Large interannual variations in the measured growth rate of atmospheric carbon dioxide (CO 2 ) originate primarily from fluctuations in carbon uptake by land ecosystems 1-3 . It remains uncertain, however, to what extent temperature and water availability control the carbon balance of land ecosystems across spatial and temporal scales 3-14 . Here we use empirical models based on eddy covariance data 15 and process-based models 16,17 to investigate the effect of changes in temperature and water availability on gross primary productivity (GPP), terrestrial ecosystem respiration (TER) and net ecosystem exchange (NEE) at local and global scales. We find that water availability is the dominant driver of the local interannual variability in GPP and TER. To a lesser extent this is true also for NEE at the local scale, but when integrated globally, temporal NEE variability is mostly driven by temperature fluctuations. We suggest that this apparent paradox can be explained by two compensatory water effects. Temporal waterdriven GPP and TER variations compensate locally, dampening water-driven NEE variability. Spatial water availability anomalies also compensate, leaving a dominant temperature signal in the yearto-year fluctuations of the land carbon sink. These findings help to reconcile seemingly contradictory reports regarding the importance of temperature and water in controlling the interannual variability of the terrestrial carbon balance 3-6,9,11,12,14 . Our study indicates that spatial climate covariation drives the global carbon cycle response.Large interannual variations in recently measured atmospheric CO 2 growth rates originate primarily from fluctuations in carbon uptake by land ecosystems, rather than from the oceans or variations in anthropogenic emissions [1][2][3] . There is a general consensus that the tropical regions contribute the most to terrestrial carbon variability 1,8,18,19 . The observed positive correlation between mean tropical land temperature and CO 2 growth rate 3,5,6,12,13 implies smaller land carbon uptake and enhanced atmospheric CO 2 growth during warmer years, with a sensitivity of about 5 gigatonnes of carbon per year per K. There is a tight relationship between this sensitivity on interannual timescales and long-term changes in terrestrial carbon per degree of warming across multiple climate carbon-cycle models 6 . Despite this strong emergent relationship with mean tropical land temperature, several studies suggest that variations in water availability have an important 8,10,11,14 , even a dominant role 4,9 , in shaping the interannual variability (IAV) of the carbon balance of extensive semi-arid and sub-tropical systems. Furthermore, the recent doubling of the tropical carbon cycle sensitivity to interannual temperature variability has been linked to interactions with changing moisture regimes 13 . A full understanding of the processes governing the climatic controls of terrestrial carbon cycling on interannual timescales and across spatial scales is therefore still lacking. H...

There were systematic differences in performance among the fluxes, with the following ascending order: net ecosystem exchange (R 2 < 0.5), ecosystem respiration (R 2 > 0.6), gross primary production (R 2 > 0.7), latent heat (R 2 > 0.7), sensible heat (R 2 > 0.7), and net radiation (R 2 > 0.8). The ML methods predicted the across-site variability and the mean seasonal cycle of the observed fluxes very well (R 2 > 0.7), while the 8-day deviations from the mean seasonal cycle were not well predicted (R 2 < 0.5). Fluxes were better predicted at forested and temperate climate sites than at sites in extreme climates or less represented by training data (e.g., the tropics). The evaluated large ensemble of ML-based models will be the basis of new global flux products.Published by Copernicus Publications on behalf of the European Geosciences Union.

&lt;p&gt;&lt;strong&gt;Abstract.&lt;/strong&gt; Spatial-temporal fields of land-atmosphere fluxes derived from data-driven models can complement simulations by process-based Land Surface Models. While a number of strategies for empirical models with eddy covariance flux data have been applied, a systematic intercomparison of these methods has been missing so far. In this study, we perform a cross-validation experiment for predicting carbon dioxide (CO&lt;sub&gt;2&lt;/sub&gt;), latent heat, sensible heat and net radiation fluxes, in different ecosystem types with eleven machine learning (ML) methods from four different classes (kernel methods, neural network, tree methods, and regression splines). We employ two complementary setups: (1) eight days average fluxes based on remotely sensed data, and (2) daily mean fluxes based on meteorological data and mean seasonal cycle of remotely sensed variables. The pattern of predictions from different ML and setups were very consistent. There were systematic differences in performance among the fluxes, with the following ascending order: net ecosystem exchange (R&lt;sub&gt;2&lt;/sub&gt; &lt; 0.5), ecosystem respiration (R&lt;sub&gt;2&lt;/sub&gt; &gt; 0.6), gross primary production (R&lt;sub&gt;2&lt;/sub&gt; &gt; 0.7), latent heat (R&lt;sub&gt;2&lt;/sub&gt; &gt; 0.7), sensible heat (R&lt;sub&gt;2&lt;/sub&gt; &gt; 0.7), net radiation (R&lt;sub&gt;2&lt;/sub&gt; &gt; 0.8). ML methods predicted very well the across sites variability and the seasonal cycle (R&lt;sub&gt;2&lt;/sub&gt; &gt; 0.7) of the observed fluxes, while the eight days deviations from the mean seasonal cycle were not well predicted (R&lt;sub&gt;2&lt;/sub&gt; &lt; 0.5). Fluxes were better predicted at forested and temperate climate sites than at ones growing in extreme climates or less representated in training data (e.g. the tropics). The large ensemble of ML based models evaluated will be the basis of new global flux products.&lt;/p&gt;