Abstract/Summary

Data-driven techniques that scale up eddy covariance (EC) fluxes from tower footprints with satellite observations and machine learning algorithms significantly advance our understanding of global carbon, water, and energy cycles. However, few upscaling approaches take a consistent approach to upscaling both carbon and energy fluxes. A lack of uniformity in the upscaling approach could lead to inconsistencies in global interannual variability of fluxes and between types of carbon and energy fluxes. Hence, this study aims to identify obstacles in flux upscaling and propose a uniform upscaling framework UFLUX for gross primary productivity (GPP), ecosystem respiration (Reco), net ecosystem exchange (NEE), sensible heat (H), and latent energy (LE). The key findings are as follows: 1) The upscaling performance exhibits a limited improvement from the use of more advanced machine learning approaches (e.g. <0.3 in R2 improvements while using deep neural networks). 2) The spatial density of EC towers is the primary factor determining the effectiveness of upscaling, explaining >50% of the upscaling uncertainty. 3) The UFLUX framework considered the interconnection between fluxes and achieved a competitive validation precision (daily R2 = 0.7 on average of five flux types) when compared with products that upscaled a subset of the fluxes. UFLUX effectively preserved the ecosystem light-use efficiency (0.83 of linear regression slope and the same after), Bowen ratio (0.8), and particularly, the water-use efficiency (0.81), when compared to the only other product (i.e. FLUXCOM) to upscale both carbon and water.