Understanding drivers and biases of simulated CO emissions by the INFERNO fire model over South America
Abstract. Integrating fire simulation into climate models enhances our understanding of ecosystem-fire-climate interactions, clarifying the role of fire in the carbon cycle and other processes. The Interactive Fires and Emissions algorithm for Natural Environments (INFERNO) is one of the new modules in the upgraded UK Earth System Model (UKESM). Here, we use a version of INFERNO coupled only with the Joint UK Land Environment Simulator (JULES) to evaluate its performance and biases over South America (SA); a region that accounts for ∼15 % of global fire carbon emissions. For this, we compared carbon monoxide (CO) estimates from INFERNO (2004–2021) with five satellite-based biomass-burning inventories, conducted sensitivity experiments and developed a machine learning (ML) model targeting biases. INFERNO was able to represent CO emissions in most of the fire-active zone in SA, particularly the southern Amazon ’Arc of Deforestation’, but overestimates emissions (∼100 %) outside them (e.g. within the Amazon forest). The ML model (R2 = 64 %) indicates that tree categories of Plant Functional Types (PFTs) and soil moisture— through its role in flammability and gross primary productivity (GPP) —significantly influence spatiotemporal biases. In northern SA, CO emissions were overestimated by approximately 300 % due to seasonal cycle inaccuracies, while INFERNO showed lower biases in southern SA emissions despite lacking seasonal representation. Both flammability and GPP underpinned the limited simulation of the seasonal cycle. Although INFERNO misrepresented emissions trends in the Arc of Deforestation, it successfully captured the increase in emissions in the eastern Andean Mountains from 2014 to 2021, albeit underestimating their magnitude. Sensitivity experiments revealed that the underlying PFT affected spatiotemporal variability (115 %) and trends (167 %) in CO emissions, while flammability influenced the seasonal cycle (116 %) and trends (158 %). These findings highlight the need for enhanced PFT accuracy and a deeper understanding of the roles of precipitation/soil moisture in GPP and flammability, as well as the consideration of landscape fragmentation to represent land management and forest fire vulnerability.