High resolution modeling of vegetation reveals large summertime biogenic CO₂ fluxes in New York City

We find that vegetation in the developed land contributes significantly to the region's biogenic CO₂ fluxes (i.e. GEE, Res, and net ecosystem exchange NEE = GEE + Res) during the summer (figure 3(a)). Specifically, compared to assumptions of no vegetation in the developed land covers, the regional total GEE increases by three (from 0.1 to 0.27 TgC month⁻¹) and four fold (from 0.1 to 0.38 TgC month⁻¹) when vegetation in the developed land is treated as forest or grassland, respectively, and total Res (heterotrophic + autotrophic) increased by factors of 6 (from 0.02 to 0.12 TgC month⁻¹) and 2 (from 0.02 to 0.04 TgC month⁻¹), respectively. Altogether, these enhanced biogenic CO₂ fluxes result in a 2–4 times (from 0.08 to 0.17 or 0.34 TgC month⁻¹) increase in regional NEE when taking into account the vegetation in the developed land (figure 3(a)). These findings indicate that assuming zero biogenic CO₂ fluxes from vegetated developed land, which is common in larger-scale weather and climate models, can lead to a significant underestimation of regional biogenic CO₂ fluxes in urban environments (figure S2 in the supplementary information).

Zoom In Zoom Out Reset image size

Vegetation in the developed land covers of low-, medium-, and high-intensity contributes differently to the regional NEE across the domain. Medium-intensity development accounts for a higher portion of land area in the study domain relative to low-intensity (30% versus 12%, figure 3(c)) and makes a larger contribution to EVI compared to high-intensity (0.28 versus 0.14, figure 3(d)), thus contributing most to total NEE in the domain. In general, given the predominance of the developed land in the study domain ($\gt$90%), the vegetation within the developed land cover categories (e.g. lawns and trees in parks, yards, and along streets) dominates the domain's total NEE. The conventional ecosystem land cover categories of 'forest' and 'grassland', that only account for $\lt$10% of the domain area collectively, contribute roughly 15% to the regional NEE in our study domain (figure 3(b)).

Because of large differences in CO₂ fluxes between grassland and deciduous broadleaf forest ecosystems, we find that the simulated regional biogenic carbon fluxes are quite sensitive to how the vegetation within developed land covers is categorized (figure 3(a)). Specifically, the regional total GEE, Res, and NEE vary by a factor of 1.4, 0.4, and 2.2, respectively, between the two scenarios (figure 3(a)). The 15 cm spatial resolution NYC land cover map suggests that the tree and grass coverage are roughly the same in the low-intensity developed land and the tree coverage in the medium- and high-intensity developed land (25% and 11%) is higher than grass (16% and 5%). Overall, the relative contribution of trees to total vegetation cover, relative to grass, increases as development density increases. Note that the 15 cm land cover map only covers the political boundaries of NYC which comprises 65% of our study domain. The developed land cover categories outside of NYC in our study domain are dominated by open space and low-intensity developed land cover categories (figure 1(a)) and therefore likely contain more urban grassland (i.e. lawn) than developed land within NYC itself. As such, we expect our DEVasDBF and DEVasGRS scenarios (figure 3) to be bookend estimates of the actual biogenic CO₂ fluxes across our study domain, with the actual fluxes being somewhere between the estimates derived from these two scenarios.

