Although deforestation in the Amazon has decreased during the last decade, processes of forest degradation have continued. Forest degradation from selective logging, fragmentation, and understory fires contributes significantly to the regional changes in forest carbon stocks and carbon cycling. However, the impact of forest degradation processes on carbon cycling remains very uncertain, and observation-based studies are often too short to characterize the long-term dynamics of degraded forests. Here, we combined airborne LiDAR with geo-referenced forest inventories to obtain calibrated estimates of biomass at 15 large study sites across the Brazilian Amazon with various levels of degradation. The comparison between LiDAR-based estimates of forest biomass with existing biomass maps exhibited large variability and strong biases for the most degraded areas, suggesting that forests near intensive human activities could be losing carbon due to recent additional degradation. We also used the Ecosystem Demography model to estimate the recovery time for three areas in the Eastern Amazon that experience similar climate but suffered different levels of degradation. Model simulations show that while degraded forests may be able to recover most of the biomass within 100 years if no further disturbances occurred, changes in the micro-environment due to tree loss could increase forest flammability and delay forest recovery. These results highlight the need to account for fine-scale, anthropogenic drivers of heterogeneity in forest structure and carbon stocks at a regional-level, and to develop model frameworks that represent forest degradation processes and the carbon cycle consequences of interactions between the forest micro-environment and global climate.