Scholarlink

<p>Flood mapping is critical for effective disaster management and risk mitigation. This study evaluates the performance of Sentinel-1, Sentinel-2, and Landsat satellite data for flood detection within the Google Earth Engine platform, using Random Forest (RF) and Minimum Distance (MD) classifiers. The analysis focuses on two distinct environments: Tetouan, Morocco (urban) and Matam, Senegal (rural floodplain). Key validation metrics, including Area Under the Curve - Receiver Operating Characteristic (AUC-ROC), Intersection over Union (IoU), Overlap Percentage, and Commission/Omission Errors, were employed to assess classification accuracy. The results were validated against UNOSAT benchmark flood maps, ensuring robust accuracy assessments. In rural Matam, Sentinel-2 demonstrated superior performance, achieving the highest AUC-ROC (~ 99%), IoU (~ 85%), and minimal commission (~ 1.4%) and omission errors (~ 14%). In contrast, Sentinel-1 underperformed in Matam, with significant underestimation of flood extents (~ 54–60% error) due to radar signal limitations in vegetated floodplains. Conversely, in urban Tetouan, Sentinel-1 outperformed optical datasets, providing inundated area estimates closest to the UNOSAT benchmark despite moderate omission errors (~ 40–44%). Sentinel-2 and Landsat showed considerable underestimation in urban areas, primarily due to challenges in detecting water bodies in densely built environments. The integration of RF and MD classifiers significantly improved flood detection accuracy, leveraging the complementary strengths of ensemble learning and boundary refinement. Additionally, iterative training sample optimization enhanced model stability, with the best results observed at 300–400 training samples. These findings underscore the critical role of multi-sensor integration and optimized machine learning techniques in enhancing flood mapping accuracy, offering valuable insights for disaster management in diverse environments.</p> Graphical Abstract <p>This study presents an inclusive framework for flood mapping by integrating multi-source satellite imagery and machine learning techniques, focusing on two hydrologically and geographically distinct African regions: urban Tetouan (Morocco) and rural Matam (Senegal). The graphical abstract illustrates the end-to-end workflow, beginning with data acquisition from Sentinel-1 (SAR), Sentinel-2, and Landsat sensors. These datasets were processed using Google Earth Engine and Python-Colab environments, where Random Forest (RF) and Minimum Distance (MD) classifiers were applied for flood detection. The resulting flood maps provide a comparative assessment of sensor performance and classifier accuracy, offering scalable insights for diverse geographical contexts. This work underscores the value of harmonizing optical and radar satellite data to improve flood monitoring capabilities in data-scarce regions. Ultimately, this study contributes significantly to supporting flood-risk management, aiding policymakers, and enhancing climate adaptation strategies for vulnerable communities—thereby promoting sustainable watershed management and strengthening environmental resilience in at-risk regions of Africa.</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

Enhancing Flood Mapping Accuracy in North and West Africa Using Multi-Sensor Satellite Data and Machine Learning in Google Earth Engine

  • 期刊论文

  • 发表日期：

    2025年9月9日

查看全文
  • Yassine Loukili,
  • Younes Lakhrissi,
  • Imad Bourian,
  • Lahcen Hassine,
  • Zakaria Kourab,
  • Tarig Ali,
  • Ahmed F. Elaksher,
  • Rabin Chakrabortty

摘要

Flood mapping is critical for effective disaster management and risk mitigation. This study evaluates the performance of Sentinel-1, Sentinel-2, and Landsat satellite data for flood detection within the Google Earth Engine platform, using Random Forest (RF) and Minimum Distance (MD) classifiers. The analysis focuses on two distinct environments: Tetouan, Morocco (urban) and Matam, Senegal (rural floodplain). Key validation metrics, including Area Under the Curve - Receiver Operating Characteristic (AUC-ROC), Intersection over Union (IoU), Overlap Percentage, and Commission/Omission Errors, were employed to assess classification accuracy. The results were validated against UNOSAT benchmark flood maps, ensuring robust accuracy assessments. In rural Matam, Sentinel-2 demonstrated superior performance, achieving the highest AUC-ROC (~ 99%), IoU (~ 85%), and minimal commission (~ 1.4%) and omission errors (~ 14%). In contrast, Sentinel-1 underperformed in Matam, with significant underestimation of flood extents (~ 54–60% error) due to radar signal limitations in vegetated floodplains. Conversely, in urban Tetouan, Sentinel-1 outperformed optical datasets, providing inundated area estimates closest to the UNOSAT benchmark despite moderate omission errors (~ 40–44%). Sentinel-2 and Landsat showed considerable underestimation in urban areas, primarily due to challenges in detecting water bodies in densely built environments. The integration of RF and MD classifiers significantly improved flood detection accuracy, leveraging the complementary strengths of ensemble learning and boundary refinement. Additionally, iterative training sample optimization enhanced model stability, with the best results observed at 300–400 training samples. These findings underscore the critical role of multi-sensor integration and optimized machine learning techniques in enhancing flood mapping accuracy, offering valuable insights for disaster management in diverse environments.

Graphical Abstract

This study presents an inclusive framework for flood mapping by integrating multi-source satellite imagery and machine learning techniques, focusing on two hydrologically and geographically distinct African regions: urban Tetouan (Morocco) and rural Matam (Senegal). The graphical abstract illustrates the end-to-end workflow, beginning with data acquisition from Sentinel-1 (SAR), Sentinel-2, and Landsat sensors. These datasets were processed using Google Earth Engine and Python-Colab environments, where Random Forest (RF) and Minimum Distance (MD) classifiers were applied for flood detection. The resulting flood maps provide a comparative assessment of sensor performance and classifier accuracy, offering scalable insights for diverse geographical contexts. This work underscores the value of harmonizing optical and radar satellite data to improve flood monitoring capabilities in data-scarce regions. Ultimately, this study contributes significantly to supporting flood-risk management, aiding policymakers, and enhancing climate adaptation strategies for vulnerable communities—thereby promoting sustainable watershed management and strengthening environmental resilience in at-risk regions of Africa.