Lawns (i.e. urban grasslands)—which are typically composed of both grasses and herbs—are oftentimes the most common herbaceous plant-dominated ecosystem type in cities and developed landscapes (Ignatieva et al 2015). However, there are currently no model parameters specifically for 'urban grasslands.' Here, we use 'grassland' model parameters that derived from the Vaira Ranch Grassland in California which is dominated by C3 annual grasses and herbs. The plants in this ecosystem with a Mediterranean climate are functional during the winter and early spring and dead during the summer (Xu and Baldocchi 2004). There are obvious differences in climate, environmental conditions, and management between a rural grassland system in California and an urban grassland in NYC. First, the ecosystem respiration of Vaira grasses is mainly driven by precipitation and shows higher respiration rates during wet and cooler season (Xu and Baldocchi 2004), suggesting a weakened relationship with Tas. However, urban lawns under management such as irrigation likely show a stronger temperature-respiration relationship, compared to the unmanaged Vaira Ranch grasses. This could lead to an underestimation of lawn ecosystem respiration by VPRM. Second, Vaira grassland has similar drivers as urban grasslands for gross primary production (GPP)—Ta, light intensity, and leaf area index (Xu and Baldocchi 2004, Miller et al 2018). Miller et al (2018) report a GPP of $6.0$ ± 1.0 gC m⁻² d⁻¹ for turf grasses and Xu and Baldocchi (2004) 4.5–5.0 gC m⁻² d⁻¹ for Vaira grasses, suggesting that the parameterization of Vaira grasses based on Xu and Baldocchi (2004) might provide a reasonable estimation of GPP for lawns. That said, the obvious differences in climate and grass community composition between Vaira and lawns in mesic temperate cities highlight the need for development of model parameterizations for urban lawns. Miller et al (2018) report an empirical light use efficiency for low-maintenance lawns, but more than 50% of homeowners fertilize their lawns (Polsky et al 2014). Parameterization for GPP and ecosystem respiration for lawns of different levels of management is still lacking. Our results highlight the need to develop a more comprehensive set of empirical parameters for urban grasslands using both site-specific and regional gridded input data sets.

There is considerable spatial variability of NEE across the domain (figure 4). Throughout June, NEE values at some locations approach −30 µmol m⁻² s⁻¹ (figure 3(c)), which is comparable to the Eddy-covariance observations in Harvard forest (Urbanski et al 2007). The NEE from developed land when all vegetation is assumed to be trees, ranges from −10 to 0 µmol m⁻² s⁻¹. In contrast, when the vegetation in developed land is assumed to be grassland, the NEE exhibits a wider range from −25 to 0 µmol m⁻² s⁻¹ with a higher frequency of more negative values. This large spatial heterogeneity of the biogenic carbon fluxes across the domain is often not captured by large-scale models with coarser spatial resolutions (Hu et al 2021, Gourdji et al 2022). Measurements of biogenic CO₂ fluxes from different urban ecosystems (e.g. lawns, urban savannas, fragmented forests, etc) and across seasons are needed to evaluate model results and improve performance of carbon flux models in urban landscapes. The domain-averaged GEE and NEE show typical diurnal variations with maxima around noon in all three scenarios, as well as the differences between cases (figure 3(d)).

Zoom In Zoom Out Reset image size

The simulated atmospheric CO₂ enhancements agree well with the observed CO₂ enhancement, especially during the daytime, when the boundary layer is well mixed (figure 2(b)). The key factors that impact the agreement include (a) correctly capturing the boundary layer dynamics in the Lagrangian atmospheric transport model STILT, (b) accurately defining the upwind background CO₂ concentrations (see details in section 2.4), and (c) accurate representation of the biogenic and anthropogenic surface CO₂ fluxes. We observe large day-to-day variability in the simulated CO₂ enhancements, likely driven by boundary layer dynamics such as wind and mixing conditions (figure 2) and more variability at night when the low nocturnal boundary layer magnifies the enhancement from biogenic respiration and anthropogenic CO₂ emissions. The first two factors above have been extensively studied and tested (e.g. Sargent et al 2018) allowing us to assess the impact of the biogenic and anthropogenic fluxes. The overall good agreement in this study highlights that a detailed representation of urban biological fluxes and knowledge of the spatial and temporal distribution of emissions are essential for detecting variability in atmospheric CO₂ concentrations.

The atmospheric CO₂ concentration enhancements are, on average, 30 ppmv in the morning and 12 ppmv in the afternoon (figure 2(b)), while nighttime enhancements are larger due to the shallow nighttime stable boundary layer. This diurnal variation in atmospheric CO₂ enhancements can also be seen in figure 2(a). The large variability in morning and nighttime CO₂ enhancements manifests challenges in both measurements and modeling under stable boundary layer conditions. In the following discussion, we focus on the afternoon period when the boundary layer is well-mixed.

