Scholarlink

<p>Small inland water bodies in Morocco, locally known as Dayas, serve vital socio-economic and ecological functions. Their sustainable management has become crucial due to recurring drought conditions; however, conventional monitoring methods are proving unsuitable. This study evaluated ten multispectral water indices for Daya surface water detection using Landsat imagery processed through GEE across three study locations. We compared two thresholding approaches: zero reference threshold (ZTR) and optimal thresholds determined through iterative testing with step values of 0.1 and 0.01. In addition to indices, three machine learning (ML) methods (Random Forest, SVM, and CART) were also implemented for comparison. Results were validated using ground reference data categorized into water and non-water classes, with accuracy assessed through Overall Accuracy (OA) and Kappa coefficients. The optimal thresholding method outperformed the ZTR across all study locations. The optimal thresholding method has demonstrated satisfactory performance compared with ML classification methods. However, when applying mean optimal thresholds for operational consistency, only three indices (MNDWI, MBWI, and WRI) maintained high accuracy (OA &gt; 97%, Kappa &gt; 92%) across all sites. These findings indicate that MNDWI, MBWI, and WRI are the most reliable indices for mapping Daya water surfaces in the study region when using optimized thresholds. Future research should focus on evaluating these indices for temporal monitoring using multi-sensor approaches to enhance our understanding of Daya dynamics under changing environmental conditions.</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

Evaluating the performance of multispectral indices and machine learning for extracting small-scale, non-permanent inland water bodies (Dayas) in Western Morocco

  • 期刊论文

  • 发表日期：

    2025年8月18日

查看全文
  • Rachid Boutafoust,
  • Abdelmejid Rahimi,
  • Abdelkrim Bouasria,
  • Yassine Bouslihim,
  • Mohammed Bounif

摘要

Small inland water bodies in Morocco, locally known as Dayas, serve vital socio-economic and ecological functions. Their sustainable management has become crucial due to recurring drought conditions; however, conventional monitoring methods are proving unsuitable. This study evaluated ten multispectral water indices for Daya surface water detection using Landsat imagery processed through GEE across three study locations. We compared two thresholding approaches: zero reference threshold (ZTR) and optimal thresholds determined through iterative testing with step values of 0.1 and 0.01. In addition to indices, three machine learning (ML) methods (Random Forest, SVM, and CART) were also implemented for comparison. Results were validated using ground reference data categorized into water and non-water classes, with accuracy assessed through Overall Accuracy (OA) and Kappa coefficients. The optimal thresholding method outperformed the ZTR across all study locations. The optimal thresholding method has demonstrated satisfactory performance compared with ML classification methods. However, when applying mean optimal thresholds for operational consistency, only three indices (MNDWI, MBWI, and WRI) maintained high accuracy (OA > 97%, Kappa > 92%) across all sites. These findings indicate that MNDWI, MBWI, and WRI are the most reliable indices for mapping Daya water surfaces in the study region when using optimized thresholds. Future research should focus on evaluating these indices for temporal monitoring using multi-sensor approaches to enhance our understanding of Daya dynamics under changing environmental conditions.