Scholarlink

<p>The Brahmaputra River in India experiences recurrent annual floods, causing widespread ecological disruption and economic loss. With recent advancements in satellite-based microwave remote sensing, particularly Synthetic Aperture Radar (SAR), it is now possible to monitor floods at higher spatial and temporal resolutions. However, accurately delineating flood extents in hydrologically complex and vegetated terrains remains a significant challenge. This study evaluates and compares the performance of six classification techniques, Index Approach, EM (Expectation-Maximization) Cluster, K-means, Random Forest, Maximum Likelihood, and Gray Level Co-occurrence Matrix (GLCM), for flood mapping in Assam’s Nagaon district using multi-temporal SAR data from three key dates: 22 July 2020, 18 June 2022, and 30 June 2022. The analysis includes VV, VH, and combined VV + VH polarizations to assess their impact on classification accuracy and flood extent estimation. Results show that the K-means and Random Forest approaches consistently outperformed other methods, with maximum accuracies of 97.87% and 97.36%, respectively, when both polarizations were used. The GLCM approach also showed high performance, particularly when texture features were enhanced by combining polarizations. Conversely, the EM Cluster and Index-based methods recorded lower accuracy and greater misclassification, especially under single-band polarizations. Flood extent estimates derived from K-means and Random Forest were closely aligned with reference data from the National Remote Sensing Centre (NRSC), further confirming their reliability. This study underscores the advantages of multi-polarization SAR data and advanced classification techniques for accurate flood delineation, and recommends future integration of hybrid models and deep learning frameworks for real-time, scalable flood monitoring and disaster management applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparative analysis of classification techniques for flood mapping using Sentinel 1 SAR data

  • 期刊论文

  • 发表日期：

    2025年7月2日

查看全文
  • Supraja Irukumati,
  • Anantha Rama Vittal,
  • Santhosh Lakkur Gurunarayan

摘要

The Brahmaputra River in India experiences recurrent annual floods, causing widespread ecological disruption and economic loss. With recent advancements in satellite-based microwave remote sensing, particularly Synthetic Aperture Radar (SAR), it is now possible to monitor floods at higher spatial and temporal resolutions. However, accurately delineating flood extents in hydrologically complex and vegetated terrains remains a significant challenge. This study evaluates and compares the performance of six classification techniques, Index Approach, EM (Expectation-Maximization) Cluster, K-means, Random Forest, Maximum Likelihood, and Gray Level Co-occurrence Matrix (GLCM), for flood mapping in Assam’s Nagaon district using multi-temporal SAR data from three key dates: 22 July 2020, 18 June 2022, and 30 June 2022. The analysis includes VV, VH, and combined VV + VH polarizations to assess their impact on classification accuracy and flood extent estimation. Results show that the K-means and Random Forest approaches consistently outperformed other methods, with maximum accuracies of 97.87% and 97.36%, respectively, when both polarizations were used. The GLCM approach also showed high performance, particularly when texture features were enhanced by combining polarizations. Conversely, the EM Cluster and Index-based methods recorded lower accuracy and greater misclassification, especially under single-band polarizations. Flood extent estimates derived from K-means and Random Forest were closely aligned with reference data from the National Remote Sensing Centre (NRSC), further confirming their reliability. This study underscores the advantages of multi-polarization SAR data and advanced classification techniques for accurate flood delineation, and recommends future integration of hybrid models and deep learning frameworks for real-time, scalable flood monitoring and disaster management applications.