A long-window/short-window data assimilation for separating temporal scales of carbon-exchange focussing on the Amazon

Carbon exchange between the terrestrial biosphere and atmosphere in Amazonia is large, variable, and likely changing due to observed rapid warming and drying across key areas. But quantification of Amazonian net carbon fluxes (i.e., the balance of sources and sinks) is difficult due the limited availability of observational data on atmospheric CO₂, as well as uncertainties in fluxes resulting from land-use changes, such as deforestation and wildfires. Here, we aim to quantify the Amazonian carbon fluxes by leveraging different observational datasets that span various spatio-temporal scales. We analyse the decadal mean and sub-seasonal variability of the net ecosystem exchange (NEE). For this, we apply a novel data-assimilation approach in which NEE across these time scales is separately optimized, to maximize the observational constraints. The system is based on CarbonTracker Europe (CTE) and also uses atmospheric CO₂ mole fractions as top-down constraint. But the new CTE-Long-Window/Short-Window (CTE-LW/SW) also uses the observed long-term rate of change of biomass [1] to inform on decadal mean NEE, OCO2-XCO₂ to separate seasonal changes in NEE over westerly forests and easterly Cerrado landscapes, MODIS NIRv to pick up drought stress, and MOPITT XCO to constrain carbon emissions from biomass burning. We will present results from CTE-LW/SW for the recent decades (2001-2020), focusing on the Amazon droughts, the east-west gradient in carbon sinks, and on the human-driven rise and decline of dry season fire emissions.

[1] https://doi.org/10.1126/sciadv.abe9829