Constraining the time of emergence of anthropogenic signal in the global land carbon sink

Abstract. The global land carbon sink has increased since the preindustrial period, driven by increasing atmospheric CO₂ concentration and climate change. However, detecting these anthropogenic signals in the global land carbon sink is challenging due to the large year-to-year variability, which can mask or amplify long-term trends, particularly on regional and decadal scales. This study aims to detect the time it takes for long-term trends driven mostly by anthropogenic signal to dominate over natural variations, that is, the "time of emergence", in the land carbon sink.

For this, we use five large ensembles of historical simulations (1851–2014) and future scenarios (2016–2100) by Earth system models. Our results show that, firstly, the anthropogenic signal in the global net land carbon sink emerges from 26 to 66 years in the period 1960–2019 (relative to the natural variations in the period of 1930–1959), depending on the ESM considered. The time of emergence is considerably shorter for the two major gross carbon fluxes: 8–13 years for gross primary productivity and 6–10 years for total ecosystem respiration. Furthermore, we find that long-term trends of net land carbon sink on most regional scales take at least 20 years more to emerge, due to larger contributions from internal climate variability at smaller scales.

Secondly, future scenarios show delayed signal detection compared to historical trends, due to a slow-down of the increasing net land carbon sink in response to emission mitigation, compared to the higher natural variability.

Thirdly, we apply dynamical adjustment to filter out the year-to-year circulation-induced variability in both the historical and future simulations. This approach allows to substantially shorten the detection time for the global net land carbon sink: between 34–39 % for the historical period and 27–54 % for the future simulations. This approach can, in principle, be applied to observational based datasets, thereby improving our ability to detect long-term trends on land carbon sink variability. Given that long-term trends are mostly associated with human impacts on the land carbon cycle, our proposed approach can offer valuable insights on the effectiveness of policy decisions and their implementation.