Scholarlink

<p>Forest fires, a growing global threat, have increased in intensity and frequency over recent decades due to climate change and human activities. These fires may destroy biodiversity and cause environmental degradation, habitat loss, and exacerbate human–animal conflicts. Certain cultural practices in countries like India further intensify the problem, especially from March to May, when dry deciduous forests accumulate combustible biomass. Despite several challenges, the early prediction of forest fires can significantly mitigate their impacts. However, most previous studies have overlooked the dynamic nature of climatic factors in their predictive models. This research aims to fill this gap by producing a 5-day early prediction map, that incorporates climatic conditions from the preceding 6&#xa0;days, as well as vegetation health and topographic factors. In this study, three different machine learning algorithms, i.e., random forest, extreme gradient boost, and support vector machine, were evaluated to identify the most effective model for early fire prediction. The models were trained using nearly 2000 ground-based fire observations collected between 2017 and 2022, and included sixteen driving factors encompassing climatic, topographic, anthropogenic, and forest health parameters. Our findings highlight the key factors influencing fire occurrences, with all models achieving over 80% accuracy in 5-day early predictions, and Random Forest demonstrating the highest overall performance. Critical determinants for fire susceptibility included elevation, temperature, and relative humidity. The practical utility of the model was further validated through three distinct case studies during March 2022, a peak period for forest fire occurrences. These case studies achieved an area under the curve (AUC) value exceeding 85% for 5-day early predictions, confirming the effectiveness of the model. The resulting 5-day early susceptibility maps provide valuable insights for timely warnings, enabling better prevention, resource allocation and optimizing management efficiency to reduce environmental, ecological, and economic losses.</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

Early prediction of forest fire events integrating multi-sensor datasets and machine learning techniques

  • 期刊论文

  • 发表日期：

    2025年8月27日

查看全文
  • Debarati Bera,
  • Subrata Raut,
  • Chandan Mahata,
  • Dipanwita Dutta,
  • Debajit Datta,
  • Arnab Kundu

摘要

Forest fires, a growing global threat, have increased in intensity and frequency over recent decades due to climate change and human activities. These fires may destroy biodiversity and cause environmental degradation, habitat loss, and exacerbate human–animal conflicts. Certain cultural practices in countries like India further intensify the problem, especially from March to May, when dry deciduous forests accumulate combustible biomass. Despite several challenges, the early prediction of forest fires can significantly mitigate their impacts. However, most previous studies have overlooked the dynamic nature of climatic factors in their predictive models. This research aims to fill this gap by producing a 5-day early prediction map, that incorporates climatic conditions from the preceding 6 days, as well as vegetation health and topographic factors. In this study, three different machine learning algorithms, i.e., random forest, extreme gradient boost, and support vector machine, were evaluated to identify the most effective model for early fire prediction. The models were trained using nearly 2000 ground-based fire observations collected between 2017 and 2022, and included sixteen driving factors encompassing climatic, topographic, anthropogenic, and forest health parameters. Our findings highlight the key factors influencing fire occurrences, with all models achieving over 80% accuracy in 5-day early predictions, and Random Forest demonstrating the highest overall performance. Critical determinants for fire susceptibility included elevation, temperature, and relative humidity. The practical utility of the model was further validated through three distinct case studies during March 2022, a peak period for forest fire occurrences. These case studies achieved an area under the curve (AUC) value exceeding 85% for 5-day early predictions, confirming the effectiveness of the model. The resulting 5-day early susceptibility maps provide valuable insights for timely warnings, enabling better prevention, resource allocation and optimizing management efficiency to reduce environmental, ecological, and economic losses.