The COVID-19 pandemic became an unintended global experiment: fossil fuel emissions dropped sharply, but how this affected atmospheric CO₂ remained unclear. Related papers in the archive explore biosignatures, technosignatures, and exoplanet atmospheres, as well as climate impacts on the biosphere, but this study stands apart: it turns to Earth and tests whether a short-term anthropogenic perturbation can be detected in real time. The question matters for monitoring systems and for verifying the emission reductions that countries pledged under the Paris Agreement.
The authors built a chain of three components. First, they created a daily fossil fuel emission dataset at high spatial resolution by combining country-level estimates from Liu et al. (2020) with ODIAC spatial distributions and TIMES diurnal cycles. Then, using the VEGAS terrestrial biosphere model, they calculated hourly carbon fluxes between land and atmosphere, while oceanic exchange came from Takahashi climatology adjusted to the global carbon budget. These fluxes were fed into the GEOS-Chem atmospheric transport model at 4°×5° resolution with 47 levels, running from January 2008 to May 2020. To isolate sub-annual anomalies, they applied the DCA method (Detrending, Climatology, Anomaly): subtracting a 12-month moving average, then the mean seasonal cycle, to obtain deviations from climatology. Additionally, they ran three sensitivity experiments: BWC (biosphere+weather+COVID), BW (no COVID effect), and B (biosphere only), to separate the contributions of each factor.
The model showed that from February to April 2020, the column CO₂ anomaly in the 50°S–50°N band was −0.13 ppm relative to the 2010–2019 climatology. The GOSAT satellite gave a similar picture: the March–April drop was the largest sub-annual deviation in the past 10 years. Sensitivity experiments attributed the contributions as follows: COVID-19, 54%; biosphere, 23%; weather, 23%. In May, the anomaly strengthened to −0.36 ppm, mainly due to COVID reductions and the biosphere shifting from a source to a carbon sink. At background stations such as Cape Kumukahi in Hawaii, anomalies of 1–2 ppm were observed, but there the biosphere and weather dominated, with the COVID-19 signal only about 0.2 ppm. In cities, the picture varied: in Beijing, the expected reduction of 5–10 ppm was hidden by weather fluctuations, while in Chengdu, a clear stepwise drop of 10–20 ppm occurred on the day of lockdown. Roadside measurements in Beijing showed a 30 ppm decrease in curb-side CO₂ during the quarantine compared with pre-COVID levels.
This work demonstrates that modern carbon monitoring systems can capture short-term anthropogenic signals even against natural variability. Unlike neighboring studies focused on other planets or long-term trends, this is the first test of detecting a sub-annual anomaly caused by a sharp emission reduction using a combination of model and satellite data. It strengthens confidence in attribution methods and paves the way for operational monitoring of climate commitments.
The topic could evolve toward higher spatial resolution in models and denser observational coverage, especially in urban areas. Improved mesoscale modeling and expanded networks of low-cost sensors would allow more confident detection of local emission reduction signals. Also promising is the use of data from new satellites such as OCO-3 and future missions for more detailed mapping of urban CO₂ plumes.
The results will affect greenhouse gas monitoring and verification systems, as well as the development of climate regulation policies at city and country levels.
Next steps include refining city-scale models, integrating more ground-based and satellite observations, and developing methods for real-time operational attribution of CO₂ anomalies.
This work directly addresses the unresolved problem of separating anthropogenic and natural carbon fluxes on short timescales. Existing monitoring systems track annual and interannual changes well, but sub-annual anomalies caused by rapid emission changes remain at the edge of detectability. This study shows that combining models and observations can overcome that barrier, but attribution accuracy is still limited by weather noise and uncertainty in biospheric fluxes.
During the lockdown in Beijing, roadside CO₂ concentrations dropped by 30 ppm, equivalent to removing about 60% of cars from the roads. In Chengdu, the drop was so sharp that it looked like a step on the graph, coinciding with the day of Chinese New Year.