Abstract. Ecological models are effective tools for simulating the distribution of global carbon sources and sinks. However, these models often suffer from substantial biases due to inaccurate simulations of complex ecological processes. We introduce a set of scaling factors (parameters) to an ecological model on the basis of plant functional type (PFT) and latitudes. A global carbon assimilation system (GCAS-DOM) is developed by employing a dual optimization method (DOM) to invert the time-dependent ecological model parameter state and the net carbon flux state simultaneously. We use GCAS-DOM to estimate the global distribution of the CO₂ flux on 1° × 1° grid cells for the period from 2001 to 2007. Results show that land and ocean absorb −3.63 ± 0.50 and −1.82 ± 0.16 Pg C yr⁻¹, respectively. North America, Europe and China contribute −0.98 ± 0.15, −0.42 ± 0.08 and −0.20 ± 0.29 Pg C yr⁻¹, respectively. The uncertainties in the flux after optimization by GCAS-DOM have been remarkably reduced by more than 60%. Through parameter optimization, GCAS-DOM can provide improved estimates of the carbon flux for each PFT. Coniferous forest (−0.97 ± 0.27 Pg C yr⁻¹) is the largest contributor to the global carbon sink. Fluxes of once-dominant deciduous forest generated by the Boreal Ecosystems Productivity Simulator (BEPS) are reduced to −0.78 ± 0.23 Pg C yr⁻¹, the third largest carbon sink.