Forest ecosystems play an increasingly important role in the global carbon cycle. However, knowledge on carbon exchanges, their spatio-temporal patterns, and the extent of the key controls that affect carbon fluxes is lacking. In this study, we employed 29-site-years of eddy covariance data to observe the state, spatio-temporal variations and climate sensitivity of carbon fluxes (gross primary productivity (GPP), ecosystem respiration (R-eco), and net ecosystem carbon exchange (NEE)) in four representative forest ecosystems in Yunnan. We found that 1) all four forest ecosystems were carbon sinks (the average NEE was -3.40 tC ha(-1) yr(-1)); 2) contrasting seasonality of the NEE among the ecosystems with a carbon sink mainly during the wet season in the Yuanjiang savanna ecosystem (YJ) but during the dry season in the Xishuangbanna tropical rainforest ecosystem (XSBN), besides an equivalent NEE uptake was observed during the wet/dry season in the Ailaoshan subtropical evergreen broad-leaved forest ecosystem(ALS) and Lijiang subalpine coniferous forest ecosystem(LJ); 3) as the GPP increased, the net ecosystem production (NEP) first increased and then decreased when the GPP > 17.5 tC ha(-1) yr(-1); 4) the precipitation determines the carbon sinks in the savanna ecosystem (e.g., YJ), while temperature did so in the tropical forest ecosystem (e.g., XSBN); 5) overall, under the circumstances of warming and decreased precipitation, the carbon sink might decrease in the YJ but maybe increase in the ALS and LJ, while future strength of the sink in the XSBN is somewhat uncertain. However, based on the redundancy analysis, the temperature and precipitation combined together explained 39.7%, 32.2%, 25.3%, and 29.6% of the variations in the NEE in the YJ, XSBN, ALS and LJ, respectively, which indicates that considerable changes in the NEE could not be explained by variations in the temperature and precipitation. Therefore, the effects of other factors (e.g., CO2 concentration, N/P deposition, aerosol and other variables) on the NEE still require extensive research and need to be considered seriously in carbon-cycle-models. (c) 2017 Published by Elsevier B.V.