Rivers are dynamic systems that play a crucial role in supporting biodiversity and providing essential ecosystem services, such as water purification, flood regulation, and habitat maintenance. Understanding river geomorphology and its evolution is critical, particularly in the context of climate change and increasing anthropogenic pressures. This study investigates the application of advanced remote sensing and deep learning techniques to analyze river dynamics, with a focus on mapping active channel changes and assessing the impact of extreme hydrological events. The research integrates satellite imagery with deep learning methodologies, employing a Fully Convolutional Network classifier to detect and analyze changes in rivers active channel in the Emilia-Romagna region, which was severely affected by the spring 2023 floods. The study evaluates the model's performance under challenging conditions, identifying both its capabilities and limitations. By comparing active channel density maps between July 2022 and July 2023, the analysis reveals key geomorphological processes, including the reduction of active river channels due to vegetation growth and significant channel modifications after flooding. However, classification errors - such as the misidentification of lakes as river channels and the underestimation of narrow watercourses due to resolution constraints - are also observed and discussed. To address these limitations, potential improvements are proposed, including expanding the training dataset, refining the model, and delineating the active channel over the entire year satellite imagery available. The findings emphasize the value of remote sensing and deep learning for advancing fluvial geomorphology studies and river management.
I fiumi sono sistemi dinamici che svolgono un ruolo cruciale nel supportare la biodiversità e nel fornire servizi ecosistemici essenziali, come la purificazione dell’acqua, la controllo delle inondazioni e il mantenimento degli habitat. Comprendere la geomorfologia fluviale e la sua evoluzione è fondamentale, soprattutto nel contesto del cambiamento climatico e delle crescenti pressioni antropiche. Questo studio indaga l’applicazione di tecniche avanzate di telerilevamento e deep learning per analizzare la dinamica dei fiumi, con un focus sulla mappatura dei cambiamenti nei canali attivi e sulla valutazione dell’impatto degli eventi idrologici estremi. La ricerca integra immagini satellitari con metodologie di deep learning, utilizzando un classificatore Fully Convolutional Network per rilevare e analizzare i cambiamenti nei canali attivi dei fiumi nella regione Emilia-Romagna, gravemente colpita dalle inondazioni della primavera 2023. Lo studio valuta le performance del modello in condizioni difficili, identificando sia le sue capacità che le sue limitazioni. Confrontando le mappe di densità dei canali attivi tra luglio 2022 e luglio 2023, l'analisi rivela i principali processi geomorfologici, inclusa la riduzione dei canali attivi dei fiumi a causa della crescita della vegetazione e significativi cambiamenti nei canali dopo le inondazioni. Tuttavia, vengono osservati e discussi anche errori di classificazione, come l’identificazione errata dei laghi come canali fluviali e la sottovalutazione dei corsi d’acqua stretti a causa delle limitazioni di risoluzione. Per affrontare queste limitazioni, vengono proposte potenziali migliorie, tra cui l'espansione del dataset di addestramento, il perfezionamento del modello e l'uso di più immagini satellitari disponibili durante l’anno. I risultati sottolineano l'importanza del telerilevamento e del deep learning per il progresso degli studi sulla geomorfologia fluviale e la gestione dei fiumi.
Detection of river channel changes after flooding: testing satellite images and deep neural network model in Emilia-Romagna
PIOVESAN, ALESSANDRO
2024/2025
Abstract
Rivers are dynamic systems that play a crucial role in supporting biodiversity and providing essential ecosystem services, such as water purification, flood regulation, and habitat maintenance. Understanding river geomorphology and its evolution is critical, particularly in the context of climate change and increasing anthropogenic pressures. This study investigates the application of advanced remote sensing and deep learning techniques to analyze river dynamics, with a focus on mapping active channel changes and assessing the impact of extreme hydrological events. The research integrates satellite imagery with deep learning methodologies, employing a Fully Convolutional Network classifier to detect and analyze changes in rivers active channel in the Emilia-Romagna region, which was severely affected by the spring 2023 floods. The study evaluates the model's performance under challenging conditions, identifying both its capabilities and limitations. By comparing active channel density maps between July 2022 and July 2023, the analysis reveals key geomorphological processes, including the reduction of active river channels due to vegetation growth and significant channel modifications after flooding. However, classification errors - such as the misidentification of lakes as river channels and the underestimation of narrow watercourses due to resolution constraints - are also observed and discussed. To address these limitations, potential improvements are proposed, including expanding the training dataset, refining the model, and delineating the active channel over the entire year satellite imagery available. The findings emphasize the value of remote sensing and deep learning for advancing fluvial geomorphology studies and river management.File | Dimensione | Formato | |
---|---|---|---|
Piovesan_Alessandro.pdf
accesso aperto
Dimensione
5.9 MB
Formato
Adobe PDF
|
5.9 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/82503