Integrating Machine Learning and Geospatial Analysis for Land Surface Temperature Reconstruction and Climate Impact Assessment on Forest Fire Dynamics in Alberta
Date:
Abstract: Accurate Land Surface Temperature (LST) data are essential for comprehending the interactions among climate variables, vegetation dynamics, and forest fire occurrences. This study introduces a machine learning framework employing CatBoost and XGBoost models to reconstruct LST across diverse land cover classes in Alberta, Canada. On the test dataset, the models demonstrated robust predictive performance: for LST-Day data, CatBoost and XGBoost achieved Median Absolute Errors (MedAE) of approximately 1.434 °C and 1.425 °C, respectively; for LST-Night data, MedAE values were around 1.186 °C for CatBoost and 1.176 °C for XGBoost. Beyond LST reconstruction, the study examines the relationships between climatic variables—LST, precipitation, and relative humidity—and forest fire occurrences across Alberta's natural subregions. The analysis reveals that elevated LST, combined with decreased precipitation and relative humidity, correlates with increased forest fire activity and subsequent vegetation changes, particularly in the Central Mixedwood, Dry Mixedwood, and Montane subregions. These findings align with observed warming and drying trends in these areas, suggesting that climate-induced alterations may amplify fire regimes and influence vegetation composition. Such shifts have significant implications for biodiversity, ecosystem services, and carbon sequestration. The integration of machine learning techniques with geospatial analysis offers a comprehensive approach to LST reconstruction and climate impact assessment on forest fire dynamics. This methodology provides valuable insights for the Geographic Information Systems (GIS) community, enhancing the understanding of climate-fire-vegetation interactions. The findings are instrumental for developing adaptive forest management strategies aimed at mitigating the adverse effects of climate change on fire regimes and vegetation dynamics in Alberta's diverse landscapes.
Conference page: https://www.mogiscon.com/