Scholarlink

<p>Monitoring surface water bodies in mountainous regions is essential for multiple disciplines, including water management, sustainable mountain development, and climate change adaptation. Despite advancements in remote sensing techniques, challenges persist in precisely detecting surface water in these complex terrains, where supervised classifiers and ratio-based indices exhibit inconsistent performances. However, no comprehensive study has yet evaluated these methods specifically for mountainous areas, where issues such as terrain complexity, snow cover, and shadows further complicate water extraction. This paper systematically evaluates and compares the performance of nine commonly used ratio-based water indices and four supervised classifiers for surface water extraction across 12 experimental sites. We further assessed the reliability of four global land cover products, namely JRC, DynamicWorld, ESA, and ESRI. Our analysis showed that no single method consistently outperformed the other techniques across all 12 experimental sites; however, supervised classifiers generally achieved the highest overall performance. On average, deep convolutional neural networks emerged as the most effective supervised classifier (precision = 96.72%, recall = 93.02%, and F1-score = 94.78%). Furthermore, the new water index was the best spectral index (precision = 86.94%, recall = 93.38%, and F1-score = 89.01%), and the Sentinel-1&#xa0;A water index was the best SAR-derived index (precision = 77.63%, recall = 90.83%, and F1-score = 83.37%). Regarding global land cover products, ESA showed the best overall performance, while DynamicWorld had the weakest. This study also highlighted the primary causes of surface water misclassification in mountains utilizing different remote sensing-based techniques and discussed their associated limitations and challenges.</p> Graphical Abstract <p>The graphical abstract presents a systematic comparison of remote sensing-based methodologies, including water indices (optical and SAR-based), machine learning algorithms, and global land cover products, employed for the detection of surface water in challenging mountainous regions. The workflow begins with the selection of 12 geographically diverse mountain sites, highlighting various challenges such as mountain shadow, snow/ice, and mixed land cover. Each approach is illustrated as a component of a multi-stage workflow, encompassing data collection, pre-processing, classification, and validation. The layout presents examples of generated surface water maps using spectral indices, SAR-based indices, and supervised classifiers across 12 experimental sites. Key accuracy metrics (precision, recall, and F1-score) are shown as comparative bar charts for each method group. The deep convolutional neural networks stands out as the most reliable, while a new water index and Sentinel-1&#xa0;A water index are marked for their strong performance. The framework includes the assessment of four global land cover datasets, wherein ESA’s results regularly demonstrates superior alignment with observed water bodies. The final panel emphasizes practical takeaways: supervised classifiers are generally superior. This visual summary captures the complexity and decision-making path researchers must consider when extracting surface water from mountainous environments using Earth observation data.</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

A Comprehensive Analysis of the Different Remote Sensing-Based Approaches for Surface Water Mapping in Mountainous Areas

  • 期刊论文

  • 发表日期：

    2025年8月26日

查看全文
  • Amin Naboureh,
  • Ainong Li,
  • Jinhu Bian,
  • Guangbin Lei,
  • Xi Nan,
  • Zhengjian Zhang

摘要

Monitoring surface water bodies in mountainous regions is essential for multiple disciplines, including water management, sustainable mountain development, and climate change adaptation. Despite advancements in remote sensing techniques, challenges persist in precisely detecting surface water in these complex terrains, where supervised classifiers and ratio-based indices exhibit inconsistent performances. However, no comprehensive study has yet evaluated these methods specifically for mountainous areas, where issues such as terrain complexity, snow cover, and shadows further complicate water extraction. This paper systematically evaluates and compares the performance of nine commonly used ratio-based water indices and four supervised classifiers for surface water extraction across 12 experimental sites. We further assessed the reliability of four global land cover products, namely JRC, DynamicWorld, ESA, and ESRI. Our analysis showed that no single method consistently outperformed the other techniques across all 12 experimental sites; however, supervised classifiers generally achieved the highest overall performance. On average, deep convolutional neural networks emerged as the most effective supervised classifier (precision = 96.72%, recall = 93.02%, and F1-score = 94.78%). Furthermore, the new water index was the best spectral index (precision = 86.94%, recall = 93.38%, and F1-score = 89.01%), and the Sentinel-1 A water index was the best SAR-derived index (precision = 77.63%, recall = 90.83%, and F1-score = 83.37%). Regarding global land cover products, ESA showed the best overall performance, while DynamicWorld had the weakest. This study also highlighted the primary causes of surface water misclassification in mountains utilizing different remote sensing-based techniques and discussed their associated limitations and challenges.

Graphical Abstract

The graphical abstract presents a systematic comparison of remote sensing-based methodologies, including water indices (optical and SAR-based), machine learning algorithms, and global land cover products, employed for the detection of surface water in challenging mountainous regions. The workflow begins with the selection of 12 geographically diverse mountain sites, highlighting various challenges such as mountain shadow, snow/ice, and mixed land cover. Each approach is illustrated as a component of a multi-stage workflow, encompassing data collection, pre-processing, classification, and validation. The layout presents examples of generated surface water maps using spectral indices, SAR-based indices, and supervised classifiers across 12 experimental sites. Key accuracy metrics (precision, recall, and F1-score) are shown as comparative bar charts for each method group. The deep convolutional neural networks stands out as the most reliable, while a new water index and Sentinel-1 A water index are marked for their strong performance. The framework includes the assessment of four global land cover datasets, wherein ESA’s results regularly demonstrates superior alignment with observed water bodies. The final panel emphasizes practical takeaways: supervised classifiers are generally superior. This visual summary captures the complexity and decision-making path researchers must consider when extracting surface water from mountainous environments using Earth observation data.