Influence of urban pollution on the production of organic particulate matter from isoprene epoxydiols in central Amazonia

Abstract. The atmospheric chemistry of isoprene contributes to the production of a substantial mass fraction of the particulate matter (PM) over tropical forests. Isoprene epoxydiols (IEPOX) produced in the gas phase by the oxidation of isoprene under HO2-dominant conditions are subsequently taken up by particles, thereby leading to production of secondary organic PM. The present study investigates possible perturbations to this pathway by urban pollution. The measurement site in central Amazonia was located 4 to 6 h downwind of Manaus, Brazil. Measurements took place from February through March 2014 of the wet season, as part of the GoAmazon2014/5 experiment. Mass spectra of organic PM collected with an Aerodyne Aerosol Mass Spectrometer were analyzed by positive-matrix factorization. One resolved statistical factor ( IEPOX-SOA factor ) was associated with PM production by the IEPOX pathway. The IEPOX-SOA factor loadings correlated with independently measured mass concentrations of tracers of IEPOX-derived PM, namely C5-alkene triols and 2-methyltetrols (R = 0. 96 and 0.78, respectively). The factor loading, as well as the ratio f of the loading to organic PM mass concentration, decreased under polluted compared to background conditions. For an increase in NOy concentration from 0.5 to 2 ppb, the factor loading and f decreased by two to three fold. Overall, sulfate concentration explained 37 % of the variability in the factor loading. After segregation of factor loading into subsets based on NOy concentration, the sulfate concentration explained up to 75 % of the variability. Considering both factors, the data sets show that the suppressing effects of increased NO concentrations dominated over the enhancing effects of higher sulfate concentrations. The pollution from Manaus elevated NOy concentrations more significantly than sulfate concentrations relative to background conditions. In this light, increased emissions of nitrogen oxides, as anticipated for some scenarios of Amazonian economic development, could significantly alter pathways of PM production that presently prevail over the tropical forest, implying changes to air quality and regional climate.

min, constituting half of the total acquired data set. 33 Due to expected low mass concentrations and to secondary need for the low-sensitivity 34 high-resolution data, the AMS was operated for most of the time in medium-resolution V-mode 35 (Δm/m = 2200 at m/z 44), which was used for mass quantification. High-resolution W-mode 36 (Δm/m = 4000 at m/z 44) was acquired for one of every six days. These data were used to aid 37 choice of ions to fit. When only V-mode data were acquired, the instrument was operated in 38 "mass spectrum" sub-mode for 180 s and in "particle-time-of-flight" sub-mode for 60 s. When 39 W-mode data were also collected, the "W-mass-spectrum" sub-mode was operated for 60 s, the 40 "V-mass-spectrum" sub-mode ran for 120 s, and "V-particle-time-of-flight" sub-mode ran for 60 41 s. Both ways of operation corresponded to 4 min for each cycle, allowing for synchronization to 42 the valve switching system. 43 The AMS sensitivity and the ammonium relative ionization efficiency (RIE) were 44 calibrated every five days using dried ammonium nitrate particles having a mobility diameter of 45 400 nm. An interpolated curve of the obtained values, corresponding to RIE = 4.3 ± 0.2, was 46 total residuals when going from the five-factor to the six-factor solution. The six-factor solution 122 also offered more meaningful factor profiles ("mass spectra") (de Sá, in preparation). Panel c 123 shows Q/Q expected as a function of the rotational ambiguity parameter f peak (Ulbrich et al., 2009) 124 for the six-factor solution. A plausible range for f peak was determined according to the best 125 practice of limiting Q/Q expected to a value that does not exceed 0.1% of the minimum value 126 (occurring at f peak = 0). The default value of f peak = 0 was chosen for the final six-factor solution, 127 since no significant improvements in the external validation of the factors were observed. 128 Moreover, the IEPOX-SOA factor resolved in the six-factor solution had a very robust time trend 129 across a range of rotations in the solution ( Figure S2d). As f peak varied, the correlation of factor 130 loading with C5-alkene triols concentrations remained approximately constant, even as some 131 features of the factor profile changed significantly, such as the relative signals of C 5 H 6 O + and 132 CO 2 + , respectively f(C 5 H 6 O) and f(CO 2 ), as well as the magnitude of factor loadings and 133 consequently the ratio f. 134 The loading of the IEPOX-SOA factor may be an overestimate or an underestimate of the 135 atmospheric concentration of the IEPOX-derived PM (brown dashed lines in Figure 1). In 136 respect to overestimate, the AMS mass spectrum observed in laboratory studies for the uptake of 137 IEPOX by acidic sulfate particles is statistically equal to that obtained for the uptake of isoprene 138 photo-oxidation products, yet IEPOX accounted for only half of those products . 