Scholarlink

<p>Forest fires seriously threaten the environment and the economy. Accurate and reliable predictive models are necessary to successfully reduce forest fires. Using multispectral satellite imagery and sophisticated algorithms, recent advances in machine learning (ML) and geospatial technology have improved forest fire susceptibility mapping. This study uses multispectral satellite imagery to map forest susceptibility to forest fires using ML and Deep Learning algorithms. The following boosting algorithms were assessed: XGBoost, CatBoost, LightGBM, and AdaBoost, as well as their ensemble combinations. In addition, DL models such as DenseNet121, MobileNetV2, ResNet, and EfficientNetB0 were used. DenseNet121 was the best DL model (0.9781), while the XGBoost, CatBoost, and LightGBM ensembles had the best ML accuracy (0.9951). These results demonstrate the potential of integrating geospatial data with contemporary machine learning and deep learning algorithms to improve wildfire susceptibility prediction under the studied 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

Harnessing geospatial intelligence for wildfire risk mapping using ML and DL techniques

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月27日

查看全文
  • Priyanka Vibhandik,
  • Suraj Sawant,
  • Vishal Mishra,
  • Ranjeet Bidwe,
  • Amit Joshi

摘要

Forest fires seriously threaten the environment and the economy. Accurate and reliable predictive models are necessary to successfully reduce forest fires. Using multispectral satellite imagery and sophisticated algorithms, recent advances in machine learning (ML) and geospatial technology have improved forest fire susceptibility mapping. This study uses multispectral satellite imagery to map forest susceptibility to forest fires using ML and Deep Learning algorithms. The following boosting algorithms were assessed: XGBoost, CatBoost, LightGBM, and AdaBoost, as well as their ensemble combinations. In addition, DL models such as DenseNet121, MobileNetV2, ResNet, and EfficientNetB0 were used. DenseNet121 was the best DL model (0.9781), while the XGBoost, CatBoost, and LightGBM ensembles had the best ML accuracy (0.9951). These results demonstrate the potential of integrating geospatial data with contemporary machine learning and deep learning algorithms to improve wildfire susceptibility prediction under the studied conditions.