Fig. 4: Carbon stocks and net fluxes associated with forest age dynamics.

a, AGCs stocks across forest age classes, distinguishing between stand-replaced forests and undisturbed ageing forests, expressed per unit area at a one-degree pixel level. The stand-replaced categories represent the AGC stock of forests at a given age class (young, maturing, mature or old) before stand replacement. b, Net carbon changes for stand-replaced forests across different forest age categories. In a and b, the median values from the 20 biomass realizations are displayed, and the spread represents the spatial variation within a given age class. c, The relationship between the fraction of old forests replaced by young stands (that is, stand-replacement extent) and changes in NEE from inversions between circa 2020 (average of 2019–2021) and 2010 (average of 2009–2011) is shown. The dark solid line represents the linear regression on the ensemble estimates, whereas the dashed grey lines indicate the regressions for the nine individual atmospheric inversion models. One-degree grid cells with less than 20% forest cover have been masked. To smooth the spatial distribution of net CO₂ fluxes, we applied a Gaussian filter (length = 500 km, equivalent to approximately four one-degree pixels). This smoothing technique reduces noise in the data and helps minimize the influence of local transport errors. To minimize noise and regional variability due to atmospheric transport errors, all spatial data were aggregated using area-weighted averaging over 5 × 5 grid cells (that is, a coarsening scale of 5° × 5°). CAMS, Copernicus Atmosphere Monitoring Service; CarboScope, Jena CarboScope Atmospheric Inversion System; CMS-Flux, Carbon Monitoring System Flux; CTE, CarbonTracker Europe; IAPCAS, Institute of Atmospheric Physics Carbon Assimilation System; ICT-NOAA, CarbonTracker NOAA; MIROC, Model for Interdisciplinary Research on Climate; NISMON-CO₂, NICAM-based Inverse Simulation for Monitoring of CO₂; UoE, University of Edinburgh Inversion System.