Scholarlink

<p>This review synthesizes recent advances in deep learning and satellite remote sensing for environmental disaster detection, with a specific focus on Kazakhstan. Drawing from 107 peer-reviewed studies (2018–mid-2025) identified through Scopus, Web of Science, and IEEE Xplore, we analyze DL applications across five major hazards: floods, wildfires, oil spills, drought, and land degradation using optical and synthetic aperture radar (SAR) imagery. Key architectures include convolutional neural networks, U-Net variants, and multi-modal sensor fusion models, with reported performance gains such as up to 18% improvement in mean intersection-over-union via SAR-optical fusion. We highlight Kazakhstan-specific challenges, including snow-water spectral confusion in the Normalized Difference Water Index (NDWI), NDVI saturation in steppe environments, and acute scarcity of locally labeled training data. Using high-resolution imagery from the national KazEOSat-1 system and Sentinel missions, we illustrate gaps in regional model adaptation through case studies of the 2024 Ural River floods and Aral Sea desertification. A comparative framework of data sources, models, and metrics is proposed to guide localized hazard analysis. We outline practical future directions including cross-regional transfer learning, multimodal SAR–optical fusion, and cloud-native processing pipelines tailored to Central Asia.</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

Satellite based deep learning approaches for detecting environmental disasters across Kazakhstan

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月24日

查看全文
  • Marat Nurtas,
  • Serik Nurakynov,
  • Aizhan Altaibek,
  • Aidana Mergembayeva,
  • Mohammed Atef Mohammed

摘要

This review synthesizes recent advances in deep learning and satellite remote sensing for environmental disaster detection, with a specific focus on Kazakhstan. Drawing from 107 peer-reviewed studies (2018–mid-2025) identified through Scopus, Web of Science, and IEEE Xplore, we analyze DL applications across five major hazards: floods, wildfires, oil spills, drought, and land degradation using optical and synthetic aperture radar (SAR) imagery. Key architectures include convolutional neural networks, U-Net variants, and multi-modal sensor fusion models, with reported performance gains such as up to 18% improvement in mean intersection-over-union via SAR-optical fusion. We highlight Kazakhstan-specific challenges, including snow-water spectral confusion in the Normalized Difference Water Index (NDWI), NDVI saturation in steppe environments, and acute scarcity of locally labeled training data. Using high-resolution imagery from the national KazEOSat-1 system and Sentinel missions, we illustrate gaps in regional model adaptation through case studies of the 2024 Ural River floods and Aral Sea desertification. A comparative framework of data sources, models, and metrics is proposed to guide localized hazard analysis. We outline practical future directions including cross-regional transfer learning, multimodal SAR–optical fusion, and cloud-native processing pipelines tailored to Central Asia.