In order to explain the amount of organic aerosol (OA) as well as its spatial distribution and 14C content, several groups have advanced the hypothesis that the secondary component (SOA) is derived from biogenic precursors but its formation depends on interactions between urban anthropogenic emissions and a larger, geographically dispersed pool of biogenic emissions, hereinafter referred to as A–B interactions (de Gouw et al., 2005; Weber et al., 2007; Goldstein et al., 2009; de Gouw and Jimenez, 2009; Carlton et al., 2010; Worton et al., 2011; Xu et al., 2015). A–B interactions lie at the intersection of two problems. First is an explanation of unexpectedly high concentrations of OA, which are most prominently noticed downwind of urban areas (e.g., Volkamer et al., 2006; Kleinman et al., 2008; Matsui et al., 2009). Second is the finding that a high fraction of OA consists of non-fossil carbon of biogenic origin, even downwind of urban areas (Schichtel et al., 2008; Marley et al., 2009; Hodzic et al., 2010; Zotter et al., 2014).

Progress has been made on the problem of models predicting lower OA than observed. Older models (ca. pre 2005) are now recognized to contain a limited set of aerosol precursors and aerosol formation mechanisms. Some part of the gap between observations and theory can be closed using updated models that contain new categories of anthropogenic compounds, chemical mechanisms, and physical interactions (e.g. Robinson et al., 2007; Hodzic et al., 2010). In order to agree with 14C measurements, the gap has to be closed with a significant fraction of non-fossil carbon, which brings us back to the possibility of A–B interactions.

Evidence for, and mechanisms of, A–B interactions have been reviewed by Hoyle et al. (2011). Pertinent findings are that (1) the major source of volatile organic compounds (VOCs), globally and in many well-studied regions, such as the southeastern US and Canadian boreal forest, is biogenic (e.g., Guenther et al., 1995; Goldstein et al., 2009; Slowik et al., 2010), (2) large biogenic emission rates and SOA yields are a possible cause for the high fraction of non-fossil carbon found in the summer in many locations (Schichtel et al., 2008; Hodzic et al., 2010), including those that are nominally urban (Weber et al., 2007; Marley et al., 2009), and (3) there are realistic mechanisms whereby biogenic SOA yields depend on the presence of anthropogenic pollutants (Carlton et al., 2010). The latter include effects of anthropogenic pollutants on oxidant levels and consequently on biogenic VOC oxidation rates (Kanakidou et al., 2000), increased partitioning of biogenic VOC oxidation products to the aerosol phase because the aerosol volume, including associated water, available for partitioning is increased by an anthropogenic component (Carlton et al., 2010; Carlton and Turpin, 2013), aqueous phase reactions (Ervens et al., 2011), and effects of sulfate on aerosol phase chemistry (Xu et al., 2015). Explanations of non-fossil carbon based on A–B interactions are constrained by the observation that the difference between modeled and observed OA concentration generally decreases as anthropogenic influence decreases (Tunved et al., 2006; Chen et al., 2009; Hodzic et al., 2010; Slowik et al., 2010)

In multiple studies it has been found that OA, SOA, and/or WSOC (water soluble organic carbon; demonstrated to be a surrogate for SOA) are highly correlated with anthropogenic tracers. High correlations have been observed even at locations where it is suspected that much if not most SOA is biogenic. In a study of the Atlanta region, aircraft flights over the urban core established that WSOC is proportional to carbon monoxide (CO), while surface observations of aerosol 14C, also in the urban core, established that 70–80 % of WSOC was non-fossil and likely biogenic (Weber et al., 2007). While there was a high correlation between WSOC and CO in Atlanta, similar to that observed in the New York City metropolitan region (Sullivan et al., 2006), there was no clear linkage between WSOC and biogenic VOCs. At the Blodgett Forest Research Station, located 25 km downwind from the Carbonaceous Aerosols and Radiative Effects Study (CARES) sampling region, OA was observed to be correlated with CO (R2=0.79) but based on 14C filter samples it was determined that the majority of aerosol carbon was non-fossil from biogenic sources (Worton et al., 2011). Although high correlations between OA and biogenic VOCs have been reported (e.g. Slowik et al., 2010), it is more typical that in regions with an urban influence, correlations are low and in general do not suggest a relation between SOA and biogenic precursors. In a set of global calculations having the objective of satisfying the dual constraints of matching observations of OA concentration and 14C content, it was found that the best of many mechanisms considered was one in which the geographic distribution of OA followed that of anthropogenic CO, but had a 14C content appropriate to a biogenic precursor (Spracklen et al., 2011).

