A global, observation-based assessment of whole-ecosystem carbon turnover times shows that the overall mean global carbon turnover time is about 23 years and that locally its spatial variability depends on precipitation at least as strongly as on temperature. Feedback between the terrestrial carbon cycle and climate is partly determined by changes in the residence time of carbon in land ecosystems. Using observation-based gross primary production estimates, remote-sensing based estimates of vegetation biomass and new estimates of total soil organic carbon to full depth, Nuno Carvalhais et al. calculate a spatially explicit estimate of global mean residence times of carbon in land ecosystems. They arrive at an overall mean global carbon turnover time of about 23 years with carbon residing, on average, in the vegetation and soil near the Equator for a shorter time period than at northern latitudes. The paper reports expected dependencies of carbon turnover time with temperature, but also identifies a strong association with precipitation, implying that future carbon cycle climate feedbacks may depend more strongly on changes in the water cycle than currently expected based on Earth system model studies. The response of the terrestrial carbon cycle to climate change is among the largest uncertainties affecting future climate change projections1,2. The feedback between the terrestrial carbon cycle and climate is partly determined by changes in the turnover time of carbon in land ecosystems, which in turn is an ecosystem property that emerges from the interplay between climate, soil and vegetation type3,4,5,6. Here we present a global, spatially explicit and observation-based assessment of whole-ecosystem carbon turnover times that combines new estimates of vegetation and soil organic carbon stocks and fluxes. We find that the overall mean global carbon turnover time is years (95 per cent confidence interval). On average, carbon resides in the vegetation and soil near the Equator for a shorter time than at latitudes north of 75° north (mean turnover times of 15 and 255 years, respectively). We identify a clear dependence of the turnover time on temperature, as expected from our present understanding of temperature controls on ecosystem dynamics. Surprisingly, our analysis also reveals a similarly strong association between turnover time and precipitation. Moreover, we find that the ecosystem carbon turnover times simulated by state-of-the-art coupled climate/carbon-cycle models vary widely and that numerical simulations, on average, tend to underestimate the global carbon turnover time by 36 per cent. The models show stronger spatial relationships with temperature than do observation-based estimates, but generally do not reproduce the strong relationships with precipitation and predict faster carbon turnover in many semi-arid regions. Our findings suggest that future climate/carbon-cycle feedbacks may depend more strongly on changes in the hydrological cycle than is expected at present and is considered in Earth system models.