Scholarlink

Background <p>Subtropical pine forests dominated by <i>Pinus roxburghii</i> in northern Pakistan are increasingly vulnerable to forest fires, posing serious ecological and management challenges. This study (i) identifies the primary environmental and anthropogenic drivers of fire risk, (ii) evaluated the performance of four machine learning models Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), and eXtreme Gradient Boosting (XGBoost) for fire susceptibility prediction, and (iii) develops a spatially explicit vulnerability map for the Malakand region using 2001–2023 fire occurrence data<b>.</b></p> Methods <p>Spatial modeling was conducted using MODIS fire products (FIRMS and MCD64A1), along with topographic, climatic, vegetation, and human activity variables. Predictor importance and inter-variable relationships were analyzed prior to model training and evaluation.</p> Results <p>NDVI, skin temperature, and population density emerged as key predictors. Among the models tested, RF and XGBoost outperformed others, with RF achieving 86.2% accuracy and AUC 95.4, and XGBoost showing 87.2% accuracy and AUC 95.2. Multicollinearity analysis confirmed variable independence (Tolerance &gt; 0.1; VIF &lt; 10).</p> Conclusion <p>The resulting fire vulnerability map delineated distinct spatial patterns, identifying 6.88% of the study area as highly susceptible to fire. These findings provide actionable insights into targeted fire prevention and sustainable forest management in subtropical Himalayan ecosystems.</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

Machine learning-based forest fire vulnerability assessment in subtropical chir pine forests of Pakistan

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月11日

查看全文
  • Sultan Muhammad,
  • Syed Moazzam Nizami,
  • Qun ou Jiang,
  • Kaleem Mehmood,
  • Majid Hussain,
  • Shoaib Ahmad Anees,
  • Fahad Shahzad,
  • Waseem Razzaq Khan

摘要

Background

Subtropical pine forests dominated by Pinus roxburghii in northern Pakistan are increasingly vulnerable to forest fires, posing serious ecological and management challenges. This study (i) identifies the primary environmental and anthropogenic drivers of fire risk, (ii) evaluated the performance of four machine learning models Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), and eXtreme Gradient Boosting (XGBoost) for fire susceptibility prediction, and (iii) develops a spatially explicit vulnerability map for the Malakand region using 2001–2023 fire occurrence data.

Methods

Spatial modeling was conducted using MODIS fire products (FIRMS and MCD64A1), along with topographic, climatic, vegetation, and human activity variables. Predictor importance and inter-variable relationships were analyzed prior to model training and evaluation.

Results

NDVI, skin temperature, and population density emerged as key predictors. Among the models tested, RF and XGBoost outperformed others, with RF achieving 86.2% accuracy and AUC 95.4, and XGBoost showing 87.2% accuracy and AUC 95.2. Multicollinearity analysis confirmed variable independence (Tolerance > 0.1; VIF < 10).

Conclusion

The resulting fire vulnerability map delineated distinct spatial patterns, identifying 6.88% of the study area as highly susceptible to fire. These findings provide actionable insights into targeted fire prevention and sustainable forest management in subtropical Himalayan ecosystems.