Scholarlink

Using cloud-minimized Sentinel-2 Level-2A scenes (August 2017 pre-restoration; August 2024 post-restoration), we mapped Water, Salt-flat, and Non-Water-Salt over Lake Urmia to evaluate changes around the Kani Sib inter-basin transfer within the Urmia Lake Restoration Program (ULRP). Three off-the-shelf classifiers, Random Forest (RF), Gradient Boosting (GB), and Decision Tree (DT), were trained on stratified samples. RF delivered the highest overall accuracies (2017: 87.6%; 2024: 80.7%), outperforming GB (85.1%, 78.4%) and DT (80.2%, 74.9%). A 3 × 3 majority (“sieve”) filter improved spatial coherence and raised overall accuracy by ~0.7–0.8 percentage points. Transition mapping shows localized water gains along the southern shoreline near the transfer corridor, but expansion of salt flats in the northern basin; pixel accounting yields a net open-water loss of ~88.6 km2 from 2017 to 2024. We infer that while targeted inflows locally replenished margins, basin-wide desiccation driven by abstraction and evaporation dominated. The findings motivate an integrated water-management strategy that couples controlled transfers with basin-wide demand reduction, real-time monitoring (e.g., Synthetic Aperture Radar (SAR), thermal infrared (TIR)), and community engagement to support climate-resilient governance in Lake Urmia and analogous endorheic systems.

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

Multi-temporal Sentinel-2 and Machine-Learning Assessment of Lake Urmia Restoration Dynamics

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hadi Kheiri Gharajeh,
  • Ola Elwasila Abdelrahman Yousif,
  • Daniel Sahebi

摘要

Using cloud-minimized Sentinel-2 Level-2A scenes (August 2017 pre-restoration; August 2024 post-restoration), we mapped Water, Salt-flat, and Non-Water-Salt over Lake Urmia to evaluate changes around the Kani Sib inter-basin transfer within the Urmia Lake Restoration Program (ULRP). Three off-the-shelf classifiers, Random Forest (RF), Gradient Boosting (GB), and Decision Tree (DT), were trained on stratified samples. RF delivered the highest overall accuracies (2017: 87.6%; 2024: 80.7%), outperforming GB (85.1%, 78.4%) and DT (80.2%, 74.9%). A 3 × 3 majority (“sieve”) filter improved spatial coherence and raised overall accuracy by ~0.7–0.8 percentage points. Transition mapping shows localized water gains along the southern shoreline near the transfer corridor, but expansion of salt flats in the northern basin; pixel accounting yields a net open-water loss of ~88.6 km2 from 2017 to 2024. We infer that while targeted inflows locally replenished margins, basin-wide desiccation driven by abstraction and evaporation dominated. The findings motivate an integrated water-management strategy that couples controlled transfers with basin-wide demand reduction, real-time monitoring (e.g., Synthetic Aperture Radar (SAR), thermal infrared (TIR)), and community engagement to support climate-resilient governance in Lake Urmia and analogous endorheic systems.