An objective of the CARES field campaign, conducted in June 2010, was to determine whether and to what extent the simultaneous occurrence of high concentrations of anthropogenic and biogenic compounds is associated with enhanced concentrations of OA. An overview of the field campaign is given by Zaveri et al. (2012) and a detailed description of the meteorology and emission source regions affecting the Sacramento plume provided by Fast et al. (2012). Effects of A–B interactions have been examined during the CARES campaign by Shilling et al. (2013) using aircraft data from the G-1 and by Setyan et al. (2012) using data from the rural T1 surface site located 40 km northeast of Sacramento. Both studies considered the ratio (OA - background) / (CO - background) and found that this ratio increased by a factor of ∼ 3 when an anthropogenically influenced air mass also contained high concentrations of biogenic VOCs. At a CO concentration, indicative of a significant anthropogenic impact, both studies find that OA concentrations are higher when the CO is accompanied by higher concentrations of biogenic VOCs.

In this study we use linear and bilinear regressions to empirically relate OA to tracer compounds of anthropogenic or biogenic origin. CO is used as a nearly inert tracer of anthropogenic emissions, that upon atmospheric oxidation lead to SOA. A comparison is made between calculations in which isoprene, methyl vinyl ketone and methacrolein (MVK + MACR), or methanol (CH3OH) were used as tracers of biogenic emissions. Isoprene and MVK + MACR, in contrast to CO, have short atmospheric lifetimes; therefore, their presence only explicitly addresses biogenic inputs to an air mass over a few hours time span. CH3OH, has a primary biogenic source and a lifetime of order 10 days (Schade and Goldstein, 2006; Wells et al., 2012) and therefore can provide information on biogenic inputs over a time span comparable to the lifetime of tropospheric aerosols.

In order to understand the roles of anthropogenic and biogenic tracers in describing SOA formation over spatial scales comparable to the Sacramento plume, correlation coefficients between OA and explanatory variables have been determined for each plume transect. For the purpose of determining the sensitivity of OA to conditions that vary over the CARES campaign, we define for each plume transect and species a background concentration and plume perturbation, Δ, and use these quantities in a regression analysis amongst transects. Transects are also split into subsets having the varying combination of low and high values for ΔCO and Δisoprene, ΔMVK + MACR or ΔCH3OH. We find that the data subset with high concentrations of both anthropogenic and biogenic tracers has uniquely high values of ΔOA. This result is similar to the A–B enhancement found by Setyan et al. (2012) and Shilling et al. (2013). We consider whether the uniquely high values of ΔOA can be explained by an independent linear dependence of ΔOA on anthropogenic and biogenic tracers, rather than a synergistic effect.

OA and its anthropogenic and biogenic precursors are expected to be mutually enhanced through a common dependence on meteorological conditions (e.g. ventilation, sunlight, and temperature) occurring in pollution episodes (Goldstein et al., 2009). Such conditions promote A–B interaction and may also give rise to an altered, non-synergistic dependence of OA on precursors. Our analysis cannot distinguish between causes of enhanced OA, leading us to equate the net effect of enhanced OA above that expected from a bilinear model of low concentration data to an A–B interaction. We find a high correlation between ΔOA and ΔO3 (Herndon et al., 2008; Wood et al., 2010). Studies of the dependence of O3 on meteorological factors and on anthropogenic and biogenic precursors have a long history and could yield insights on SOA production.

Although there is a high anthropogenic, high biogenic, subset that stands out as having high concentrations of ΔOA, most of the spatial variability of OA within a transect and ΔOA amongst transect can be explained by CO or ΔCO, respectively. These observations suggest a primarily anthropogenic origin for OA produced in the Sacramento plume. In contrast, the variability of background OA is much better explained by background CH3OH, which suggests a biogenic origin. As background OA is more abundant than OA formed in the Sacramento plume, the plume OA, though correlating best with CO, is expected to have a 14C signature of non-fossil, biogenic carbon. The plume composition would then resemble that observed in Atlanta (Weber et al., 2007) and that observed in the afternoon outflow from Los Angeles (Zotter et al., 2014).

Data for the CARES field campaign including the G-1 data used in this study can be accessed from DOE's ARM (Atmospheric Radiation Measurement) Climate Research archive at http://www.archive.arm.gov. AMS (Aerosol Mass Spectrometer) and PTR-MS (Proton Transfer Reaction Mass Spectrometer) data were recorded at the instrument cycle time of 13 and 4 s, respectively. Most other measurements were collected at 1 Hz. Data used in this study was interpolated or averaged to a 10 s time base. Units for aerosol concentration are µg m-3 at 1013 mb and 23 ∘C. Trace gas abundances are expressed as mixing ratios in units of ppbv. When referring to aerosol and gas phase species collectively, the term concentration is used.

