Large fire years in which >1% of the landscape burns are becoming more frequent in the Alaskan (USA) interior, with four large fire years in the past 10 years, and 79 000 km² (17% of the region) burned since 2000. We modeled fire severity conditions for the entire area burned in large fires during a large fire year (2004) to determine the factors that are most important in estimating severity and to identify areas affected by deep‐burning fires. In addition to standard methods of assessing severity using spectral information, we incorporated information regarding topography, spatial pattern of burning, and instantaneous characteristics such as fire weather and fire radiative power. Ensemble techniques using regression trees as a base learner were able to determine fire severity successfully using spectral data in concert with other relevant geospatial data. This method was successful in estimating average conditions, but it underestimated the range of severity.

This new approach was used to identify black spruce stands that experienced intermediate‐ to high‐severity fires in 2004 and are therefore susceptible to a shift in regrowth toward deciduous dominance or mixed dominance. Based on the output of the severity model, we estimate that 39% (∼4000 km²) of all burned black spruce stands in 2004 had <10 cm of residual organic layer and may be susceptible a postfire shift in plant functional type dominance, as well as permafrost loss. If the fraction of area susceptible to deciduous regeneration is constant for large fire years, the effect of such years in the most recent decade has been to reduce black spruce stands by 4.2% and to increase areas dominated or co‐dominated by deciduous forest stands by 20%. Such disturbance‐driven modifications have the potential to affect the carbon cycle and climate system at regional to global scales.