Scholarlink

<p>Oasis fires have become increasingly frequent due to climate change, threatening fragile ecosystems and agricultural livelihoods, particularly in the semi-arid region. This study develops a data-driven fire susceptibility mapping framework for Errachidia province, Morocco, integrating Sentinel-2 satellite imagery and machine learning (ML) models, including support vector machines (SVM), random forest (RF), and eXtreme gradient boosting (XGBoost). Unlike previous studies focusing on Mediterranean and arid landscapes, this research explicitly addresses wildfire susceptibility mapping in Moroccan oasis ecosystems, where limited scientific assessments exist. The study introduces an innovative ML-based fire susceptibility mapping framework for palm oasis ecosystems, which compiles multi-source geospatial datasets, including topographic, meteorological, and anthropogenic variables, to identify fire-prone areas from 2019 to 2024. Advanced feature selection and data preprocessing techniques were employed to enhance model efficiency. The dataset was split into 70% training and 30% testing, with model performance assessed using multiple metrics, including ROC-AUC, precision, recall, and F1-score. This work’s innovative hybrid SVM–XGBoost model demonstrated superior predictive accuracy, achieving an impressive 0.996 AUC, outperforming stand-alone models. The susceptibility maps identified high-risk hotspots near Oued Ziz, Ouled Chaker, and Aoufous Oases, where vegetated loss exceeded 30 hectares based on NDVI analysis. This study contributes to wildfire risk management by providing a robust, validated, and transferable methodology for fire prediction in arid and semi-arid ecosystems. The findings offer critical insights for targeted mitigation strategies, including optimized firebreak placement and community-driven fire awareness programs. Integrating remote sensing with ML, the proposed framework enhances early warning systems and adaptive wildfire resilience strategies, making it applicable to global fire-prone ecosystems facing similar climatic challenges.</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

Hybrid SVM–XGBoost framework for wildfire susceptibility mapping in palm oasis ecosystems to enhance risk assessment in arid and semi-arid regions

  • 期刊论文

  • 发表日期：

    2026年1月3日

查看全文
  • Brahim Benzougagh,
  • Youssef Bammou,
  • Halah Kadhim Tayyeh,
  • Ahmed Mageed Hussein,
  • Mohammed El Brahimi,
  • Tarik Amraoui,
  • Driss Sadkaoui,
  • Khaled Mohamed Khedher,
  • Salma Abdullah Hassan Alghurabi,
  • Khadeijah Yahya Faqeih,
  • Lizny Jaufer,
  • Shuraik Kader

摘要

Oasis fires have become increasingly frequent due to climate change, threatening fragile ecosystems and agricultural livelihoods, particularly in the semi-arid region. This study develops a data-driven fire susceptibility mapping framework for Errachidia province, Morocco, integrating Sentinel-2 satellite imagery and machine learning (ML) models, including support vector machines (SVM), random forest (RF), and eXtreme gradient boosting (XGBoost). Unlike previous studies focusing on Mediterranean and arid landscapes, this research explicitly addresses wildfire susceptibility mapping in Moroccan oasis ecosystems, where limited scientific assessments exist. The study introduces an innovative ML-based fire susceptibility mapping framework for palm oasis ecosystems, which compiles multi-source geospatial datasets, including topographic, meteorological, and anthropogenic variables, to identify fire-prone areas from 2019 to 2024. Advanced feature selection and data preprocessing techniques were employed to enhance model efficiency. The dataset was split into 70% training and 30% testing, with model performance assessed using multiple metrics, including ROC-AUC, precision, recall, and F1-score. This work’s innovative hybrid SVM–XGBoost model demonstrated superior predictive accuracy, achieving an impressive 0.996 AUC, outperforming stand-alone models. The susceptibility maps identified high-risk hotspots near Oued Ziz, Ouled Chaker, and Aoufous Oases, where vegetated loss exceeded 30 hectares based on NDVI analysis. This study contributes to wildfire risk management by providing a robust, validated, and transferable methodology for fire prediction in arid and semi-arid ecosystems. The findings offer critical insights for targeted mitigation strategies, including optimized firebreak placement and community-driven fire awareness programs. Integrating remote sensing with ML, the proposed framework enhances early warning systems and adaptive wildfire resilience strategies, making it applicable to global fire-prone ecosystems facing similar climatic challenges.