In our analysis of the G-1 data set we consider variations in OA observed along a plume transect and a larger-scale variability that includes the effects of day-to-day changes in meteorology. Variations in OA are correlated with anthropogenic and biogenic explanatory variables, which are surrogates for the actual compounds that form SOA.

Tables 3 and 4 summarize chemical measurements from the G-1, providing backgrounds and plume perturbations for OA, CO, isoprene, MVK + MACR, CH3OH, and O3 averaged over transects according to location. Background concentrations of the long-lived constituents, OA, CO, and CH3OH are to the first order about the same on all five legs. The short-lived species, isoprene and MVK + MACR, increase by approximately an order of magnitude from east to west, following changes in emissions. Ozone is an intermediate case. Calling the 10th percentile of concentration “background” comes closest to matching the traditional definition for species with long atmospheric residence time. Even so, backgrounds for OA, CO, and CH3OH have significant variability, which we will take advantage of for source attribution of OA.

From T0 eastward, ΔCO is elevated relative to Upwind values. Biogenic mixing ratios have their major increase east of the City-Edge transect. Effects of dilution are apparent in the decrease of ΔCO and ΔOA at the Foothills transect, albeit based on only five transects from three flights.

Three analysis methods were used. In method 1, spatial variations in OA are compared with spatial variations of CO and Bio on a single transect. Our metric for a successful explanatory variable is that it accounts for a high fraction of variance in OA and that the standardized regression slopes, i.e. βs in model M4S, indicate that the explained variance is not spurious. By these criteria the average dependence of OA on CO is greater than its dependence on the biogenic tracer CH3OH, and much greater than its dependence on isoprene or MVK + MACR.

Although, on average, most of the explained variance in OA is due to CO, there are transects in which OA is highly correlated with isoprene, MVK + MACR or CH3OH (Figs. 3, S1–S3). Examples for MVK + MACR are the sequential-in-time City-Edge and T1 transects from the 608b flight shown in Figs. 4 and 5. Complicating the interpretation of method 1 in general and the 608a transects in particular, is the circumstance that the anthropogenic and biogenic tracers, CO and MVK + MACR, are themselves often highly correlated. As can be seen by comparing the standardized coefficients in Table 6 with values of R2 for models M1–M4, small difference in correlation coefficients can be associated with large changes in the relative importance of CO and MVK + MACR as judged by the standardized regression slopes. By most measures an R2 of 0.89 for OA vs. MVK + MACR on the 608b City-Edge transect would be considered excellent. But, because of the slightly higher R2 for OA vs. CO (0.94) and the high correlation between CO and MVK + MACR (0.89), a bilinear model assigns most of the variance in OA to CO. An opposite conclusion is reached for the 608b, T1 transect.

The spatial correlation between CO and, for example, MVK + MACR on individual transects is highly variable (Figs. 3b and S2). On average it is somewhat greater than the correlation between OA and MVK + MACR. It is not clear what processes cause a sometimes high but variable correlation between anthropogenic and biogenic VOCs. A possible explanation is that it is an accident of geography whereby under certain wind directions the plumes from urban and forested areas line up, while under other wind directions the overlap is lessened or eliminated. Faster oxidation of isoprene in the high NOx anthropogenic Sacramento plume could be a contributing factor. Perhaps related is the observation by Dreyfus et al. (2002) of the co-advection of anthropogenic and biogenic compounds from the direction of Sacramento to the Blodgett Forest Research Station.

In Method 2, plume perturbations (Δs) are defined on each downwind transect and for each species of interest. Correlation coefficients calculated amongst transects quantify the extent to which linear and bilinear combinations of Δ tracers (CO, isoprene, MVK + MACR, CH3OH, and O3) can explain the variations of ΔOA that occur as chemical and meteorological conditions vary from flight to flight and with transect location. As with method 1, we find that CO is more successful in explaining the variability of OA, in comparison to biogenic tracers. This can be seen from the scatter diagrams in Fig. 7 and from the standardized coefficients for the bilinear model M4S in Table 7.

