Adoption of Machine Learning Technique in Nile River Islands Classification. Issue 2 (9th March 2022)

• Record Type: Journal Article
• Authors: Kamal, Noha; El-Banna, Ahmad A. Aziz; El-Ashmawy, Nagwa
• Is Part Of: Journal of hydroinformatics. Volume 24:Issue 2(2022)
• Page Start: 464
• Page End: 480
• Publication Date: 2022-03-09
• Subjects: fluvial islands -- machine Learning (ML) -- Nile River -- Random Forest (RF) -- satellite images classification; Hydrology -- Data processing -- Periodicals; Geographic information systems -- Periodicals; Geographic information systems; Hydrology -- Data processing; Electronic journals; Periodicals; 551.480285
• DOI: 10.2166/hydro.2022.154
• Languages: English
• ISSNs: 1464-7141

Abstract

Abstract: Wider adoption of machine learning methods in water resources has the potential to greatly accelerate the efficiency and quality of analysis. The Nile River is one of the major fluvial hydro-systems in the world. Fluvial islands are present in nearly all natural and regulated rivers. The Nile River is characterized by numerous natural phenomena and human interventions represented in multiple islands characteristics. This paper investigates the formation and development of the Nile River islands in the fourth reach, which extends between Assuit and Delta barrages. A machine learning (ML) technique, with the Random Forest (RF) algorithm, has been introduced as a potential technique to replace the traditional ones, to extract and classify the land cover types and the geometrical characteristics of the Nile River islands. The assessment of the results of extracting the Nile River islands and the land cover types are included. The accuracy of the extracted boundaries of the islands is assessed using field surveying data. The classification of the islands based on the islands' geometric characteristics represented that 70% of the extracted islands are classified as Wide Island, 20% are classified as Equal Island, and 10% as Narrow Island. The islands' classification, based on the land cover, results show that there is only 5% of the islands that are urban areas, 5% of the islands are mixed class (both vegetation and urban), and the rest of the islands 90% have a vegetation land cover type. The accuracy assessment was performed using the error matrix, the results show that the overall accuracy of the land cover classification is greater than 84%. The proposed islands' classification scheme can become an important tool that provides the decision-makers with more detailed information to improve the planning of the Nile River islands development projects. Furthermore, this schema can be expanded to other climatic and topographic regions. HIGHLIGHTS: The study investigates the applicability of using the Machine Learning technique in Nile island boundaries determination out of the satellite images. The study utilizes the Random Forest algorithm to extract the land cover types for the Nile islands and the surrounding area. The study introduces new schemes for Nile island classification based on geometric and land cover aspects. Graphical Abstract