Scholarlink

<p>Riyadh, the capital of Saudi Arabia, is currently undergoing one of the most rapid periods of urbanization and growth in the Gulf region. An accurate assessment of Land Use/Land Cover (LULC) mapping faces significant challenges in a unique semi-arid environment. Traditional pixel-based remote sensing classification methods often struggle in such contexts due to spectral confusion, particularly between built-up areas and desert landscapes. For this reason, this study aims to evaluate and investigate the performance and limitations of supervised (Maximum Likelihood) and unsupervised (K-means) classification techniques using Sentinel-2 level-2 imagery at 10&#xa0;m resolution for semi-arid to arid regions and propose the most effective approach. Despite the moderate performance of these methods, a notable research gap exists in applying advanced techniques tailored for arid environments, where vegetation classification remains particularly challenging, yielding user accuracies of 28–44%. The findings indicate that while supervised classification outperformed unsupervised methods, achieving an overall accuracy of 64% (Kappa = 0.52) compared to 49% (Kappa = 0.32) for unsupervised approaches, both methods struggled with classifying spectrally similar regions, such as built-up and desert areas, yielding user accuracies of 28–44%. This research suggests that exploring advanced deep learning methods, specifically convolutional neural networks, could potentially address the limitations observed in this study and enhance classification accuracy in Riyadh’s complex urban landscape This study not only establishes a critical benchmark for future LULC studies in semi-arid regions but also advocates for innovative methodologies that can significantly contribute to sustainable urban planning amidst rapid expansion, ultimately addressing the unique challenges of arid landscapes.</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

Uncertainties in supervised or unsupervised classification using sentinel-2 imagery for land use detection in the semi-arid region

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月1日

查看全文
  • Shahadat Baser,
  • Muhammad Bilal,
  • Ardiansyah Koeshidayatullah

摘要

Riyadh, the capital of Saudi Arabia, is currently undergoing one of the most rapid periods of urbanization and growth in the Gulf region. An accurate assessment of Land Use/Land Cover (LULC) mapping faces significant challenges in a unique semi-arid environment. Traditional pixel-based remote sensing classification methods often struggle in such contexts due to spectral confusion, particularly between built-up areas and desert landscapes. For this reason, this study aims to evaluate and investigate the performance and limitations of supervised (Maximum Likelihood) and unsupervised (K-means) classification techniques using Sentinel-2 level-2 imagery at 10 m resolution for semi-arid to arid regions and propose the most effective approach. Despite the moderate performance of these methods, a notable research gap exists in applying advanced techniques tailored for arid environments, where vegetation classification remains particularly challenging, yielding user accuracies of 28–44%. The findings indicate that while supervised classification outperformed unsupervised methods, achieving an overall accuracy of 64% (Kappa = 0.52) compared to 49% (Kappa = 0.32) for unsupervised approaches, both methods struggled with classifying spectrally similar regions, such as built-up and desert areas, yielding user accuracies of 28–44%. This research suggests that exploring advanced deep learning methods, specifically convolutional neural networks, could potentially address the limitations observed in this study and enhance classification accuracy in Riyadh’s complex urban landscape This study not only establishes a critical benchmark for future LULC studies in semi-arid regions but also advocates for innovative methodologies that can significantly contribute to sustainable urban planning amidst rapid expansion, ultimately addressing the unique challenges of arid landscapes.