A notable feature of the correlations amongst transects is that ΔO3 explains 88 % of the variance in ΔOA. A display of Fig. 7j on a log–log scale (not shown) indicates that the ΔOA–ΔO3 relation stays approximately constant at low concentrations. Figure 7j also shows that the same relation holds at each of the four downwind transect locations. The high ΔOA–ΔO3 correlation is evidence that most OA above background originates from secondary chemistry. This assignment is in agreement with findings from many locations that SOA is correlated with other oxidized species and in agreement with the analysis of CARES observations by Setyan et al. (2012) and Shilling et al. (2013). A mechanistic reason for a relation between OA and O3 was proposed by Herndon et al. (2008) based on a chemical mechanism in which OH + VOC was the rate limiting step in O3 production, and also in forming low volatility VOC oxidation products that partition to the aerosol phase. The least-squares fit in Fig. 7j has a slope of 225 µg m-3 per ppm O3, as compared with the 104–180 µg m-3 per ppm O3 slopes found by Herndon et al. (2008) in Mexico City outflow. Differences could reflect a ±30 % uncertainty in AMS measurements (Canagaratna et al., 2007) or variations in processing conditions.

Because of the mismatch in atmospheric lifetimes between isoprene or MVK+MACR and OA, it is not clear whether a correlation should be expected. Shilling et al. (2013) have made an argument based on changes between pseudo Lagrangian morning and afternoon measurements, that an anti-correlation is consistent with OA formation from biogenics that have been depleted by oxidation reactions. Robust relations between SOA and biogenic tracers are typically not found. For example, de Gouw et al. (2005) present aerosol and trace gas measurements from the Ronald H. Brown, taken off the coast of a region in Maine with high biogenic emission rates. Sub-µm OA was observed to be highly correlated with the anthropogenic tracer, iso-propyl nitrate, but not correlated with either monoterpenes or isoprene. A similar lack of correlation was found by Sullivan et al. (2006) and Weber et al. (2007). In contrast, Slowik et al. (2010) showed an R2 of 0.71 for a regression of less oxygenated SOA (OOA-2; O / C = 0.46) vs. MVK + MACR, concluding on the basis of the MVK + MACR lifetime that OOA-2 was formed in the relatively recent past. Setyan et al. (2012) also observed a high correlation (R2 = 0.61) between MVK + MACR and a more-oxidized OOA (O / C = 0.54) at T1.

Unlike isoprene and MVK + MACR, the atmospheric lifetime of CH3OH is comparable to that of SOA. Standardized regression coefficients averaged over all transects (Table 5) indicate that CH3OH captures the spatial variance in OA better than short-lived biogenic tracers. The same is true for regressions amongst transects based on Δ values (Table 7). However, in methods 1 and 2, most of the variance in OA or ΔOA is still captured by CO. Correlations and standardized regression coefficients, for individual transects and amongst transects, are therefore consistent with an anthropogenic source for most of the OA formed downwind of Sacramento.

Figure 7 shows that the variance in background OA is uniquely captured by background CH3OH, with an explained variance of 82 % compared to 23 to 27 % for background CO, isoprene, or MVK + MACR. The excellent fit between background OA and background CH3OH and the poor fit with CO is what would be expected if background SOA is primarily biogenic.

In method 3, concentrations in the form of Δ values are used to define plume-transect subsets, which have low or high mixing ratios of CO and either isoprene, MVK + MACR, or CH3OH. Of the four combinations of low and high values of ΔCO and ΔBio, only the subset with a high mixing ratio of both anthropogenic and biogenic tracers has high average ΔOA. However, according to method 1, most transects have a correlation between OA and biogenic tracers that is low and/or spurious, such that the addition of a biogenic to an OA vs. CO model, produces only a modest improvement in explaining the variance of OA during a plume transect. In method 2, there is a poor correlation between ΔOA and Δisoprene or ΔMVK + MACR. The correlation between ΔOA and ΔCH3OH is somewhat higher but still lower than that between ΔOA and ΔCO. We recognize the deficiencies in using short-lived biogenic tracers, yet in method 3 these same biogenic tracers as well as the long-lived CH3OH can split the data set into subsets with and without high ΔOA.

The association between simultaneous high mixing ratios of anthropogenic and biogenic tracers and high ΔOA is heavily reliant on data from two flights on 28 June as this was the only day in which ΔOA exceeded 6 µg m-3. What was distinctive about this day? According to Fast et al. (2012), from 22 June till the end of the field campaign on 28 June, winds at 700 hPa were light and variable. After 25 June there was a steady increase in maximum day time temperature; 27 and 28 June had the warmest temperatures, approaching 40 ∘C at T0 on the 28th. Ozone also reached its highest value of 90 ppb at T0. Wind speeds at the G-1 altitude were ∼ 2 m s-1. A detailed description of chemical conditions on 28 June is given by Shilling et al. (2013). In brief, during these two flights isoprene reached 13 ppb, O3 approached 120 ppb, CO was in excess of 260 ppb, and OA more than 25 µg m-3. Except for CO, these values are the highest observed from the G-1 during CARES.

