Seasonal Variability in Local Carbon Dioxide Biomass Burning Sources Over Central and Eastern US Using Airborne In Situ Enhancement Ratios

We present observations of local enhancements in carbon dioxide (CO2) from local emissions sources over three eastern US regions during four deployments of the Atmospheric Carbon Transport‐America (ACT‐America) campaign between summer 2016 and spring 2018. Local CO2 emissions were characterized by carbon monoxide (CO) to CO2 enhancement ratios (i.e., ΔCO/ΔCO2) in air mass mixing observed during aircraft transects within the planetary boundary layer. By analyzing regional‐scale variability of CO2 enhancements as a function of ΔCO/ΔCO2 enhancement ratios, observed relative contributions to CO2 emissions were separated into fossil fuel and biomass burning (BB) regimes across regions and seasons. CO2 emission contributions attributed to biomass burning (ΔCO/ΔCO2 > 4%) were negligible during summer and fall in all regions but climbed to ∼9%–11% of observed combustion contributions in the South during winter and spring. Relative CO2 fire emission trends matched observed winter and spring BB contributions, but conflictingly predicted similar levels of BB during the fall. Satellite fire data from MODIS and VIIRS suggested the use of higher spatial resolution fire data that might improve modeled BB emissions but were not able to explain the bulk of the discrepancy.

primary source of CO is through combustion, other sources can include direct biological emission and oxidation of volatile organic compounds (VOCs) (Griffin et al., 2007). Enhancement ratios of CO vs CO 2 (ΔCO/ΔCO 2 ) can be particularly powerful for evaluating combustion sources, as the ratio of a plume from a single point source provides information about the source's combustion efficiency (CE). High efficiency (fuel-lean, high temperature) combustion produces relatively little CO, as the fuel carbon is nearly completely converted to CO 2 . Low efficiency (fuel-rich, lower temperature) combustion converts less of the fuel carbon to CO 2 , resulting in the release of greater amounts of intermediate combustion products, such as CO and organic compounds. For example, vehicular emissions in the United States typically have emission ratios in the range of <2% ΔCO/ΔCO 2 (Djuricin et al., 2010;Graven et al., 2009;LaFranchi et al., 2013;Turnbull et al., 2011), whereas modern power plant emissions typically are much more efficient, less than <0.1% ΔCO/ΔCO 2 (Peischl et al., 2010;USEPA, 2010). BB emissions typically have emission ratios on the order of 4% or higher ΔCO/ΔCO 2 (Akagi et al., 2011;Andreae & Merlet, 2001;Suntharalingam et al., 2004), which makes ΔCO/ΔCO 2 enhancement ratios a reliable marker for distinguishing uncontrolled BB from controlled fossil fuel (FF) combustion.
The recent NASA Atmospheric Carbon Transport-America (ACT-America) airborne study provides an ideal test case to examine trends in CO 2 sources over central and eastern United States and to compare them to model-derived seasonal and regional trends. In particular, the high-resolution, concurrent measurements of CO and CO 2 allow for a bottom-up perspective of the influence of various combustion sources on CO 2 emissions with respect to the inferred CE, allowing for apportionment between BB and FF emissions. We present a technique similar to that reported by Halliday et al. (2019), which utilized distributions of ΔCO/ΔCO 2 determined using a rolling background to examine relative combustion regime contributions and sources, expanding the method in order to ascertain contributions to local CO 2 emissions. The aim is to examine the seasonal and regional variability in CO 2 emissions over three regions in the eastern and central US using aircraft measurements obtained within the seasonal ACT-America field campaigns. These results are then compared to spatially and temporally concurrent modeled BB CO 2 emission products, as well as the satellite fire products that drive those emissions, in order to evaluate the model emissions response with respect to season and region in the eastern half of the United States.

