Drivers of and uncertainty in Amazon carbon sink long-term and interannual variability in CMIP6 models

The Amazon basin rainforest is a critical component of the climate system, currently representing 25 % of terrestrial carbon gains and storing 150x109 to 200x109 t of carbon. Whether the Amazon rainforest will remain a net carbon sink is an open scientific question: while its future stability and functioning may be compromised by climate change and anthropogenic pressures, Earth system model (ESM) divergence in the projections undermines the models' reliability in simulating its future evolution. In this study, we examined the contribution of different climatic drivers behind the long-term and interannual variability evolution of the carbon sink within the Amazon basin using 11 CMIP6 ESMs, shedding light on the main factors contributing to inter-model diversity. By adopting the carbon-cycle feedback framework with C4MIP experiments, our results underscore the dominant role of CO2 fertilization in driving the long-term Amazon carbon sink trend and uncertainty. We also address the variability in carbon fluxes at the interannual timescale using a multivariate predictive model with historical and SSP5-8.5 ScenarioMIP simulations. In this respect, we emphasize the contribution of gross primary productivity (GPP) modulation by shortwave incoming radiation, which dominates net biome productivity (NBP) divergence across the ESM ensemble. Additionally, we demonstrate that temperature-driven anomalies will be the main mechanism responsible for the higher Amazon carbon sink sensitivity to the El Ni & ntilde;o-Southern Oscillation (ENSO) under sustained global warming, predominantly as a result of the amplification of NBP sensitivity to temperature anomalies. Being that the representation of terrestrial carbon-cycle processes is still one of the main uncertainties undermining ESM projections, we therefore advocate for increased focus from modelling groups on a more accurate and consistent representation of land processes and parameterizations, which will hopefully lead to reduced uncertainties in simulations from the next generation of ESMs.