Assessment of Regional Fire Carbon Emissions during 2003–23 Simulated by Multiple CMIP6 Models

Wildfires play a critical role in the Earth system, affecting climate, carbon cycling, air quality, and human livelihoods. Phase 6 of the Coupled Model Intercomparison Project (CMIP6) provides multi-model simulations of historical and future wildfire carbon emissions, yet substantial uncertainties remain. Here, we evaluate fire carbon emissions from 10 CMIP6 models over 2003–23 using three satellite-based datasets (GFED, GFAS, QFED) across 14 subregions. The multi-model ensemble (MME) largely reproduces the spatial patterns, with a pattern correlation coefficient of 0.64 and normalized standard deviation of 1.00, but overestimates emissions in most regions, particularly the tropics, leading to inflated global totals. Only 5 out of 14 subregions show good agreement with observations. Seasonality is reasonably captured, with most models simulating peak fire months within ±1 month. Sensitivity analysis reveals strong regional variation in fire responses to near-surface air temperature (TAS) and surface soil moisture (SM). The MME correctly identifies TAS as the dominant driver in most high and middle latitudes but misrepresents the relative influence of TAS and SM in key tropical regions. Observationally, widespread declines in tropical fire emissions contribute to an overall global decrease (−0.02 PgC yr⁻²), but CMIP6 models fail to capture these trends, instead simulating spurious increases that drive unrealistic upward global trends (0.03 PgC yr⁻²). These discrepancies are largely attributable to inadequate representation of anthropogenic fire suppression and overestimated temperature sensitivity. Our results highlight the need for improved fire-process parameterizations to enhance the reliability of future projections of fire-related carbon emissions under climate change.