Anthropogenic climate change is expected to catalyze forest conversion to grass and shrublands due to more extreme fire behavior and hotter and drier post-fire conditions. However, field surveys of wilderness areas in the Northern Rocky Mountains of the United States show robust conifer regeneration on burned sites. This study utilizes a machine learning (GBM) approach to systematically monitor canopy cover on burned areas in two large wilderness areas from 1985 to 2021. The predictive model was developed from coincident LiDAR and Landsat observations and used to create time series of canopy cover on 352 burned sites. Fire impact, as measured by canopy cover loss relative to pre-fire condition, was highly correlated with burn severity. Recovery was characterized by two metrics: whether or not a site exhibited signs of recovery, and the rate at which a site is recovering. Eighty-five percent of the land area studied showed evidence of recovery. Burned areas that are failing to recover are occurring more recently than their recovering counterparts, with 60% of non-recovering sites burning for the first time after 2003. However, the 5-year probability of recovery is similar among recent burns and for those that burned earlier in the record, suggesting that they may recover with more time. Once sites begin recovering, median time to reach pre-fire state is 40 years. Seven sites have projected recovery times greater than two hundred years, six of which burned for the first time after 2006. While fires that are failing to recover or recovering slowly make up proportionally small portions of the landscape, they may be of particular management interest as harbingers of future forest conversion, particularly under hotter and drier future climate scenarios. This work provides a framework for systematic monitoring into the future and establishes a baseline of recovery in the mountains of western Montana and northern Idaho.