Article Open access Clara Deser , William R. Wieder , Danica L. Lombardozzi & Flavio Lehner npj Climate and Atmospheric Science 7 , Article number: 320 (2024) Cite this article Metrics Abstract Internal climate variability (ICV) creates a range of climate trajectories, which are superimposed upon the forced response. A single climate model realization may not represent forced change alone and may diverge from other realizations, as well as observations, due to ICV. We use an initial-condition large ensemble of simulations with the Community Earth System Model (CESM2) to show that ICV produces a range of outcomes in the terrestrial carbon cycle. Trends in gross primary production (GPP) from 1991 to 2020 differ among ensemble members due to the different climate trajectories resulting from ICV. We quantify how ICV imparts on GPP trends and apply our methodology to the observational record. Observed changes in GPP at two long-running eddy covariance flux towers are consistent with ICV, challenging the understanding of forced changes in the carbon cycle at these locations. A probabilistic framework that accounts for ICV is needed to interpret carbon cycle trends. Introduction Climate change is evident in numerous disparate observations of the Earth system. The increase in atmospheric CO 2 measured at Mauna Loa, Hawaii, since 1959, is one of the iconic records of global change 1 , 2 , as is the planetary warming seen in the time series of surface temperature measurements 3 , 4 , 5 and reflected in Arctic sea–ice loss 6 . Multidecadal changes in the biosphere, both greening and browning, are found in satellite-derived vegetation indices 7 , 8 , 9 . Further evidence for a changing biosphere is obtained from the worldwide network of eddy covariance flux towers, which measure energy, water, and CO 2 exchange between the biosphere and […]