Terrestrial gross primary production (GPP) is the largest global CO(2) flux driving several ecosystem functions. We provide an observation-based estimate of this flux at 123 +/- 8 petagrams of carbon per year (Pg C year(-1)) using eddy covariance flux data and various diagnostic models. Tropical forests and savannahs account for 60%. GPP over 40% of the vegetated land is associated with precipitation. State-of-the-art process-oriented biosphere models used for climate predictions exhibit a large between-model variation of GPP's latitudinal patterns and show higher spatial correlations between GPP and precipitation, suggesting the existence of missing processes or feedback mechanisms which attenuate the vegetation response to climate. Our estimates of spatially distributed GPP and its covariation with climate can help improve coupled climate-carbon cycle process models.

[1] We upscaled FLUXNET observations of carbon dioxide, water, and energy fluxes to the global scale using the machine learning technique, model tree ensembles (MTE). We trained MTE to predict site-level gross primary productivity (GPP), terrestrial ecosystem respiration (TER), net ecosystem exchange (NEE), latent energy (LE), and sensible heat (H) based on remote sensing indices, climate and meteorological data, and information on land use. We applied the trained MTEs to generate global flux fields at a 0.5°× 0.5°spatial resolution and a monthly temporal resolution from 1982 to 2008. Cross-validation analyses revealed good performance of MTE in predicting among-site flux variability with modeling efficiencies (MEf) between 0.64 and 0.84, except for NEE (MEf = 0.32). Performance was also good for predicting seasonal patterns (MEf between 0.84 and 0.89, except for NEE (0.64)). By comparison, predictions of monthly anomalies were not as strong (MEf between 0.29 and 0.52). Improved accounting of disturbance and lagged environmental effects, along with improved characterization of errors in the training data set, would contribute most to further reducing uncertainties. Our global estimates of LE (158 ± 7 J × 10 18 yr −1 ), H (164 ± 15 J × 10 18 yr −1), and GPP (119 ± 6 Pg C yr ) were similar to independent estimates. Our global TER estimate (96 ± 6 Pg C yr −1 ) was likely underestimated by 5-10%. Hot spot regions of interannual variability in carbon fluxes occurred in semiarid to semihumid regions and were controlled by moisture supply. Overall, GPP was more important to interannual variability in NEE than TER. Our empirically derived fluxes may be used for calibration and evaluation of land surface process models and for exploratory and diagnostic assessments of the biosphere.

Deforestation in mid-to high latitudes is hypothesized to have the potential to cool the Earth's surface by altering biophysical processes [1][2][3] . In climate models of continental-scale land clearing, the cooling is triggered by increases in surface albedo and is reinforced by a land albedo-sea ice feedback 4,5 . This feedback is crucial in the model predictions; without it other biophysical processes may overwhelm the albedo effect to generate warming instead 5 . Ongoing land-use activities, such as land management for climate mitigation, are occurring at local scales (hectares) presumably too small to generate the feedback, and it is not known whether the intrinsic biophysical mechanism on its own can change the surface temperature in a consistent manner 6,7 . Nor has the effect of deforestation on climate been demonstrated over large areas from direct observations. Here we show that surface air temperature is lower in open land than in nearby forested land. The effect is 0.85 6 0.44 K (mean 6 one standard deviation) northwards of 456 N and 0.21 6 0.53 K southwards. Below 356 N there is weak evidence that deforestation leads to warming. Results are based on comparisons of temperature at forested eddy covariance towers in the USA and Canada and, as a proxy for small areas of cleared land, nearby surface weather stations. Night-time temperature changes unrelated to changes in surface albedo are an important contributor to the overall cooling effect. The observed latitudinal dependence is consistent with theoretical expectation of changes in energy loss from convection and radiation across latitudes in both the daytime and night-time phase of the diurnal cycle, the latter of which remains uncertain in climate models 8 .The latitudinal gradient of land-use impact is evident in the comparison of the surface air temperature recorded at FLUXNET (www.fluxnet.ornl.gov) forest towers 9 (Supplementary Table 1 and Supplementary Fig. 1) and surface weather stations in North America (Fig. 1a). Here we use the surface stations as proxies for cleared land. In accordance with the requirement of the World Meteorological Organization, these stations are located in open grassy fields that have biophysical characteristics similar to those of open land, such as being covered by snow in northern latitudes in the winter 10 . Latitude accounts for 31% of the variations in the temperature difference DT between the forest sites and the adjacent open lands (number of site pairs n 5 37). The rate of change in DT with latitude is 20.070 6 0.010 K per degree (mean 6 one standard error, s.e., P , 0.005). At these sites, the annual net all-wave radiation R n