ACT-America
The ACT-America campaign was a NASA Earth Venture Suborbital project focused on reducing errors in inversion models of the transport and emissions of atmospheric carbon dioxide and methane over the continental United States (Davis et al., 2021;Wei et al., 2021). ACT-America consisted of seasonal six-week intensives with three deployments per intensive; sampling locations and dates are shown in Table 1. Each deployment is sampled over three separate regions as shown in Figure 1. Measurements were collected using two aircraft: the NASA Wallops Flight Facility C-130 (N436NA) and the NASA Langley Research Center B-200 King Air (N529NA). The C-130 was instrumented with in situ chemistry measurements (Kostinek et al., 2019), whole air sampling (Baier et al., 2020), and remote sensing measurements Pal et al., 2020). The B-200 was instrumented with in situ measurements (Weibring et al., 2020) and whole air sampling measurements. The two aircraft combined to collect a mixture of planetary boundary layer (PBL) as well as lower and upper free tropospheric data. Three types of flight patterns were flown: frontal passage flights aimed at describing the transport of greenhouse gases (GHGs) by midlatitude cyclones , fair weather flights aimed at constraining regional fluxes, and OCO-2 validation flights aimed at quantifying CO 2 levels over multiple altitudes under an  (Bell et al., 2020). Flight patterns primarily focused on level altitude legs at either 300 m above ground level (AGL) for PBL measurements or constant pressure altitude flight levels above the PBL ranging up to 9 km MSL. Flights were conducted primarily in midday conditions. While flights were at times conducted in areas with broad, regional plumes, individual sources were not targeted.

In Situ Airborne Measurements
The two aircraft contained identical payloads for measuring in situ gas phase carbon species as described in Wei et al. (2021) and summarized here in brief. CO 2 , CO, and methane (CH 4 ) dry mole fractions were measured using a commercial cavity ringdown spectrometer (G2401-m, PICARRO, Inc.) with a custom gas sampling and calibration system ( Figure S1 in Supporting Information S1). The spectrometer cycled between measurements of each species sequentially every 2.5 s. The instrument temporal response through the gas system and instrument was measured to be typically ∼2-3 s. The calibration gas was humidified to the same level as the dried ambient sample air (typically 0.03%-0.05%), thus avoiding water vapor-dependent calibration discrepancies (Reum et al., 2019). Single concentration calibrations were performed hourly during flight to assess instrument offsets. Linear slope calibrations were conducted weekly on the ground through three-point calibrations over a broader concentration range. All calibration gases were traceable to the CO 2 X2007 (Tans et al., 2017), CO X2014A (Novelli et al., 1991), and CH 4 X2004A (Dlugokencky et al., 2005) WMO scales (NOAA ESRL). Inflight wingtip-to-wingtip in situ comparisons as well as in-flight in situ/flask comparisons (Baier et al., 2020), all agreed within experimental error. Measurement precision was 0.1 ppm, 5 ppb, and 1 ppb in 2.5 s for CO 2 , CO, and CH 4 , respectively. Measurement accuracy was 0.1 ppm, 2%, and 1 ppb for CO 2 , CO, and CH 4 , respectively.

Airborne ΔCO/ΔCO 2 Analysis
ΔCO/ΔCO 2 were derived using a short-term sliding slope window (Halliday et al., 2019;Smith et al., 2015). Using a sliding fixed-time bin window over the CO and CO 2 time series measured at ∼2.5 s intervals and binned at 5 s intervals, a linear regression of CO vs CO 2 is calculated for each period. This results in a linear slope equivalent to ΔCO/ΔCO 2 and a coefficient of determination (r 2 ) for each bin, where r 2 can then be used to filter For this work, running-bin linear regressions of ΔCO/ΔCO 2 were calculated using weighted orthogonal distance regression (ODRPACK95 -IGOR Pro v7), which Wu and Yu (2018) found to return equivalent unbiased slopes compared to the York regression used by Halliday et al. (2019). For each fit, CO and CO 2 mole fractions were weighted by the measurement precisions of 0.1 ppm for CO 2 and 5 ppb for CO. In order to focus on the relative enhancement ratios of local sources in the PBL, all data above 1 km AGL were rejected in order to focus on the 300 m level altitude flight legs. Values of ΔCO/ΔCO 2 were calculated using data from the ACT-America 5 s merge (Davis et al., 2018).
This resulted in binned frequency distributions of ΔCO/ΔCO 2 enhancement ratio slopes at 0.1% resolution. In Halliday et al. (2019), while the raw frequency distribution depended heavily on the choice of r 2 cutoff and bin window size, the normalized distribution was insensitive to both factors. To estimate the variability due to these factors, sensitivity tests were performed for each parameter over a range of r 2 values (0.3, 0.4, 0.5, 0.6, 0.7, and 0.8) and bin windows (30, 45, 60, 90, and 120 s), for a total of 30 different values. Figures S2-S5 in Supporting Information S1 shows line histograms of the distribution of ΔCO/ΔCO 2 observed during the four campaigns.
The results were similar to those observed by Halliday et al. (2019) in that there was wide variability in the raw frequency distribution intensities, but very similar normalized frequencies regardless of parameter value.
