Text S1 Simulation Experiment for Background Influence
Here, we examine the sensitivity of ACT BL samples to background influences, and evaluate a method to account for background conditions empirically using data collected in the FT. Atmospheric BL signals in North America contain a mixture of background air arriving from distant oceanic and terrestrial sources outside of North America and more local- to regional- sources originating within North America. These latter surface influences (Figure 4) are a significant but incomplete component of observed variability. Background air flowing into the WRF-Chem model domain contains substantial seasonal variability that is synchronized with upstream surface influences, and thus can amplify surface-driven signals. Background air also contains emissions from oceanic sources, which can offset biospheric uptake signals.
We examine the influence of background air on predicted signals by sampling observationally constrained global atmospheric models using BL end points from the 500-particle back trajectories. Here, CO2 and CO fields are determined by running posterior fluxes through GEOS-Chem and saving output every 3 hours at the native horizontal grid (4° x 5° and 2° x 2.5°, respectively). Atmospheric COS fields are determined by an independent 4DVar data assimilation system of the TM5 chemistry transport model (TM5-4DVAR), which infers surface COS flux from NOAA surface observations, and projects optimized fluxes into the atmosphere over the period 2000-2019 (4° x 6°; Ma et al., 2020). The back trajectory 500-particle ensembles contain a mixture of particles that remain within the North America WRF domain over the 10 day period, as well as particles that reach the boundary and exit (Figure S1). Most particles (> 90% on average) exit the domain in winter time, while a larger percentage of particles remain within the domain in the summer (25-50% on average) under a weaker and less advective jet-stream. We then average all particle endpoints together, and repeat for each ACT-America BL flask sample receptor. Seasonal varations for each region and tracer show a consistent pattern of peak concentration in spring, and gradual drawdown through later summer (Figure S2). COS drawdown continues into fall, at which time CO2 becomes enriched and secondary production of CO apparently increases. We find similar seasonal and regional patterns under fair weather condions as are found for cold and warm air masses (top vs bottom row in Figure S2, respectively), with the exception of amplified CO and COS enrichment in summer 2016 in the NE.
We then ask the question: How representative is FT air sampled by ACT-America of these background influences? The main assumption here is that FT air within the continental interior is influenced more by large scale horizontal advection (from boundaries) than by vertical mixing of underlying regional surface exchange. To address this question, we compare seasonal variations of background air from BL particle end points to samples from FT particle start points, using the same atmospheric models. Seasonal variations of FT air (particle receptors) are indicated by markers in Figure S2. The two approaches agree with respect to the timing and magnitude of seasonal drawdown and enrichment, including diverging patterns between CO2/CO and COS in fall. Moreover, we find higher agreement between the approaches under fair weather conditions than under cold and warm air masses, as indicated by a near doubling of RMSE values driven by the Northeast region. .
These results suggest that BL and FT samples are more representative of the same air mass on fair weather days, and consequently, that removing FT values provides a viable approach for estimating observed BL enhancements in the ACT-America data with improved accuracy for fair weather conditions. More precisely, the vertical difference, referred to here as tracer “enhancements” and defined as the difference between BL and FT concentrations, gives a robust measure of observed regional surface flux influences, enabling direct comparison between observed and predicted signals. We note that differences in transport between the sampled fields (from GEOS-Chem and TM5) and particle back trajectories (from WRF-Chem) are unlikely to have a significant influence on these results due to (1) our focus on regional scale conditions, (2) sampling of 500 particle end points, (3) the use of observationally constrained transport fields.
References
Ma, J. et al. Inverse modelling of carbonyl sulfide: implementation, evaluation and implications for the global budget. Atmos Chem Phys Discuss 2020, 1–39 (2020). https://doi.org/10.5194/acp-2020-603