Centennial-scale decline in global fire emissions driven by land use and population growth

Fires and their carbon emissions have substantial impacts on land surface, climate systems, and air quality. However, long-term datasets with detailed spatiotemporal fire records remain limited, due to insufficient understanding of the climatic, ecosystem, and societal drivers of fire processes in current process-based models. Here, we employ a data-driven approach that integrates machine learning algorithms with outputs from eight fire models within the Fire Model Intercomparison Project (FireMIP) to reconstruct fire CO₂ emissions from 1901 to 2012 and to assess the respective impacts of human activities, climate, and land cover change. Our MLA-based dataset reveals a global decline in fire-emitted CO₂ at -7.45 ± 0.12 Tg C yr^-2 (-0.29% yr^-1), mainly in South America and Africa. Land use change emerges as the primary driver, reducing fire CO₂ emissions by -6.07 ± 0.23 Tg C yr^-2, followed by population growth, which contributes -3.60 ± 0.54 Tg C yr^-2. Population growth typically suppresses fires in agricultural and urban areas but raises fire risks at rainforest edges where deforestation occurs. Although climate change has a limited impact on global fire CO₂ reduction (-0.39 ± 0.19 Tg C yr^-2), it remains a key driver for boreal fires, strongly influenced by precipitation changes. These findings underscore the need for robust data and informed management to support fire prevention and climate change mitigation efforts.