To examine the relationship between CO 2 and CE, an extension of the technique is required. Thus, each observed slope was binned by both ΔCO/ΔCO 2 and the total ΔCO 2 in the bin, the latter used as a metric for the CO 2 intensity of the emission. The result is a 2D heat map representing the enhancement in CO 2 as a function of ΔCO/ ΔCO 2 enhancement ratio ( Figure 2a). To calculate the ΔCO 2 -weighted distribution with respect to ΔCO/ΔCO 2 , the data were summed with respect to ΔCO 2 for each ΔCO/ΔCO 2 bin: where NWF is the normalized ΔCO 2 -weighted bin frequency as a function of ΔCO/ΔCO 2 , while n i,j is the number of points in the ith bin of ΔCO 2 and the jth bin of ΔCO/ΔCO 2 . Thus, NWF represents the relative contribution toward CO 2 emissions of a given ΔCO/ΔCO 2 enhancement ratio, and integrated NWF values over a range of ΔCO/ΔCO 2 enhancement ratios represent the relative contribution in CO 2 emissions from the CE source represented by that range. Figures 2b and 2c shows the resultant NWF distribution for the measurements collected during the spring campaign at less than 1 km altitude. The same sensitivity analyses to r 2 and bin width were performed as with the unweighted normalized method. NWF values were similarly insensitive to the choice of r 2 cutoff and bin size (Figures S6-S9 in Supporting Information S1), though with somewhat more variability than the unweighted method. Thus, the final NWF value was calculated as the average of the 30 values from the sensitivity analyses over the different combinations of r 2 and bin width parameters listed above, with the gray shaded areas in Figures 2b and 2c showing the full variability from these different cutoff choices. Instrument error was neglected for the NWF analysis (other than in the fits), as it is a relatively small contribution compared to the cutoff error (Halliday et al., 2019).
One of the key advantages of this technique is that, by focusing on ΔCO/ΔCO 2 slopes, it does not rely on broad regional assumptions about homogeneous background levels of CO and CO 2 . CO and CO 2 background mole fractions are calculated on a bin-sized time frame (i.e., 30-120 s). The aircraft ground speed was typically from 100-120 m/s at these altitudes, thus the spatial extent of the bin windows varied between 3 and 14 km. This smaller background scale results in more robust enhancement factors that do not experience the background error biases of a regional background method, as any background variability on spatial scales greater than 14 km would have no influence on the calculated enhancement factors. Additionally, as the NWF distributions were not strongly dependent on bin window size, any biases in calculated enhancement factors due to background variability between 3 and 14 km are captured by the NWF error. As a result, this method is rigorous to any changes in background on scales >3 km, though it is still subject to variability on smaller scales.
Another important caveat of the technique is that very high CE sources with very low ΔCO/ΔCO 2 emission ratios (e.g., power generation plants; Peischl et al., 2010;USEPA, 2010) could be missed if the measured enhancements were below the instrument precision. In addition, this technique does not describe the total amount of CO 2 emissions, only the relative contributions nearby point sources with different enhancement ratios from the background. Thus, the technique is internally consistent across seasons, well suited at looking at relative differences in contributions from different CE sources, but not a good predictor of absolute CO 2 emissions from BB and FF combustion. A near-field source would be observed as a stronger contribution than a more distant source, making the method more biased toward near-field sources. This should be somewhat mitigated by the tendency for more distant plumes to have broadened signatures, which would translate to a greater count frequency, albeit weaker, provided it has a significantly different ratio from the background. This mitigation should be less effective at higher bin widths, which may account for some of the greater variability in the ΔCO 2 -weighted NWF compared to the unweighted normalized frequency. This same effect makes it impossible to define an exact receptor footprint for the results other than this weighting effect on source distance.

Modeled BB CO 2 Emissions
Modeled CO 2 fire flux components were obtained from NOAA's CarbonTracker: version CT2017 (Carbon-Tracker Team, 2018) for the summer 2016 campaign and CT-Near Real Time (NRT).v2019-2 release (Peters et al., 2007) for the other 3 seasons. The fire module in both CT2017 and CT-NRT.v2019-2 models pyrogenic CO 2 emissions using the GFED4.1s and GEFD_CMS fire module (Giglio et al., 2013;van der Werf et al., 2017), which uses MODIS 1° fire products to detect fires and the CASA model to convert burned area to a CO 2 flux. Fire emission fluxes were calculated from the average of the outputs of the GFED4.1s and GEFD_CMS modules with 3 h time resolution and 1° latitude by 1° longitude spatial resolution over the same dates and years as the observations. To enhance the near-field comparison with the observed data, fire fluxes were subsampled to a 27 km 2 grid, roughly double the 14 km maximum bin window spatial extent of the ΔCO/ΔCO 2 NWF analysis. Regional uncertainty was estimated to be ∼50%, with possibly higher values in regions with a high prevalence of smaller fires (van der Werf et al., 2017).