139 The implication is that the uptake of non-IEPOX species can lead to a similar AMS spectrum. 140 Pathways of PM production from condensation of multifunctional hydroperoxides lead to a 141 distinct mass spectrum from IEPOX pathways, and are not expected to be highly active under the 142 acidic particle conditions of these experiments , leaving a large mass fraction 143 of produced PM unexplained. The combination of AMS vaporization at 600 °C and ionization by 144 electron impact at 70 eV may convert IEPOX-derived and non-IEPOX-derived molecules into 145 similar groups of ions, which then give rise to a similar mass spectrum. The SV-TAG, which 146 uses desorption temperatures up to 310 °C and thus can also induce thermal decomposition of 147 some molecules, might also result in an in-common analyte (i.e., tracer) between IEPOX-derived 148 and non-IEPOX-derived molecules, thereby precluding a constraint on any possible overestimate 149 by the AMS factor (Isaacman-VanWertz et al., 2016;Lopez-Hilfiker et al., 2016). For these 150 reasons, the loading of the IEPOX-SOA factor might overestimate IEPOX-derived PM 151 concentrations by accounting for other isoprene oxidation products that are not produced through 152 the IEPOX intermediate. 153 In respect to underestimate, the loading of the IEPOX-SOA factor may not capture the 154 entire particle-phase carbon footprint that originated from IEPOX uptake. Extensive atmospheric 155 processing, such as reactions with hydroxyl radicals or photolysis, can partly eliminate the initial 156 products of IEPOX uptake (Kroll et al., 2009;Bateman et al., 2011;Epstein et al., 2014;Hu et 157 al., 2015;Hu et al., 2016). The IEPOX-originated carbon can still be inside the particle, yet it no 158 longer contributes to the loading of the IEPOX-SOA factor because of an altered mass spectrum 159 for some molecules. Atmospheric reactions gradually homogenize particle composition and 160 properties, and the AMS spectra can become more uniform . Specifically, 161 the ratio of signal intensity at m/z 44 to that at m/z 43 increases, and the relative intensity of m/z 162 82 decreases (Ng et al., 2011;Hu et al., 2015). This modified organic material, which originally 163 entered the particle phase through IEPOX uptake, may then contribute to the loading of PMF 164 factors other than IEPOX-SOA, such as the oxidized organic factors broadly labeled as "OOA" 165 (Zhang et al., 2005). For these several reasons, the IEPOX-SOA factor loading might be an 166 underestimate of IEPOX-derived PM concentrations. 167

S3. Comparison of background and polluted cases 168
The presence of a pollution plume at T3 is indicated by a combination of several external 169 measured variables, including particle number, ozone, and NO y concentrations. The definition of 170 background and polluted cases aimed at selecting afternoons that were associated with extreme 171 values of those variables. Conditions entailing concentration of ozone at around 10 ppb or less, 172 particle number concentration of less than 500 cm -3 , and NO y concentrations of less than 1 ppb 173 were collectively a strong indicative of a background case. Conditions including ozone 174 concentrations upward of 30 ppb, particle number concentration above 2000 cm -3 , and NO y 175 concentrations around 1.5 ppb or above indicated pollution. Measurements of these variables 176 onboard the G-1 aircraft confirmed what the ground measurements suggested on the several days 177 that the G-1 flew overhead. 178 Examples of these supporting data are illustrated in Figure  For the polluted cases shown in Figure S4, NO y concentrations were variable between 202 and within cases, ranging from 1 ppb up to 7 ppb. Larger sulfate mass concentrations (from left 203 to right) were associated with larger absolute and relative factor loadings, analogous to what was 204 observed for background cases. A comparison between March 3 and 13 (i.e., background and 205 polluted; Figure 4) shows that for similar sulfate concentrations but higher NO y levels March 13 206 had considerably lower factor loadings. The case of February 9 (polluted) illustrates that the 207 factor loadings did not exceed 0.4 µg m -3 and f did not exceed 0.2 even at rarely high sulfate 208 concentrations in the wet season, reaching 0.9 µg m -3 . This finding is attributed to the high NO 209 concentrations, as implied by NO y concentrations of 3 ppb and greater, that suppressed IEPOX 210 production. 211 These cases illustrate the possible wide range of observed sulfate mass concentrations 212 under background conditions, in great part overlapping with typical values of polluted 213 conditions. The cases also demonstrate that the trend in observed IEPOX-SOA factor loadings 214 and ratio f, both within each category (background or polluted conditions) and between them, can 215 be consistently explained by the roles that sulfate and NO exert on the production of IEPOX-216 derived PM. 217

