The signature of internal variability in the terrestrial carbon cycle

Uncertainty in model initial states produces uncertainty in climate simulations because of unforced variability internal to the climate system. Climate scientists use initial-condition ensembles to separate the forced signal of climate change from the unforced internal variability. Our analysis of an 11-member initial-condition ensemble from the Community Earth System Model Version 2 that spans the period 1850-2014 shows that a similar ensemble approach is needed to robustly assess trends in the terrestrial carbon cycle. Uncertainty in model initialization gives rise to internal variability that masks trends in carbon fluxes, and also creates spurious unforced trends, during the period 1960-2014 across North America, meaning that a single model realization can diverge from the observational record or from other models simply because of random behavior. The forced response is, however, evident in the ensemble mean and emerges from the noise of unforced variability at decadal timescales. Our results suggest that trends in the observational record must be interpreted with caution because of multiple possible histories that would have been observed if the sequence of internal variability had unfolded differently. Furthermore, internal variability produces irreducible uncertainty in the carbon cycle, leading to ambiguity in the magnitude and sign of carbon cycle trends, especially at small spatial scales and short timescales. The small spread in initial land carbon pools at 1850 suggests that internal climate variability arising from atmospheric and oceanic initialization, not the biogeochemical initialization, is the predominant cause of carbon cycle variability among ensemble members. Initial-condition ensembles with other Earth system models are needed to develop a multi-model understanding of internal variability in the terrestrial carbon cycle.