The Vulcan inventory provides hourly CO 2 emissions at 1 km 2 resolution for the years 2010-2015 (Gurney et al., 2020b). CO 2 emissions are separated into 10 sectors: onroad (vehicles), electricity production, residential, nonroad (off-road vehicles), airport, commercial, industrial, commercial marine vehicles, rail, and cement. Emission data were sourced from various inventories, primarily the US Environmental Protection Agency National Emission Inventory. For comparison with the modeled fire emissions in this analysis, the 2015 hourly 1 km 2 Vulcan emissions were averaged spatially to the same 27 km 2 grid as the fire emissions. Cement sources were not included in the analysis, as the carbonate decomposition process that produces the majority of CO 2 emissions does not result in strong coemissions of CO (Andrew, 2018). Vulcan emissions were also averaged temporally between 0900 and 1700 local time to align with the aircraft flight times and minimize any biasing effects of the diurnal cycle of CO 2 emissions (Turnbull et al., 2015).

Seasonal ΔCO/ΔCO 2 Variability
NWF is weighted by the magnitude of the enhancement in CO 2 , and thus changes in the NWF distribution across various ΔCO/ΔCO 2 enhancement ratios can be used to evaluate the relative contributions of those CO 2 emission sources and their inferred CE. In this analysis, we classified the NWF distributions into regimes with respect to enhancement factor and use the variability between these regimes to infer the relative strength of the CO 2 combustion sources. Figure 3 shows the total integrated NWF contributions vs ΔCO/ΔCO 2 slope distribution for each deployment during different seasons, while Table S1 in Supporting Information S1 shows the numerical integrated NWF contributions in each regime.
Regimes were delineated by the natural minima in the NWF distributions. Common minima over all seasons exist at ΔCO/ΔCO 2 enhancement factors of 0% and ∼4%. The 0% minimum is a result of the technique, as enhancement ratios of 0% would have a low correlation. The 4% minimum is consistent with the lower end of reported biomass burning CE (Akagi et al., 2011;Andreae & Merlet, 2001;Suntharalingam et al., 2004). As a result, the FF regime is defined as ΔCO/ΔCO 2 enhancement ratios between 0% and 4%, and the BB regime is defined as enhancement ratios greater than 4%. A third regime consists of enhancement ratios with negative slopes (NS regime) or ΔCO/ΔCO 2 less than 0%. Since a negative enhancement ratio cannot be explained in the context of combustion efficiency, its existence implies a more complex mixing process likely influenced by CO 2 biogenic uptake as non-photochemical CO sinks are not known to be common. In particular, these negative enhancement ratios have been hypothesized to be associated with ecosystem uptake (Halliday et al., 2019;Silva et al., 2013) with either a photochemical or a well-mixed anthropogenic CO source. The implications of the NS regime are further discussed in Section 4.1. Efficient controlled combustion of non-fossil fuels (e.g., high-temperature wood fired furnace) could result in lower enhancement ratios akin to those typically expected for FF (Venkataraman & Rao, 2001), while inefficient combustion of FF (e.g., uncontrolled open oil burning) could result in higher enhancement ratios akin to those expected from BB (Middlebrook et al., 2012). These sources are relatively rare compared with the ubiquity of typical FF and BB combustion and thus will be neglected for the purposes of this analysis.
NWF contributions from NS regime peaked during the summer campaign, exhibited smaller contributions in spring and fall, respectively, and were negligible in winter. This pattern is consistent with the expected CO 2 biogenic processing behavior in each season. FF regime contributions were consistently the majority in all seasons, though relatively more prevalent during the fall due to the lack of BB regime contributions and lower NS contributions compared to summer. While in summer, enhancement ratios within the FF regime were predominantly between 0% and 1%, while NWF contributions at enhancement ratios between 1% and 4% became much more significant particularly in the winter and spring. It remains uncertain whether this is a real variability from differing seasonal FF CE, but it is likely influenced by biogenic CO 2 uptake.
