Scholarlink

<p>Detecting extreme climate events such as heatwaves, cold spells, and heavy rainfall is crucial for disaster preparedness and risk management. This study develops a multi-model machine learning framework integrating Random Forest (RF), XGBoost (XGB), and Multi-Layer Perceptron (MLP) to classify extreme weather events in mainland Southeast Asia. Using ERA5-Land reanalysis data (0.1° resolution), an ensemble learning approach enhances prediction accuracy and classification stability. Results show that tree-based models (RF and XGB) outperform MLP, achieving perfect precision, recall, F1-score, and accuracy (1.00) across all event categories, while MLP struggles with class imbalance, particularly in detecting heatwaves and heavy rainfall. The ensemble model reduces false positives by 33.3%, improves classification stability by 46.7%, and increases confidence by 10.2%. Feature analysis highlights event intensity and duration as key predictors, with heatwave intensity (0.19) and maximum temperature (0.17) most influential for heatwaves, while minimum temperature (0.14) and cold spell duration (0.13) drive cold spell classification. Rainfall intensity (0.21) and precipitation-temperature interactions (0.18) are crucial for heavy rain detection. Spatial analysis reveals that heatwaves dominate southern lowlands, cold spells concentrate in northern highlands, and heavy rainfall peaks in coastal regions, emphasizing regional climate risks. Long-term trends (1980–2014) indicate rising maximum temperatures (0.01&#xa0;°C/year) and increasing rainfall variability, intensifying extreme events. This ensemble-based framework supports real-time climate monitoring and early warning systems, with future work focusing on probabilistic forecasting and climate adaptation strategies.</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 extreme weather event classification ensemble machine learning modelling: an application to Mainland Southeast Asia region

  • 期刊论文

  • 发表日期：

    2025年12月12日

查看全文
  • Teerachai Amnuaylojaroen

摘要

Detecting extreme climate events such as heatwaves, cold spells, and heavy rainfall is crucial for disaster preparedness and risk management. This study develops a multi-model machine learning framework integrating Random Forest (RF), XGBoost (XGB), and Multi-Layer Perceptron (MLP) to classify extreme weather events in mainland Southeast Asia. Using ERA5-Land reanalysis data (0.1° resolution), an ensemble learning approach enhances prediction accuracy and classification stability. Results show that tree-based models (RF and XGB) outperform MLP, achieving perfect precision, recall, F1-score, and accuracy (1.00) across all event categories, while MLP struggles with class imbalance, particularly in detecting heatwaves and heavy rainfall. The ensemble model reduces false positives by 33.3%, improves classification stability by 46.7%, and increases confidence by 10.2%. Feature analysis highlights event intensity and duration as key predictors, with heatwave intensity (0.19) and maximum temperature (0.17) most influential for heatwaves, while minimum temperature (0.14) and cold spell duration (0.13) drive cold spell classification. Rainfall intensity (0.21) and precipitation-temperature interactions (0.18) are crucial for heavy rain detection. Spatial analysis reveals that heatwaves dominate southern lowlands, cold spells concentrate in northern highlands, and heavy rainfall peaks in coastal regions, emphasizing regional climate risks. Long-term trends (1980–2014) indicate rising maximum temperatures (0.01 °C/year) and increasing rainfall variability, intensifying extreme events. This ensemble-based framework supports real-time climate monitoring and early warning systems, with future work focusing on probabilistic forecasting and climate adaptation strategies.