Reforestation is an important strategy for nature-based climate solutions and identifying carbon storage potential of different locations is critical to its success. Applying average carbon values from forest inventories ignores the spatial heterogeneity in forest carbon and the effects of forest edges on carbon storage degradation. Here we show how spatially-explicit, predictive carbon modeling, that leverages satellite, social and biogeophysical datasets, can be used to identify more efficient restoration opportunities for climate mitigation than area-based carbon stock averages. Accounting for regeneration of forest edges, in addition to reforestation, boosts estimates of potential carbon gains by more than 20%. The total potential carbon gain that could be achieved through reforestation at the level indicated by the Bonn Challenge (350Mha) is 51 Gt CO2-eq, but the "missing carbon" in our current forests accounts for 64.6 Gt CO2-eq globally; the greatest potential carbon gains are found in areas of high fragmentation.