BB regime NWF contributions were observed to be very weak in the summer and fall but were ∼4% and ∼2.5% of observed NWF in winter and spring, respectively. This finding is consistent with visual observations of BB during the campaign, as agricultural fires were observed to be common especially during the winter campaign. The primary BB regime peaks at enhancement ratios between 5% and 8%, in both winter and spring and is indicative of higher CE flaming BB (Andreae & Merlet, 2001). However, weaker peaks at higher enhancement ratios were observed between 10%-14% in winter and 9%-12% in spring, indicating the additional presence of smoldering and/or mixed phase sources.

Regional ΔCO/ΔCO 2 Variability
Observed enhancement ratios were also segregated into three ACT-America flight domains: Mid-Atlantic, Midwest, and South ( Figure 1) to examine how well the calculated CE aligns with ACT-America observations and our current knowledge of activities in these three regions. Figure 4 shows the relative seasonal NWF contributions for each regime within the PBL both in total and for each region. In summer, the strong NWF contributions in the NS regime (Figure 4c) are primarily driven by observations in the Mid-Atlantic and Midwest regions, with South region NS contributions only about 25% of that of the other regions. This general trend is consistent with previously reported regional trends in measurements from the Orbiting Carbon Observatory (OCO-2) of solar-induced fluorescence (SIF). Sun et al. (2018) discussed averaged SIF retrievals for June-August 2015 and higher SIF levels were observed over the East and Midwest regions compared to the South. As SIF has been shown to be correlated with gross primary production (Sun et al., 2018), which is related to the rate of bio- sphere carbon fixation, higher SIF levels imply greater biospheric CO 2 uptake. This is further evidence, albeit circumstantial, of the relationship between the NS regime and biogenic uptake. Southern region NS contributions were at their highest during spring, possibly due to the earlier start to the agricultural growing season. Fall and winter NS contributions were much smaller than the other seasons in all regions, with <10% average NWF contributions.
In the BB NWF regime (Figure 4a) . During fall, winter, and spring, FF NWF regime contributions were similar across all regions but were weaker during summer in the Mid-Atlantic and Midwest compared to in the South, corresponding to the weaker NS regime contributions observed in the South during summer.

Biogenic Uptake and the NS Regime
Biogenic uptake can affect the calculation of the ΔCO/ΔCO 2 enhancement ratios in several ways. For instance, if the CO 2 background fluctuates due to uptake on the scale of the bin window, then these fluctuations would bias the fit slope potentially both positively and negatively, as well as reducing the goodness of fit. The agreement of NWF in the sensitivity analysis with respect to bin window size, equating to a range of 3-14 km, provides confidence that this variable was not a strong factor in the results, and the net effects of any CO 2 background variability within this range would be represented by the NWF uncertainty. Variability below 3 km could still affect the analysis, but those scales approach the size of the PBL during ACT-America (typically 1-2 km; Gonzalez et al., 2021), likely resulting in any variability at those small scales to be well-mixed. Thus, this form of the influence of biogenic uptake of CO 2 is not expected to be a significant influence on enhancement ratios in this analysis. Additionally, oxidation of biogenic volatile organic compounds (VOCs) can lead to significant CO production. As biogenic VOC emissions should correlate with biogenic CO 2 uptake, this would result in increased observations of negatively sloped enhancement ratios. Gonzalez et al. (2021) found that in an analysis of North American summer CO sources, including the ACT-America domains, the combination of biogenic VOC oxidation and fire CO sources represented a plurality (40%-45%) of above background enhancements in CO compared to only 9%-16% from FF sources. This contrasted with other seasons where FF remained the largest regional CO source. This trend circumstantially agrees with the strong NS NWF contribution observed during the summer compared to other seasons, in which case it would be expected that the enhancements in the NS regime are independent of the FF regime.
Another effect involves the mixing of air with different biogenic uptake history. For example, if an air mass with a well-mixed combination of biogenic CO 2 uptake and FF emissions mixes with a clean air mass with less biogenic CO 2 uptake, the difference in CO 2 due to the uptake signature could artificially increase the observed enhancement ratio of the FF source compared to its actual CE. If the cleaner source experienced more biogenic uptake than the FF-influenced source, the apparent enhancement ratio would be artificially decreased compared to the FF source's CE. This explanation is also consistent with the seasonal trend in the NS regime but has different implications. Unlike the biogenic VOC oxidation theory, the mixing theory would affect both positive and negative enhancement ratios. It also provides insight into what extent this bias may play in the results of this analysis. In Figure 3, the NS regime NWF contributions are at ΔCO/ΔCO 2 slopes > −1%, which suggests that the extent of this effect is likely to shift enhancement ratios no more than 1%-2%. Combined with the distinct distribution of the BB NWF contributions at ΔCO/ΔCO 2 slopes >4% vs. the FF regime, we can infer that any bias predominantly affects our calculation of the FF regime. While the enhancement ratios of the BB regime may still be affected by this form of bias, the distinct BB regime distribution would negate this effect on the integrated BB regime contribution.
