Diagnosing CO2 Transport in the North Atlantic Upper Troposphere: Evaluation of ICON-ART and IFS using IAGOS Observations

Vertical transport processes, such as for example associated with Warm Conveyor Belt airstreams (WCBs) which is defined as a coherent strongly ascending airstream associated with extratropical cyclones, play a critical role in determining the distribution of greenhouse gases within the Upper Troposphere and Lower Stratosphere (UTLS). This study evaluates the performance of two global numerical weather prediction models, ICON-ART (ICOsahedral Nonhydrostatic model with Aerosol and Reactive Trace gases) and IFS (Integrated Forecasting System), in simulating CO₂ mixing ratios during the winter of 2022. Model outputs with different resolutions are compared against in situ measurements from the IAGOS (The In-service Aircraft for a Global Observing System, https://iagos.aeris-data.fr/) infrastructure during transatlantic flights during a period characterised by strong latitudinal CO₂ gradients and vigorous synoptic activity.

The analysis focuses on specific flight campaigns where measured CO₂ mixing ratios exhibited distinct enhancements of 4–6 ppm above background levels in the mid-Atlantic UTLS region. To attribute these anomalies to specific meteorological features, a multi-diagnostic approach is employed. A machine learning algorithm to detect footprints of WCB inflow, ascent and outflow regions (ELIAS 2.0; Quinting et al., 2022) is utilised alongside HYSPLIT Lagrangian backward trajectories initialised from flight coordinates to characterise air mass origin relative to cyclone evolution.

Results reveal persistent model–data discrepancies during January–February 2022, with both ICON-ART and IFS underestimating observed CO₂ spikes by 1–5 ppm. Our analyses show a spatial proximity between WCB activity and elevated CO₂ anomalies suggesting vertical transport of air with distinct chemical signatures from the boundary layer into the upper troposphere through the WCB air stream. Specifically, we find co-located high WCB ascent probabilities (0.4 – 0.8). Moreover, trajectory origins over eastern North America confirm that surface-influenced air masses are lifted via the WCB airstream. We hypothesise that systematic biases in simulated CO₂ distributions stem from model misrepresentation of vertical transport processes and/or uncertainties in emission inventories and natural fluxes, as well as missing chemical production of CO₂ in both modelling frameworks.

These findings highlight the value of combining machine learning-based flow identification with in situ observations to diagnose transport errors in atmospheric models. As WCB activity is projected to intensify under climate change scenarios, improved representation of both synoptic-scale ascent and parametrised turbulent mixing is critical for reducing uncertainties in modelled CO₂ distributions and constraining the global carbon budget.