Decadal changes in carbon Emissions and sinks optimized using a Bayesian fusion of multiple observations.

Abstract:

A better understanding of current human perturbation global carbon cycle is fundamental to support the projection of future climate change and designing better informed climate policy. making. Here, we develop a Bayesian approach that fuses different data-streams to optimize fossil fuel emission (F), land use change emission (L), land sink (B) and ocean sink (O) at a 5-year time step from 1980 to 2014. Here we show that the optimization decreases uncertainties compared with those of the carbon budget established by Global Carbon Project (GCP) and the IPCC where single datasets are used and the land sink deduced as a residual from other terms. Uncertainties of F, L, B and O are reduced by 11%, 26, 36% and 40% compared to those of the GCP budget. Fossil fuel and land use change emissions show a significantly increasing and decreasing trend, respectively, while no significant trend is detected for land and ocean uptakes. We infer with a 93% confidence that the net biosphere uptake with land use change (BL) has increased since 1980. After a further decomposition of the net land flux, sum of L and B, into gross terrestrial fluxes of primary productivity (GPP) and ecosystem respiration (TER), we attribute the increase of B+L to a faster growth of GPP than TER from 1980~1984 to 1990~1994 and a lower decreasing rate of GPP than TER in the 2000s. This approach can be potentially implemented on a yearly basis in future with growing observation-based datasets.