The BB regime can still be affected indirectly in terms of relative intensity. Since our analysis technique yields relative distributions, any bias in the FF regime would change the scale of the total signal in the BB regime. Thus, a sensitivity test using the NS regime data can help constrain some of this FF regime bias that we observe. For this test, we examined three cases. The base case involved normalizing the integrated BB regime by the integrated FF regime, neglecting the NS regime. This case represents one in which the NS regime was exclusively driven by observations of biogenic VOC oxidation-sourced CO plumes combined with CO 2 biogenic uptake, where mixing with these two regimes would have no effect on the positive enhancement ratios. The second case involved normalizing the BB regime by the sum of the FF and NS regimes. This case represents an NS regime entirely driven by FF emissions that were shifted toward negative slopes by the changes in biogenic background. The third case involved normalizing the BB regime by the difference between the FF and NS regimes. This case represents an NS regime driven by mixing, but one where this affected the negative slopes and positive slopes equally. In other words, the combined mixing and biogenic uptake would create an equal bias in both the positive and negative sloped regimes. All of these cases are meant to be extremes, which can be used to constrain the potential influence on the BB regime.
For each of these cases, the normalized BB regime was calculated for each of the same 30 combinations of r 2 and bin window settings as discussed in Section 2.3. Figure 5 shows the results of this sensitivity test. Despite summer having the largest contributions to the NS regime, the BB/FF ratio for all cases in each region remained negligible due to the small BB regime contributions. One exception was in the South, where BB was normalized by the difference between the FF and NS regimes. In this case, for some combinations of r 2 and bin width, the NS regime contributions were larger than those from the FF regime, resulting in a prediction of negative BB/FF. As it is unlikely that no signatures of FF combustion were detected, combined with the lack of absolute detections of BB enhancement ratios, this appears to be a statistical artifact of this worst-case scenario. During fall and winter, the difference between the three cases is minimal, as NS regime contributions were low during both seasons. In contrast, the spring season had the largest variability between cases due to the combined high BB regime contributions in the South and significant NS regime contributions. Overall, the systematic trends between regions and seasons stand over all cases. As the base case uncertainty encompasses the majority of the variability within the other two extreme cases, the remaining discussion will focus on results using the base case.

Modeled CO 2 Fire Emissions Comparison
CT2017 and CT-NRT.v2019-2 simulated fire emissions were subsampled along the ACT-America flight track at 5 s intervals from the 27 km 2 and 3 h resolution pixels and for flight legs below 1 km AGL, then averaged seasonally and by region (Figure 6b). The largest modeled fire contribution in the Mid-Atlantic region was during summer at 10 ± 5 mol CO 2 /km 2 *h, with other seasons averaging less than 1/3 the fire emissions of summer. Midwestern modeled fire average contributions were highest in spring at ∼7.5 ± 3.8 mol CO 2 /km 2 *h, with emissions in other seasons weaker by an order of magnitude. The Southern region had the highest overall average fire emissions during the fall, winter, and spring seasons, ranging near 20 ± 10 mol CO 2 /km 2 *h, with a strong drop during summer to ∼6 ± 3 mol CO 2 /km 2 *h.
As the airborne ΔCO/ΔCO 2 analysis yielded relative CO 2 contributions from BB compared to overall combustion, the magnitude of these emissions cannot be directly compared to the modeled fire contribution. This can be mitigated by normalizing the fire product by the expected FF emissions in each region. Figure 6c shows the same average modeled fire CO 2 emissions as in Figure 6b but normalized by the average 2015 Vulcan modeled total FF CO 2 emissions in each region and season in order to account for the sector variability in overall FF emission. In the same fashion as the simulated fire emissions, data from each sector in the Vulcan inventory (Gurney et al., 2020b) were subsampled to 27 km 2 resolution pixels and interpolated at 5 s intervals along ACT-America flight tracks below 1 km AGL. As seen comparing Figures 6b and 6c, the effect of the FF-normalization did not qualitatively change the trends observed in the fire product averages. This minimal effect combined with the low variability in annual total US FF CO 2 emissions estimated by Vulcan between 2010 and 2015 (99%-106% of 2015 . Patterns indicate sensitivity case, where "FF-NS" is normalization by difference between the FF and NS regime contributions, "base case" is normalization by just the FF regime, and "FF + NS" is normalization by the sum of the FF and NS regime contributions. Error bars denote range in variability from r 2 and bin window sensitivity analysis. emissions; Gurney et al., 2020b) supports the validity of applying the 2015 CO 2 emissions during our 2016-2018 campaign years.
