Per entrambi i casi di test, le mappe prodotte di area bruciata sono state confrontate con perimetri di riferimento derivati da immagini ad alta risoluzione Planet per la validazione. In Portogallo, la validazione ha evidenziato errori di omissione e commissione variabili a seconda dell'operatore OWA considerato nel range [8,3-48,5%] e [0.3-6.4%], rispettivamente. In Turchia, gli errori di omissioni sono comparabili [3,6-60%] mentre la commissione è significativamente maggiore soprattutto per il caso degli operatori OWA_almost_or (19,5%) e OWA_or (23,2%). L’operatore OWA_and ha evidenziato valori più bassi del coefficiente di Dice tra tutti gli operatori testati (DC=0.57). Per il caso test in Turchia, l’operatore che ha fornito globalmente i risultati migliori è stato OWA_average con oe=8.9%, ce=11.1% e Dice=90%. I risultati prodotti in questa tesi, combinati con evidenze già riportate in letteratura (Sali et al., 2021) hanno confermato che l'algoritmo implementato in QGIS è in grado di rilevare con soddisfacente accuratezza le aree percorse dal fuoco in ambiente Mediterraneo in modo semi-automatico.
Implementing an algorithm based on soft computing for burned area mapping from Sentinel-2 data in a user-friendly QGIS workflow
ZAF, RUKEN DILARA
2020/2021
Abstract
The idea of this thesis is implementing a Burned Area (BA) mapping algorithm into a QGIS workflow. The approach for burned area identification is based on soft computing techniques and on the integration of features (spectral bands and indices) derived from Sentinel-2 multispectral satellite images with ordered weighted average operators (OWAs). The algorithm was previously implemented only in R environment and the core objective of this thesis was to exploit the QGIS user-friendly environment for displaying and visualisation of the results. Further, the workflow created and implemented in QGIS is first tested on the previously analysed fire site Pedrógão Grande, Portugal (2017), and then run over a new Mediterranean fire site to assess the exportability of the algorithm output. The new test region is chosen among as one of the most recent fire events in Antalya, Turkey (2021). In both cases, BA maps are compared to fire reference perimeters derived from very high-resolution Planet images for accuracy assessment. In Portugal, for all applied ordered weighted averages (OWAs), the omission error (oe), ranging in [8,3-48,5%], is greater than the commission error (ce), [0.3-6.4%]. On the other hand, in Turkey, while the omission errors are in a similar range [3,6-60%] as Portugal, commission errors are significantly greater especially for OWA_almost_or (19,5%) and OWA_or (23,2%). This increase leads to a decrease of the Dice coefficient with the lowest value of 0.57 for the OWA_and operator. Despite that, the OWA_average operator, which results with the best values in Turkey, still provides an acceptable oe (8.9%) and ce (11.1) with a high Dice coefficient (90%). Finally, results of the previous study on Mediterranean regions with the same algorithm (Sali et al., 2021) and our results together show that the algorithm is successfully implemented in QGIS and it is suitable to detect burned areas in Mediterranean region even without further calibration.| File | Dimensione | Formato | |
|---|---|---|---|
|
2022_4_Zaf.pdf
Open Access dal 31/03/2023
Descrizione: Ruken Dilara Zaf Thesis File
Dimensione
3.1 MB
Formato
Adobe PDF
|
3.1 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/10589/186087