The world is experiencing remarkable climate change, which alters vegetation structural and functional status, and the biosphere feedback to the climate might amplify or dampen regional and global climate change. The exchange of energy and mass across the land-atmosphere interface is an essential indicator of the interaction between ecosystem and environment, and it has been widely measured at landscape scale since the establishment of FLUXNET, a global network of micrometeorological flux measurement sites (Baldocchi, 2008; Baldocchi et al., 2001). Furthermore, these site observations have been upscaled to global scale in conjunction with satellite remote sensing data, providing opportunities to investigate the spatial and temporal variations of carbon and water cycles from a macro perspective (Beer et al., 2010; Martin Jung et al., 2010). However, such upscaling approaches are based on data-driven models, which need to be well calibrated using site observations (M. Jung, Reichstein, & Bondeau, 2009; Papale & Valentini, 2003; Xiao et al., 2010; Yang et al., 2007). Consequently, those models are limited by the representativeness, quantity and quality of the training datasets (Sundareshwar et al., 2006), as well as the lack of independent reference datasets for validation. Furthermore, by using the upscaling approach predictors are directly linked with target fluxes through machine learning, while the intermediate variables along with intrinsic mechanisms are hidden.