During summer afternoons, vegetation uptake of CO₂ across our study domain is large enough to completely offset the respective CO₂ concentration enhancements produced by many of the individual (i.e. not the sum) anthropogenic sectors such as energy for buildings, the power industry, combustion for manufacturing, or on-road traffic (figure 2(a)). Depending on the vegetation type assumed for the developed land (i.e. the DEVasDBF or DEVasGRS scenario), NEE offsets 20%–40% of CO₂ emissions attributed to total anthropogenic emissions during summer afternoons. While these offsets are for the most photosynthetically active time of year, and thus an upper limit of carbon offset possible from urban vegetation, we find it surprising that even in a city with the development density and CO₂ emissions as high as NYC, vegetated systems still make a large contribution to urban CO₂ fluxes. This large biogenic flux greatly confounds attempts to quantify city-wide carbon budgets, resulting in much greater uncertainty.

NYC has the third largest anthropogenic CO₂ emissions globally (Moran et al 2018) and has a relatively low tree canopy coverage of only 22% (Treglia et al 2021). The Boston urban region has a similar climate and a tree canopy of 25% (Raciti et al 2014) as NYC, but lower citywide anthropogenic CO₂ emissions of 0.92 kgC m⁻² yr⁻¹ (Sargent et al 2018) compared to 9.4 kgC m⁻² yr⁻¹ in our study domain (based on EDGAR). Sargent et al (2018) found that biogenic uptake almost completely balances anthropogenic CO₂ emissions on summer afternoons in Boston. Given the similar tree canopy cover and ten times higher anthropogenic emissions in our study domain, one might expect the vegetation to offset roughly 10% of emissions, all else being equal. However, our results suggest a vegetation offset between 20% and 40% on summer afternoons (e.g. figure 2). The importance of biogenic CO₂ fluxes in mediating urban enhancements of CO₂ even in a city of low tree canopy cover as NYC underscores the large impacts of biology on carbon cycling for cities of all sizes, particularly as many move to increase vegetation cover. The biogenic CO₂ fluxes should be accurately evaluated as carbon mitigation policies are introduced.

We find that CO₂ enhancements by on-road traffic could be entirely offset by vegetation uptake in a summer afternoon, which indicates that biogenic fluxes become more important when using top-down approaches for quantifying emissions from individual economic sectors. The transportation sector generates the largest share (27%) of greenhouse gas emissions nationwide, followed by electricity (25%), industry (24%), commercial & residential (13%), and agriculture (11%) (EPA 2022). Although the on-road traffic is overwhelmed by power industry and energy for buildings due to its geographic proximity to a number of power plants and old building infrastructure respectively, NYC is still the most congested city in the US in 2021 as traffic levels are beginning to increase again compared to pre-pandemic levels in 2019 (Wang et al 2021). EDGAR suggests on-road traffic accounts for 10% of total anthropogenic CO₂ emissions, but a more recent emission inventory (Gately and Hutyra 2022) estimates a higher share (25%) in NYC. As the vegetation offsets up to twice the CO₂ enhancements by on-road traffic in the most congested NYC, we expect this is also the case for less congested cities especially with an increasing electric car share.

The net sign of biogenic CO₂ fluxes exhibits large diurnal and seasonal variations (e.g. large uptake during summer daytime) (figures 2(a) and (c)) that are relevant to accurate assessment of carbon mitigation policies. Remotely sensed variability in atmospheric CO₂ concentrations are often used to validate 'bottom-up' methods and to evaluate the carbon mitigation policies. However, these satellites pass the urban areas at a certain time of day, often in the afternoons when the biogenic influences on the CO₂ concentrations are strongest. Therefore, an accurate assessment of changes in observed atmospheric CO₂ concentrations attributed to anthropogenic emissions needs to tease out the influences of biogenic CO₂ fluxes.