The link between temperature and photochemistry is complex, involving changes in PAN lifetime, anthropogenic and biogenic emission rates, photolysis rates, and water vapor (Sillman and Samson, 1995). In addition, high temperatures are often accompanied by some measure of poor ventilation, promoting the buildup of primary pollutants such as observed on 28 June. Daily maximum ozone, obtained from monitoring networks, show a strong positive correlation with temperature at most locations, including Sacramento (Steiner et al., 2006, 2010). Over most of the temperature range of interest, emission rates for isoprene and terpenes increase exponentially with leaf temperature (Guenther et al., 1993), consistent with the temperature trend for daily maximum isoprene mixing ratio observed in Sacramento (Steiner et al., 2010) and at Blodgett Forrest (Dreyfus et al., 2002). All of the factors necessary to create simultaneous high concentrations of ΔOA, ΔCO, Δisoprene, ΔCH3OH, and ΔO3 are in place on 28 June. The driving meteorological factors, however, do not necessarily affect OA and OA precursors equally, with the result that trends in the parsed data can exist without strong correlations between ΔOA and ΔBio.

According to our definition of plume perturbations (Δs) approximately 60 % of peak OA in the Sacramento plume is background. This value is determined by a 56 transect average of the ratio OA(10th percentile) / OA (90th percentile). A high contribution from background is consistent with transport simulations reported by Fast et al. (2012). As background SOA would be aged, its presence is consistent with the O to C ratios reported by Shilling et al. (2013). Correlations between background OA and its precursors, in particular the strong correlation with background CH3OH and weak correlation with background CO, suggest that background OA is primarily biogenic. Background aerosol will foster the formation of OA in the Sacramento plume by more than doubling the aerosol volume potentially available for gas to particle partitioning. The increase in partitioning realized will depend on the volatility of gas phase precursors and aerosol viscosity and phase (e.g., Zaveri et al., 2014; Madronich et al., 2015). In so far as isoprene and its oxidation products are expected to dominant VOC reactivity, these compounds will affect the processing of anthropogenic carbon and may be seen as contributing to A–B interactions.

The circumstance that ∼ 60 % of peak OA is background, and appears to be primarily biogenic, has the consequence that the fraction of non-fossil carbon in the Sacramento plume will be significantly greater than half, especially since plume OA is partially biogenic and approximately 20–30 % of urban anthropogenic emissions are estimated to be from non-fossil sources, such as cooking, trash burning, and biofuel use (Hodzic et al., 2011; Zotter et al., 2014). There were no 14C measurements to compare with our correlation-based predictions, but given observations in areas with lower biogenic emission rates (Marley et al., 2009), and given observations made at Blodgett Forest (Worton et al., 2011), it would be surprising if non-fossil carbon did not predominate.

We have used aircraft observations obtained during the CARES field campaign to infer whether background OA and OA produced in the Sacramento plume are primarily anthropogenic or biogenic, whether there is enhanced OA formation caused by interactions between anthropogenic and biogenic compounds, and if observations are consistent with 14C measurements from other locations. Linear and bilinear correlations for 56 individual transects show how well OA follows explanatory variables over small spatial scales, whilst correlations amongst transects yield information on how ΔOA (plume perturbation) varies due to changes in location and day-to-day variations in meteorological conditions. Following Setyan et al. (2012) and Shilling et al. (2013), transects are split into subsets in order to determine whether high OA concentrations are uniquely associated with high mixing ratios of anthropogenic and biogenic tracers.

Explanatory variables were CO as a tracer of anthropogenic OA precursors and isoprene, MVK+MACR, and CH3OH as tracers of biogenic emissions. In addition O3 was used as a surrogate for photochemical activity. Tracers should ideally have lifetimes comparable to the timescales over which OA is formed. CO is long lived. CH3OH is used here (to the best of our knowledge, for the first time) as an analogous long-lived biogenic tracer with an atmospheric lifetime of ∼ 10 days. In contrast, isoprene and MVK+MACR have lifetimes ranging from less than an hour to a few hours.

We have found that

Linear and bilinear regression models indicate that most of the explained variance in OA is explained by CO rather than a biogenic compound. This is true for within transect and amongst transect correlations. Adding a biogenic tracer to OA vs. CO to yield a bilinear model results in only a modest increase in explained variance.

By its very nature conclusions based on correlations are inferential. It would be highly desirable to test our results against a high-resolution chemical transport model. Model calculations are essential for comparisons between CARES and other regions; in particular, the well-studied eastern USA in which aerosols have higher relative and absolute amounts of inorganic constituents and there is a greater abundance of liquid water, perhaps enabling A–B mechanisms to a greater extent than observed in the CARES region.