Global lightning-ignited wildfires prediction and climate change projections based on explainable machine learning...

Area of Science:

• Environmental Science
• Climate Science
• Computational Science

Background:

• Wildfires are a major natural disaster and contributor to climate change, with extreme events increasing in frequency.
• Lightning-ignited wildfires, while less frequent globally than human-caused ones, significantly impact carbon emissions and burned areas in specific regions.
• Existing predictive models for lightning-ignited wildfires are often region-specific, limiting their global applicability.

Purpose of the Study:

• To develop and present machine learning models for characterizing and predicting lightning-ignited wildfires on a global scale.
• To differentiate between lightning-ignited and anthropogenic wildfires and accurately estimate ignition probability based on various factors.
• To analyze the impact of climate change on the spatial and seasonal trends of lightning-ignited wildfires.

Main Methods:

• Development of machine learning models for global wildfire prediction.
• Classification of wildfires into lightning-ignited versus anthropogenic origins.
• Estimation of lightning fire ignition probability using meteorological and vegetation data.
• Application of eXplainable Artificial Intelligence (XAI) to analyze model feature influence.

Main Results:

• Significant global differences identified between anthropogenic and lightning-ignited wildfires.
• Machine learning models achieved high accuracy in classifying fire origins and predicting ignition probability.
• Analysis revealed that climate change has demonstrably increased the global risk of lightning-ignited wildfires within a decade.

Conclusions:

• Dedicated predictive models and fire weather indices are crucial for distinguishing and managing different wildfire types.
• The increasing risk of lightning-ignited wildfires due to climate change necessitates tailored mitigation and prediction strategies.
• Global-scale machine learning models offer a powerful tool for understanding and addressing wildfire dynamics in a changing climate.