Canada’s extreme wildfires dominate the decline in global land carbon sinks in 2023

Terrestrial land carbon sinks are strongly influenced by climate extremes, and 2023 is the warmest year on record, accompanied by an El Niño event, extreme wildfires, and extreme precipitation and drought, but their impact on global land sinks in 2023 remains unclear. Here, we used the Global Carbon Assimilation System, version 2, to estimate recent global land sinks by assimilating the OCO-2 ACOS v11.1 XCO2 retrievals. We estimate the global land sink to be -1.63 ± 0.52 PgC/yr in 2023. Compared to 2017-2022, it decreases by 0.59 PgC/yr, in which net ecosystem exchange decreases by only 0.14 PgC/yr, but wildfire emissions increase significantly by 0.45 PgC/yr, mainly in Canada. Our findings suggest that extreme wildfires are an important threat to land sinks under global warming.

Methods
We generate this dataset using the Global Carbon Assimilation System, version 2 (GCASv2) by assimilating OCO-2 ACOS v11.1 XCO2 retrievals. GCAS v2 was developed to estimate gridded surface carbon fluxes by assimilating primarily satellite XCO2 retrievals, which consists of the Model for Ozone and Related Chemical Tracers, version 4 (MOZART-4) and the ensemble square root filter (EnSRF). It has good performance in assimilating GOSAT and OCO-2 XCO2 retrievals, revealing the response of terrestrial ecosystems to extreme climate events, and has participated in the inversion of the Global Carbon Budget 2023.

For this dataset, GCASv2 was run from 6 September 2014 to 6 January 2024 with a DA window of 1 week. Only the terrestrial net ecosystem exchange (NEE) and the atmosphere-ocean carbon exchange (OCN) were optimized, and the carbon emissions from fossil fuels and cement (FFC) and from wildfires were prescribed. Usually, the choice of a priori NEE has an important impact on the inversion of terrestrial carbon sinks, thus we used the simulations of two terrestrial ecosystem models (i.e., BEPS and CASA) as a priori NEE. In addition, there are large discrepancies among different wildfire CO2 emission datasets, indicating that the estimation of wildfire emissions is also subject to considerable uncertainty. To reduce the impact of such uncertainty on our inversion, two wildfire emission datasets (i.e., GFED 4.1s and GFAS) were adopted. Therefore, we performed four inversion experiments (Exp_BEPS&GFED, Exp_BEPS&GFAS, Exp_CASA&GFED, and Exp_CASA&GFAS) with different prior NEEs and wildfire emissions to obtain a more robust estimate of the NBE (=NEE + wildfires).