S4. Sulfate and particle acidity estimates in the context of field studies 218
The underlying relative importance of direct compared to indirect roles of sulfate on the 219 formation of IEPOX-derived PM is not well understood. Sulfate can play a direct role as a 220 nucleophile in the reaction of formation of organosulfates from IEPOX (Surratt et al., 2007b;221 Nguyen et al., 2014). Organosulfates, however, are believed to constitute only a fraction of the 222 IEPOX-derived PM (Hu et al., 2015). Particle acidity, an indirect effect of sulfate, has been 223 shown to drive IEPOX-derived PM production in several lab studies (Surratt et al., 2007a;224 Kuwata et al., 2015;Lewandowski et al., 2015). Nevertheless, the acidity effect observed in 225 laboratories is not as clearly observed in field studies, wherein pH estimates have typically been 226 employed as a proxy for acidity (Budisulistiorini et al., 2013;Lin et al., 2013;Worton et al., 227 2013;Budisulistiorini et al., 2015;Xu et al., 2015). The present study corroborates those findings 228 and further argues that this apparent conflict can be reasoned by taking into account that both the 229 estimate and the end-use of pH may be problematic in the context of field studies. 230 some difficulties were imposed by data availability. Gas-phase measurements of NH 3 or HNO 3 232 were not available for performing "forward" mode calculations in thermodynamic models or gas-233 particle phase partitioning calculations, which have been suggested as the best method to predict 234 pH (Hennigan et al., 2015). Co-located independent measurements of ion concentrations (e.g., by 235 chromatograph) were not available to confirm the ion balance obtained by AMS measurements 236 ( Figure S7). 237 Bearing these caveats in mind, the analysis presented in the main text using sulfate as a 238 predictor for IEPOX-SOA was replicated here using pH in place of sulfate. Figure S8 is 239 analogous to Figure 6a. pH was estimated for IOP1 using AMS measurements of mass 240 concentrations of inorganic ions (sulfate, ammonium, nitrate, and chloride) and measurements of 241 RH and temperature. The E-AIM model II (Clegg et al., 1998) was employed. The final pH was 242 calculated taking into account both the inorganic water predicted by the model and the organic 243 water estimated from organic hygroscopicity org  values (Thalman, in preparation), in a similar 244 fashion as described by Guo et al. (2015). Figure S8 shows that pH, as calculated herein and for 245 the caveats herein, does not work well as a predictor for IEPOX-SOA factor loading. Figure S9  246 is analogous to Figure 7. Figure S9 shows that, although the overall dependencies on NO y are 247 similar to analysis of the data by sulfate, the separation by pH yields groups that have less 248 distinct trends and ranges among them. 249 In addition to difficulties associated with generating accurate estimates of pH, there is an 250 inconsistency between the timescales of estimated particle acidity and IEPOX-SOA factor 251 loadings that may preclude underlying correlations to emerge. The estimated pH makes use of 252 instantaneous RH and temperature values and is therefore an instantaneous estimate of pH. They 253 can change in timescales of 15 min or less giving mixing in the boundary layer, for example. On 254 the other hand, sulfate mass concentrations and IEPOX-SOA factor loadings are variables 255 representative of processes of longer time scales of hours and days. By not containing any 256 information on the particle acidity history in the past hours or days, the calculated instantaneous 257 pH may fail in capturing the true effect of acidity on the chemical formation of IEPOX-derived 258 PM. The RH cycling history of sulfate particles has been demonstrated to mediate the extent of 259 IEPOX-derived PM production (Wong et al., 2015). Moreover, sulfate, by being intrinsically 260 related to particle acidity and at the same time a species of congruent lifetime with secondary 261 organic material, may in fact be a better proxy to capture the history of particle acidity than 262 estimates of pH. For these different reasons, sulfate rather than pH is used in the analysis herein. 263 The understanding, however, is that sulfate represents effects beyond those of the direct chemical 264 role of sulfate. In analogy to pH, this discussion also extends to the use of sulfate rather than 265 instantaneous particle water content as the predictor of IEPOX-SOA factor loadings. 266

S5. Five subsets of data based on NO y concentrations 267
