Evaluation of global fire simulations in CMIP6 Earth system models

       Fire is the primary form of terrestrial ecosystem disturbance globally and a critical Earth system process. So far, most Earth system models (ESMs) have incorporated fire modeling, with 19 out of them submitted fire simulations to the CMIP6. Transitioning from CMIP5 to CMIP6, much more models submitted fire simulations and the dominant fire scheme has evolved from GlobFIRM to the Li scheme. However, it remains unknown how well CMIP6 ESMs perform in fire simulations. This study provides the first comprehensive evaluation of CMIP6 fire simulations, through comparisons with multiple satellite-based datasets and the Reading Paleofire Database of global charcoal records (RPD).

        Our results show that most CMIP6 models simulate the global amounts of present-day burned area and fire carbon emissions within the range of satellite-based products, and reproduce observed major features of spatial pattern and seasonal cycle as well as the relationships of fires with precipitation and population density, except for models employing the GlobFIRM fire scheme. Additionally, most CMIP6 models can reproduce the response of interannual variability of tropical fires to ENSO, except for some models incorporating the SPITFIRE fire scheme. From 1850 to 2015, CMIP6 models generally agree with RPD, with some discrepancies in southern South America before 1920 and in temperate and eastern boreal North America, Europe, and boreal Asia after 1990. Compared with CMIP5, CMIP6 has solved the serious issues of CMIP5 which simulates the global burned area less than half of observations, fails to capture the high burned area fraction in Africa, and underestimates seasonal variability. CMIP6 fire carbon emissions simulations are also closer to RPD. However, CMIP6 models still fail to capture the present-day significant decline in observed global burned area and fire carbon emissions partly due to underestimation in anthropogenic fire suppression, and fail to reproduce the spring peak in NH mid-latitudes mainly due to an underestimation of crop fires. Based on our findings, we identify potential biases in fire and carbon projection based on CMIP6 models. We also provide suggestions for the fire scheme development, and bias correction methods when generating multi-source merged fire products.