The FF-normalized modeled fire emission ratios (Fire/FF) are directly comparable to the airborne BB/FF ratios. Seasonal and regional trends in the normalized Fire/FF ratios were very similar to those that form the fire product. Figure 6a shows the base case results of the sensitivity analysis described in Section 4.1. The airborne BB/FF ratio values were <0.5% for all regions during summer and fall. Comparably high BB/FF ratios were observed in the South during winter and spring, low ratios were observed in the Mid-Atlantic region in both seasons and the Midwestern spring, whereas Midwestern winter ratios were between the two scenarios. The modeled Fire/FF ratios captured the high airborne BB/FF ratios during winter and spring in the South compared to other regions, with a modeled Fire/FF winter to spring ratio of 1.05 ± 0.74 compared to an airborne FF/BB of . However, there are three major discrepancies to highlight. The largest discrepancy is the low Fire/FF emission ratio predicted by the model in the South during fall. The modeled winter to fall Fire/FF ratio in the South was ∼1.4 ± 1.0, while the equivalent airborne winter to fall fire ratio was much higher at   296 35 47 E . Another strong discrepancy was the relative winter to spring ratio in the Midwest, as the modeled Fire/FF spring to winter ratio was 17 ± 12 while the opposite was true for the airborne data, with a BB/FF spring to winter emission ratio of   0.27 0.11 0.11 E due to the relative lack of observed fire emissions in the spring. The final major discrepancy was the marked abundance of modeled emissions in the summer in both the Mid-Atlantic and Southern regions, as the airborne BB/FF ratios were negligible during that season in all three regions.

MODIS and VIIRS Fire Data
As the modeled fire product is parameterized using MODIS 1° fire counts (Jacobson et al., 2020), examining trends in satellite fire count could help understand the differences between the modeled results and aircraft observations. MODIS 1 km (FIRMS, 2020a) and VIIRS-SNPP 375 m (FIRMS, 2020b) fire radiative power (FRP), a measure of the irradiative intensity of the fire, was analyzed over each ACT-America campaign season and region. FRP is used by models to determine the amount of combusted organic matter, and thus should scale with CO 2 emission (Kaiser et al., 2012). Data were filtered for flight days and aircraft altitudes below 1 km AGL. For MODIS, data were filtered for fires detected within 50 km (∼±0.5°) of the aircraft flight tracks, the same resolution as the MODIS product used to drive the fire emissions. For VIIRS, data were filtered for fires detected within 14 km of the aircraft flight tracks, comparable to the 27 km 2 resolution of the modeled data. Figures 6d and 6e summarize the FRP-weighted sum of fire counts from each instrument by season and region for MODIS and VIIRS, respectively, while Figures S12 and S13 in Supporting Information S1 show the full spatial distribution of the fire counts and FRP. This is a much-simplified approach to methods described in the literature used to translate FRP to gas emissions but is applied here as a tool to provide insight into the model/airborne agreement. Broadly, the MODIS fire products agreed well with the fire product (Figure 6b). The highest number of fire counts were in the South for all seasons, and there were many fewer counts in the South during summer compared to the other seasons, both matching the modeled fire emissions. This is consistent with Zhang et al. (2010), who found strong correlations in the southeastern US between levoglucosan concentrations, a biomass burning tracer and MODIS fire counts for all seasons except December and January. This discrepancy was hypothesized to be due to the increased occurrence of small-scale residential burning undetectable by MODIS. One of the biggest discrepancies between MODIS and the modeled fire emissions was during summer. While the modeled fire emissions were highest in the Mid-Atlantic region during summer, the MODIS weighted counts were lowest in the Mid-Atlantic. Additionally, the modeled fire emissions in the Midwest during winter were a factor of ∼7 smaller compared to those from the Mid-Atlantic region, and the two regions had comparable MODIS weighted counts. The causes for this may be attributable to differences in the very simple FRP weighting approach used here and the more complex analysis performed by the GFED and CASA modules.
