Recent Pattern of Seasonal and Interannual Variability Inferred From Data-Driven Global Terrestrial Net CO2 Exchange

Abstract:

Terrestrial CO₂ exchange is the key indicator of climate change due to increasing atmospheric CO₂ level. Currently, two principal modelling approaches are used to estimate terrestrial CO₂ exchange: top-down and bottom-up approaches (e.g., CO₂ inversions and ecosystem model simulations, respectively). Since the early 2000s, the research community has invested substantial effort in the development of bottom-up and top-down CO₂ exchange estimations; however, current estimates of terrestrial CO₂ exchange by the two approaches remain inconsistent. To overcome this issue, we adopted more data-driven methods for both top-down and bottom-up approaches: specifically, integration of a global coverage of column averaged CO₂ concentration by the Greenhouse gases Observing SATellite (GOSAT) to CO₂ inversion, and empirical upscaling of eddy flux observations by support vector regression. In this study, we focused on biosphere flux, net CO₂ exchange resulted only from natural vegetation (no fire and fossil fuel emissions), and referred fluxes estimated by the two methods as GOSAT and SVR biosphere fluxes, respectively.

We evaluated seasonal variability of biosphere fluxes estimated by these two methods at 42 terrestrial regions, and found consistent results in the northern mid-high latitudes. Meanwhile, the two biosphere fluxes differed largely in magnitude of seasonal variability in tropical regions; SVR biosphere flux tended toward strong CO₂ sink than GOSAT biosphere flux. Despite the large offset in magnitude, however, seasonal patterns (in the form of anomaly) were similar even in the tropical regions. This result implies that current data-driven methods are limited in producing consistent CO₂ budget, but they can still produce consistent seasonal patterns of CO₂ exchange. Consistency in seasonal pattern between the two methods further suggests that reliable trends of interannual variability may be inferred by the data-driven methods.