Does vapor pressure deficit drive the seasonality of δ¹³C of the net land-atmosphere CO₂ exchange across the United States?

The seasonal pattern of the carbon isotope content (δ¹³C) of atmospheric CO₂ depends on local and nonlocal land-atmosphere exchange and atmospheric transport. Previous studies suggested that the δ¹³C of the net land-atmosphere CO₂ flux (δ_source) varies seasonally as stomatal conductance of plants responds to vapor pressure deficit of air (VPD). We studied the variation of δ_source at seven sites across the United States representing forests, grasslands, and an urban center. Using a two-part mixing model, we calculated the seasonal δ_source for each site after removing background influence and, when possible, removing δ¹³C variation of nonlocal sources. Compared to previous analyses, we found a reduced seasonal (March–September) variation in δ_source at the forest sites (0.5‰ variation). We did not find a consistent seasonal relationship between VPD and δ_source across forest (or other) sites, providing evidence that stomatal response to VPD was not the cause of the global, coherent seasonal pattern in δ_source. In contrast to the forest sites, grassland and urban sites had a larger seasonal variation in δ_source (5‰) dominated by seasonal transitions in C₃/C₄ grass productivity and in fossil fuel emissions, respectively. Our findings were sensitive to the location used to account for atmospheric background variation within the mixing model method that determined δ_source. Special consideration should be given to background location depending on whether the intent is to understand site level dynamics or regional scale impacts of land-atmosphere exchange. The seasonal amplitude in δ¹³C of land-atmosphere CO₂ exchange (δ_source) varied across land cover types and was not driven by seasonal changes in vapor pressure deficit. The largest seasonal amplitudes of δ_source were at grassland and urban sites, driven by changes in C₃/C₄ grass productivity and fossil fuel emissions, respectively. Mixing model approaches may incorrectly calculate δ_source when background atmospheric observations are remote and/or prone to anthropogenic influence.