Global water resources are under increasing pressure due to demands from population growth and climate change. As a result, the regime of the rivers is changing and their ecosystems are threatened. Therefore, for effective water management and mitigation of hazards, it is crucial to frequently and accurately map the surface area of river water. Synthetic Aperture Radar (SAR) backscatter images at high temporal resolution are nowadays available. However, mapping the surface water of narrow water bodies, such as rivers, remains challenging due to the SAR spatial resolution (few tens of meters). Conversely, Multi-Spectral Instrument (MSI) images have a higher spatial resolution (few meters) but are affected by cloud coverage. In this paper, we present a new method for automatic detection and mapping of the surface water of rivers. The method is based on the convolutional neural network known as U-Net. To develop the proposed approach, two datasets are needed: (i) a set of Sentinel-2 MSI images, used to achieve target values; (ii) a set of Sentinel-1A SAR backscatter images, used as input values. The proposed method has been experimented to map the surface water of the Mijares river (Spain) from April 2019 to September 2022. Experimental results show that the proposed approach computes the total surface area covered by the river water with a mean absolute error equal to 0.072, which is very promising for the target application. To encourage scientific collaborations, the source code used for this work has been made publicly available.

U-Nets and Multispectral Images for Detecting the Surface Water of Rivers via SAR Images

Diana Orlandi;Federico A. Galatolo;Alessandro La Rosa;Mario G. C. A. Cimino;Pierfrancesco Foglia;Carolina Pagli;Cosimo A. Prete
2024-01-01

Abstract

Global water resources are under increasing pressure due to demands from population growth and climate change. As a result, the regime of the rivers is changing and their ecosystems are threatened. Therefore, for effective water management and mitigation of hazards, it is crucial to frequently and accurately map the surface area of river water. Synthetic Aperture Radar (SAR) backscatter images at high temporal resolution are nowadays available. However, mapping the surface water of narrow water bodies, such as rivers, remains challenging due to the SAR spatial resolution (few tens of meters). Conversely, Multi-Spectral Instrument (MSI) images have a higher spatial resolution (few meters) but are affected by cloud coverage. In this paper, we present a new method for automatic detection and mapping of the surface water of rivers. The method is based on the convolutional neural network known as U-Net. To develop the proposed approach, two datasets are needed: (i) a set of Sentinel-2 MSI images, used to achieve target values; (ii) a set of Sentinel-1A SAR backscatter images, used as input values. The proposed method has been experimented to map the surface water of the Mijares river (Spain) from April 2019 to September 2022. Experimental results show that the proposed approach computes the total surface area covered by the river water with a mean absolute error equal to 0.072, which is very promising for the target application. To encourage scientific collaborations, the source code used for this work has been made publicly available.
2024
Orlandi, Diana; Galatolo, Federico A.; LA ROSA, Alessandro; Cimino, Mario G. C. A.; Foglia, Pierfrancesco; Pagli, Carolina; Prete, Cosimo A.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

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/11568/1237398
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
social impact