Results using the VIIRS weighted counts were significantly different from MODIS. The ratio of Southern spring to winter weighted counts was ∼90% from MODIS compared to ∼40% from VIIRS, and the ratio of Southern fall to winter weighted counts dropped from ∼115% from MODIS to ∼45% from VIIRS. Additionally, the ratio of winter to spring weighted counts in the Midwest increased from ∼55% with MODIS to ∼300% with VIIRS. As two of the largest discrepancies between the modeled and airborne emissions were the modeled high emissions in the South during fall and the ratio of winter to spring emissions in the Midwest, these shifts provide some circumstantial evidence that spatial resolution of either the satellite product or model may be contributing to those discrepancies. Given the uncertainty in the exact range of sensitivity of the airborne NWF method, it is possible as well that the 14 and 50 km cutoffs were too small. Figures S14 and S15 in Supporting Information S1 show a sensitivity analysis looking at total FRP from MODIS and VIIRS, respectively. While these larger scales do result in higher relative total FRP in the Southern region across all seasons, they still fail to reproduce the much-reduced level of fire activity during fall observed in the airborne FF/BB ratios.

Conclusions
In this study, we used airborne measurements conducted during the ACT-America campaigns of CO and CO 2 in the PBL to examine the relative frequency of regional and seasonal CO 2 enhancements as a function of ΔCO/ ΔCO 2 enhancement ratio, used as a proxy for CE, over central and eastern US through weighted sliding correlations. Observed enhancement ratios were separated into three regimes: biomass burning (BB), fossil fuel (FF) combustion, and those with apparent negatively sloped (NS) enhancement ratios. Contributions from the NS regime were high in the summer, complicating the analysis. However, sensitivity analyses with the seemingly biosphere-related NS regime show that the trends in the normalized BB/FF contributions were not strongly affected by this interference. Based on the airborne observations, PBL CO 2 BB emission contributions (ΔCO/ΔCO 2 > 4%) relative to FF were seen to be strongest in South in winter and spring, with Mid-Atlantic BB contributions very low or negligible for all seasons. Modeled CO 2 Fire/FF emissions were calculated using modeled fire CO 2 emissions from NOAA's CarbonTracker, versions CT2017 and CT-NRT-v2019-2 and modeled FF CO 2 emissions from the Vulcan v3.0 2015 inventory. These modeled CO 2 Fire/FF emissions agreed with these high relative fire emissions in the South, but also predicted enhanced Fire/FF emissions in the South during fall, in the Midwest during spring, and in the Mid-Atlantic region in the summer. Analysis of FRP-weighted satellite showed that while the 1 km MODIS fire data more accurately reproduced the modeled fire emissions, the 375 m VIIRS fire data reduced the overpredictions during the Southern fall and Midwestern spring. This suggests that the spatial resolution of the satellite products driving the model affects the measurement/model discrepancy, though does not explain the discrepancy in the Mid-Atlantic summer. These results imply that a combination of factors, such as undetected smaller agricultural fires below satellite product resolution or insufficiently constrained biosphere data, may cause significant biases in predictions of BB CO 2 emissions in the US. Additionally, as air quality models use similar modules to drive BB VOC and CO emissions, these same biases would likely affect predictions of regional air quality as well. The Atmospheric Carbon and Transport (ACT)-America project is a NASA Earth Venture Suborbital-2 project funded by NASA's Earth Science Division (Grant NNX15AG76G to Penn State). Co-author SP was supported by NASA Grant Number 80NSSC19K0730 and Texas Tech University. We would like to thank James Plant, James Geiger, and Ali Aknan for their assistance with in situ instrument payload preparation and data collection. Thanks to Michael Obland, Byron Meadows, and Bing Lin for mission support and coordination. Thanks to Jocelyn Turnbull for helpful suggestions throughout the review process. Thanks to NASA Wallops Flight Facility and NASA Langley Research Center for aircraft operations and logistics support. Thanks to Melissa Yang and BAER/NSRC for aircraft navigational and meteorological data support. Thanks to Duncan Aviation and TAC Air for FBO and logistics support during the ACT-America campaigns. CarbonTracker CT2017 and CT-NRT. v2019-2 results provided by NOAA ESRL, Boulder, Colorado, USA from the website at http://carbontracker.noaa.gov.