Comparison of the data‐driven top‐down and bottom‐up global terrestrial CO₂ exchanges: GOSAT CO₂ inversion and empirical eddy flux upscaling

Abstract

We examined the consistency between terrestrial biosphere fluxes (terrestrial CO 2 exchanges) from data-driven top-down (GOSAT CO 2 inversion) and bottom-up (empirical eddy flux upscaling based on a support vector regression (SVR) model) approaches over 42 global terrestrial regions from June 2009 to October 2011. Seasonal variations of the biosphere fluxes by the two approaches agreed well in boreal and temperate regions across the Northern Hemisphere. Both fluxes also exhibited strong anomalous signals in response to contrasting anomalous spring temperatures observed in North America and boreal Eurasia. This indicates that the CO 2 concentration data integrated in the GOSAT inversion and the meteorological and vegetation data in the SVR models are equally effective in producing spatiotemporal variations of biosphere flux. Meanwhile, large differences in seasonality were found in subtropical and tropical South America, South Asia, and Africa. The GOSAT inversion showed seasonal variations that pivoted around CO 2 neutral, while the SVR model showed seasonal variations that tended toward CO 2 sink. Thus, a large difference in CO 2 budget was identified between the two approaches in subtropical and tropical regions across the Southern Hemisphere. Examination of the integrated data revealed that the large tropical sink of CO 2 by the SVR model was an artifact due to the underrepresented biosphere fluxes predicted by limited eddy flux data for tropical biomes. Because of the global coverage of CO 2 concentration data, the GOSAT inversion provides better estimates of continental CO 2 flux than the SVR model in the Southern Hemisphere.

Reviews and syntheses: An empirical spatiotemporal description of the global surface–atmosphere carbon fluxes: opportunities and data limitations

“…The comparison suggests that CO 2 fluxes are comparably well constrained in the mid-latitudes where bottom-up and top-down approaches agree. Similar results have been obtained in a comparison of a bottom-up upscaling approach with a more recent inversion based on CO 2 concentration data from the Greenhouse gases Observing SATellite (GOSAT; Kondo et al, 2015). The temporal evolution between both estimates show little agreement except the trend towards more net C uptake by the Earth's surface (Fig.…”

The Effect of GOSAT Observations on Estimates of Net CO₂ Flux in Semi-Arid Regions of the Southern Hemisphere

“…6d). However, this result does not necessarily suggest that seasonal variations are always unaffected by the GOSAT X CO2 retrievals, because, contrary to this study, the inclusion of the GOSAT X CO2 retrievals significantly enhanced anomalies of net CO 2 flux in response to increased temperature in the case of Siberia (Kondo et al 2015). To fully characterize the effect of the GOSAT X CO2 retrievals, it is important to continue validation studies, including evaluation of the spatio-temporal effects on biases inherited in the GOSAT X CO2 on net CO 2 flux estimation.…”

“…In conjunction with previous work, which also analyzed the effect of the GOSAT X CO2 retrievals (e.g., Guerlet et al 2013;Parazoo et al 2013;Saeki et al 2013;Basu et al 2014;Deng et al 2014;Kondo et al 2015), the result of this study facilitates our understanding of the effect of satellite observations on CO 2 flux estimates of the terrestrial carbon cycle. In particular for regions undergoing rapid change (e.g., semi-arid regions), it is important to continue monitoring trends in net CO 2 flux with GOSAT, also using observations from additional satellites dedicated to CO 2 monitoring, such as the Orbiting Carbon Observatory-2 (OCO-2).…”

Fluxes all of the time? A primer on the temporal representativeness of FLUXNET

“…To achieve the goal of providing flux information "everywhere and all of the time" [Baldocchi, 2008], machine-learning techniques (e.g., neural networks, regression trees, and kernel methods) were adopted to extrapolate the location-and timeconstrained measurements to space-and time-explicit products, such as global long-term maps of CO 2 , H 2 O, and energy fluxes [Beer et al, 2010;Jung et al, 2009Jung et al, , 2011Papale and Valentini, 2003;Tramontana et al, 2015Tramontana et al, , 2016Xiao et al, 2008;Yang et al, 2007Yang et al, , 2006. Such gridded and harmonized products provide valuable information of terrestrial fluxes at the desired spatial and temporal scales, which have been used extensively in parameterizing, calibrating, and validating other models such as those from satellite-based remote sensing, land-atmosphere climate models, and global CO 2 flask measurements combined with atmospheric inversions [Anav et al, 2015;Beer et al, 2010;Bonan et al, 2011;Guanter et al, 2014;Jiménez et al, 2011;Kondo et al, 2015]. space-and time-explicit extent, i.e., the so-called representativeness issue [Schimel et al, 2015;Sulkava et al, 2011;Sundareshwar et al, 2007].…”