Among the many natural disturbances that affect Canada's boreal and temperate forest biomes, wildfire has the greatest impact on forest productivity, landscape structure, timber supply and greenhouse gas emissions. Fire severity represents the fate of biomass, partitioning survival, consumption, and mortality without consumption. Currently, the impact of fire on carbon stocks is limited to a single parametrization of fire severity that assumes only high severity fires, despite evidence of widespread mixed-severity fires. This paper describes a sub-model, termed FireDMs (Fire Disturbance Matrix: severity), of the National Forest Carbon Monitoring Accounting and Reporting System for Canada (NFCMARS). In this sub-model, field measurements of biomass consumption are related to satellite-derived burn severity maps and are interpreted from a fire physics and ecology perspective to derive estimates of the forest greenhouse gas emissions in the immediate aftermath of fires. The sub-model also quantifies fire-killed but uncombusted biomass as a set of distinct pools. Model outputs indicate total direct carbon emissions range from a 11 t C/ha in Boreal Shield West forests of Saskatchewan following low severity fire to over 60 t C/ha in Pacific Maritime forests of British Columbia under high severity fire. The existing approach to emissions in NFC-MARS yields regional CO2e emissions that are 10 to 25 percent higher than this new method, owing to lower overall canopy consumption with mixed-severity fires, which is only partially offset by increased estimates of forest floor consumption in this new approach. Comparisons against directly observed fire plume emissions ratios as well as against annualized carbon emissions for Canada's 2023 fire season show good model agreement with observations.