Scholarlink

<p>This study evaluates the effectiveness of the Harmonic Regression (HR) model for seasonal flood classification using Synthetic Aperture Radar (SAR) and optical data. The HR model reduces noise and extracts SAR features for flood detection, while optical images aid in segmentation and validation. Achieving over 90% accuracy, it effectively classifies shallow and deep floods, non-flooded areas, and other non-seasonal objects in agricultural regions. Compared to traditional flood classification methods, this approach provides a more refined distinction between flood types, enhancing flood mapping precision. The method was tested in Vietnam’s upper Mekong region, a complex and human-influenced zone, demonstrating strong capability in distinguishing floodwater from other water bodies. This improves flood map accuracy and supports agricultural water management and irrigation cost estimation. While HR requires threshold adjustments for varying flood conditions, its straightforward implementation and adaptability make it a valuable approach. Future work should explore deep learning techniques and integrate multiple Sentinel-1 signals to enhance its applicability.</p> Graphical Abstract <p>This study focuses on detecting and classifying shallow and deep flooding in agricultural landscapes within Vietnam’s Upper Mekong Region by leveraging Harmonic Regression (HR) and Synthetic Aperture Radar (SAR) time series data. Tested in a complex, human-influenced flood-prone area, the proposed approach achieved over 90% accuracy in distinguishing shallow floods, deep floods, non-flooded zones, and other water bodies. Unlike conventional flood mapping techniques, which often struggle to distinguish between non-seasonal permanent water bodies and seasonal flood or irrigation patterns, this method provides a more refined classification tailored to the dynamics of agricultural landscapes. By integrating SAR time series with optical imagery, the data processing workflow significantly reduces noise, enhances flood feature extraction, and improves detection precision. In particular, the model effectively separates floodwater from permanent water bodies, offering valuable insights for agricultural water management and irrigation cost estimation. The findings contribute practical support for policymakers and stakeholders involved in flood-prone agricultural planning and risk mitigation.</p> <p></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

Refined Detection and Classification of Shallow and Deep Flooding in Agricultural Landscapes: Integrating Harmonic Regression with SAR Time Series

  • 期刊论文

  • 发表日期：

    2025年6月3日

查看全文
  • Tuong Quang Vo,
  • Luan Hong Pham,
  • Khuong H. Tran,
  • Phat Huu Pham,
  • Van P. D. Tri,
  • Seung-Oh Lee,
  • Hong Joon Shin,
  • Jongho Kim

摘要

This study evaluates the effectiveness of the Harmonic Regression (HR) model for seasonal flood classification using Synthetic Aperture Radar (SAR) and optical data. The HR model reduces noise and extracts SAR features for flood detection, while optical images aid in segmentation and validation. Achieving over 90% accuracy, it effectively classifies shallow and deep floods, non-flooded areas, and other non-seasonal objects in agricultural regions. Compared to traditional flood classification methods, this approach provides a more refined distinction between flood types, enhancing flood mapping precision. The method was tested in Vietnam’s upper Mekong region, a complex and human-influenced zone, demonstrating strong capability in distinguishing floodwater from other water bodies. This improves flood map accuracy and supports agricultural water management and irrigation cost estimation. While HR requires threshold adjustments for varying flood conditions, its straightforward implementation and adaptability make it a valuable approach. Future work should explore deep learning techniques and integrate multiple Sentinel-1 signals to enhance its applicability.

Graphical Abstract

This study focuses on detecting and classifying shallow and deep flooding in agricultural landscapes within Vietnam’s Upper Mekong Region by leveraging Harmonic Regression (HR) and Synthetic Aperture Radar (SAR) time series data. Tested in a complex, human-influenced flood-prone area, the proposed approach achieved over 90% accuracy in distinguishing shallow floods, deep floods, non-flooded zones, and other water bodies. Unlike conventional flood mapping techniques, which often struggle to distinguish between non-seasonal permanent water bodies and seasonal flood or irrigation patterns, this method provides a more refined classification tailored to the dynamics of agricultural landscapes. By integrating SAR time series with optical imagery, the data processing workflow significantly reduces noise, enhances flood feature extraction, and improves detection precision. In particular, the model effectively separates floodwater from permanent water bodies, offering valuable insights for agricultural water management and irrigation cost estimation. The findings contribute practical support for policymakers and stakeholders involved in flood-prone agricultural planning and risk mitigation.