Advanced Wildfire Prediction and Severity Estimation Across Australian Biomes Using Multisource Environmental Data and Machine Learning Models

Wildfires present significant environmental and societal challenges, making accurate prediction essential for effective mitigation strategies. This paper develops a comprehensive framework for predicting wildfire occurrence and estimating their severity using random forest (RF) and XGBoost methodologies. The framework integrates historical meteorological data, vegetation indices (NDVI), land classification, and drought indicators (SPI and SPEI). Feature engineering techniques, including lagged NDVI variables and optimized thresholds for wildfire occurrence and severity classification, enhanced model performance. The results show that XGBoost slightly performed better than RF in both occurrence prediction and severity estimation, with temperature, precipitation, NDVI trends, and wind speed emerging as the most influential predictors. The findings demonstrate the potential of machine learning for supporting proactive wildfire management and resource allocation.