The data subsets and fits shown in Figure 6 are shown in separate panels in Figure S5. 268 Once a trend of decreasing fit slope with increasing NO y concentration was identified, the 269 number of data subsets was defined as the minimum necessary to have subsets of at least 100 270 data points (of a total of 888) to allow for robust statistics and that were also cohesive (as 271 measured by R 2 ) and non-redundant (i.e., of different fit lines). 272 The coloring by date in Figure S5 shows that there is no apparent correlation between 273 levels of NO y concentration or goodness of fit with different time periods within IOP1. 274 Meteorological variables such as solar radiation, temperature, and RH (not shown) are also not 275 able to delineate any clear pattern in the data, either within or among groups. 276 as reflected in the relationship between IEPOX-derived PM and NO y concentrations. Moreover, 279 sulfate and NO y concentrations were not independent variables. For these reasons, a linear 280 bivariate analysis is not appropriate, and a subset analysis was pursued instead. 281

S6. Details and assumptions of the model 282
The solution to the differential Equation 1 is as follows: 283 for which the subscript 0 indicates initial (background) conditions, i.e., immediately before the 285 airmass passes over Manaus. For the transport from Manaus to T3, t = τ tr , and the variable M 286 represents the IEPOX-derived PM mass concentrations at T3. 287 The zero-order production rate coefficient k P and the first-order loss rate coefficient k L 288 are lumped parameters representative of several production and loss processes, respectively. A 289 first assumption is that they are constant over the course of four hours. In hand with that 290 assumption, a constant boundary layer height throughout the integration time is assumed. The τ 291 values represent the time required under afternoon conditions to significantly affect the IEPOX-292 derived PM mass concentration by the corresponding processes. For afternoon time periods, 293 observations show that dM/dt > 0 over tropical forests in the absence of pollution (Chen et al., 294 2009;Chen et al., 2015).The parameter τ P , corresponding to a first order process, therefore 295 represents an instantaneous quantity in the transient system. For simplicity of presentation, τ P is 296 defined in reference to M 0 , although defining it in relation to M bg or M pol does not alter the main 297 conclusions presented herein. 298 be lost by three main mechanisms: heterogeneous oxidation against OH, condensed-phase 300 reactions, and dry deposition. The lifetime of IEPOX-derived PM against heterogeneous 301 oxidation is estimated at around two weeks for an OH concentration of 10 6 molecules cm -3 (Hu et 302 al., 2016). Lifetime of IEPOX products against particle-phase reactions has not yet been reported 303 but is expected to be at least several days. A value of one week is used here. The lifetime of PM 304 against deposition is on the order of a week. The overall loss process is then approximated in the 305 model as the sum of these three processes, which leads to an estimate of overall loss rate 306 coefficient k L = 0.015 h -1 (overall characteristic time of 2.8 days). 307 In respect to wet deposition along the track from Manaus to the T3 site, strong convection 308 imports background regional air, and for this reason strong wet deposition is mathematically 309 equivalent in the model developed herein to a trajectory that does not pass over Manaus, i.e., 310 background conditions. Weak wet deposition represents a mixing of polluted and background air 311 masses, giving rise to intermediate NO y concentrations. Entrainment on the plume edges as well 312 as with the free troposphere is mathematically similar to wet deposition in the model framework. 313 Thus entrainment and wet deposition, without directly contributing to k L , are indirectly 314 incorporated in the developed model based on their effects on NO y concentration. 315 In terms of production processes, represented by the rate coefficient k P , IEPOX-derived 316 PM is produced by multigenerational chemistry of isoprene photooxidation and reactive uptake. 317 Model Case 1 investigated the sensitivity of pollution enhancement ratio and absolute mass 318 concentration of IEPOX-derived PM to its production rate coefficient. After constrained by 319 observations, the estimated interval for k P was [0.07, 0.13] μg m -3 h -1 . In addition, values of k P > 320 0.2 μg m -3 h -1 are unlikely given the rare observation (<1%) of M bg > 1μg m -3 . 321 material from diameter growth rates. Firstly, a relative production rate of 1:3 for IEPOX-derived 323 PM to total organic PM is assumed based on the following. IEPOX-derived PM is estimated to 324 contribute 34% on average to total organic PM in central Amazonia under background 325 conditions . For assumptions of equal first order loss rate coefficients for all 326 organic material, a mass concentration ratio of 1:3 for IEPOX-derived PM to total organic 327 material implies a ratio of 1:3 in their production rate coefficients k P . As a consequence, the 328 estimated range for total organic material is [0.21, 0.39] μg m -3 h -1 . For the estimate of organic 329 material production based on growth rates, an average growth rate of 10 nm h -1 (Kulmala et al., 330 2004) is assumed. Further assumptions are a range of particle number concentration of 500 to 331 1000 cm -3 and of initial diameter of 20 to 100 nm, typical of background conditions in the 332 Amazon. The obtained range of organic material production from growth rate estimates is 333 therefore 0.02 to 0.32 μg m -3 h -1 , which is comparable to the range constrained by the model. 