Global warming intensifies wildfires and exacerbates greenhouse-gas and pollutant emissions1. However, global projections remain incomplete, hindering effective policy interventions amid uncertain warming scenarios2. Here we developed an interpretable machine learning framework to project global burned areas and wildfire emissions. This framework accounts for the effects of future climate change on fire activity and quantifies the associated number of premature deaths and radiative forcing from fire-induced particulate matter (fine particulate matter less than 2.5 μm in diameter (PM2.5)). Here we show that from 2010–2014 to 2095–2099, fire carbon emissions are projected to increase by 23% under Shared Socio-economic Pathway 2-4.5. Increased fire-related aerosols reduce the 0.06 W m−2 cooling effect north of 60° N. Projections show a surge in the number of premature deaths from wildfire smoke, reaching 1.40 (95% confidence interval = 0.66–2.25) million people annually during 2095–2099—roughly six times higher than current levels. Africa is projected to experience the greatest rise in fire-related deaths (11-fold), driven by emission changes and an ageing population. Europe and the USA would experience a one to twofold increase under Shared Socio-economic Pathway 2-4.5, linked to rising fire occurrences in the mid-latitude Northern Hemisphere. Overall, the health burden would become more evenly distributed across nations of differing development levels than present patterns, underscoring the need for coordinated efforts.