334 In terms of the influence of Manaus plume on the production and loss processes, the 335 following assumptions were made. The acceleration of the oxidant cycle in the plume implies 336 that α L > 1. Under plume conditions, OH concentrations observed at the T3 site increased by a 337 factor of three compared to background conditions (Martin et al., 2016a;Kim, in preparation). A 338 proportional increase in the loss rate of IEPOX-derived PM by OH heterogeneous chemistry in 339 the plume is expected. While the OH loss mechanism (of characteristic time of two weeks) is 340 accelerated by three fold in the plume, dry deposition and condensed phase reactions (both of 341 assumed characteristic times of a week) are held constant. As a result, the overall loss rate is 342 enhanced by two fold, i.e., α L = 2 is assumed. 343 NO in the pollution plume as well as the faster consumption of intermediate gas-phase species by 345 the enhanced OH and O 3 concentrations implies α P < 1. The assumption is that the production of 346 IEPOX almost halts in the plume, and α P = 0.1. This assumption is supported by measured gas-347 phase concentrations of ISOPOOH at the T3 site, which dropped by 90% when NO y 348 concentrations increased from 0.5 ppb to 2 ppb (Liu et al., 2016). This observation and the 349 associated model assumption are a reflection of the lifetimes of the chemical species discussed: 350 in contrast to the abovementioned lifetimes on order of a week for organic particle material 351 against loss processes, the lifetimes of the gaseous species are significantly shorter. Isoprene and 352 ISOPOOH have a lifetime on the order of a few hours (Eddingsaas et al., 2010;St. Clair et al., 353 2015), and IEPOX has a lifetime of a few hours to a day for an OH concentration of 10 6 354 molecules cm -3 (Jacobs et al., 2013;Bates et al., 2014). 355 Figure S1. Profile of the IEPOX-SOA factor resolved by PMF analysis of the time series of AMS organic mass spectra collected in the wet season of 2014 (IOP1) at the T3 site (red), and in the wet season of 2008 (green) as part of AMAZE-08 experiment at the T0t site . Also plotted is the mass spectrum of secondary organic material produced in the Harvard Environmental Chamber from β-IEPOX photooxidation onto acidic ammonium sulfate seed particles under HO 2 -dominant conditions and RH < 5% (blue) . Pearson correlation coefficients R between the PMF factor of this study and the other spectra were: R = 0.99 for the AMAZE-08 PMF factor, R = 0.81 for the chamber spectrum with all ions included, and R = 0.95 for the chamber spectrum with m/z 44 and 28 excluded. Dependence of the quality-of-fit parameter Q/Q expected on the number of factors for f peak = 0, (c) Dependence of the quality-of-fit parameter Q/Q expected on f peak for number of factors = 6. The red line represents Q/Q expected that exceeds in 0.1% the minimum value at f peak = 0. This limit determines the range of plausible f peak values as indicated by the dashed black lines, (d) For the six-factor solution, dependence on the f peak parameter of the Pearson correlation coefficient R between the IEPOX-SOA factor loadings and independently measured C 5 -alkene triols (on the right vertical axis), the mean f ratio of IEPOX-SOA factor loading to total organic PM mass concentration, and the relative intensities f(CO 2 ) and f(C 5 H 6 O)of the ions CO 2 + and C 5 H 6 O + , respectively (on the left vertical axis).   . Scatter plots of sulfate mass concentration and IEPOX-SOA factor loading for local afternoon (12:00-16:00 local time; 16:00-20:00 UTC) for five different ranges of NO y concentrations. Panels a-e correspond to groups labeled 1-5 according to Table 1. Table 1 presents the parameters of the six least-squares linear fits represented by the lines in the figure. Data is colored by date.   Figure S8. Scatter plot of estimated pH and IEPOX-SOA factor loading for local afternoon (12:00-16:00 local time; 16:00-20:00 UTC). The data sets were collected into five subsets, colored and labeled 1 to 5, based on NO y concentration (analogous to analysis shown in Figure  6a). Figure S9. Dependence on NO y concentration of (a) IEPOX-SOA factor loading, (b) organic mass concentration, and (c) the ratio f of the IEPOX-SOA factor loading to the organic PM concentration. Data are segregated by low (< 2.2) and high (> 2.8) pH and grouped into five levels of NO y concentration (Figure 7). Squares represent medians of each group. Interquartile ranges are represented by whiskers along the abscissa and shading along the ordinate. The plotted data sets were recorded during local afternoon (12:00-16:00 local time; 16:00-20:00 UTC).