.Origin of tropospheric NOx over subarctic eastern Canada

. The origin of NO(cid:127) in the summertime troposphere over subarctic eastern Canada is investigated by photochemical modeling of aircraft and ground-based measurements from the Arctic Boundary Layer Expedition (ABLE 3B). It is found that decomposition of peroxyacetyl nitrate (PAN) can account for most of the NO(cid:127) observed betw(cid:127)n the surface and 6.2 km altitude (aircraft ceiling). Forest fires represent the principal source of PAN in the region, implying the same origin for NO(cid:127). There is, however, evidence for an unidentified source of NO(cid:127) in occasional air masses subsiding from the upper troposphere. Isoprene emissions from boreal forests maintain high NO(cid:127) concentrations in the continental boundary layer over eastern Canada by scavenging OH and NO3, thus slowing down conversion of NO(cid:127) to HNO3, both in the daytime and at night. This effect is partly compensated by the production of CH3CO3 radicals during isoprene oxidation, which slows down the decomposition of PAN subsiding from the fr(cid:127) troposphere. The peroxy radical concentrations estimated from concurrent measurements of NO and NO2 concentrations during ABLE 3B are consistent with values computed from our photochemical model below 4 km, but model values are low at higher altitudes. The discrepancy may reflect either a missing radical source in the model or interferences in the NO2 measurement.

posphere, yielding an ensemble of snapshots of local photochemistry along the aircraft flight tracks from which regional statistics for the NO,, budget can be obtained. The 1-D model calculations are used for the continental boundary layer (CBL) over the boreal woodland and account for diel variations in vertical mixing and biogenic isoprene emissions. The woodlands of eastern Canada are large sources of isoprene [Blake et al., 1994]. The ABLE 3B data provide a rare chance to study the chemistry of isoprene under the low NO,, conditions which are characteristic of the CBL over remote regions.
The NO,, budgets in the free troposphere and in the continental boundary layer are presented in sections 3 and 4, respectively. Conclusions are in section 5. The appendix provides discussion on the feasibility of using the concurrent measurements of NO and NO2 in ABLE 3B to calculate the concentration of peroxy radicals and test the accuracy of photochemical models. Figure 1. Air masses of various chemical compositions were encountered by the aircraft, reflecting influences from forest and tundra fires, industrial and urban pollution, stratospheric intrusions, and tropical outflow ( Table 1). The ABLE 3B mission design placed particular emphasis on sampling biomass fire plumes. Air masses influenced by biomass fire emissions, as diagnosed by CO concentrations greater than 120 ppb [Talbot et al., 1994], accounted for about 30% of the aircraft observations in the free troposphere. Lidar measurements of aerosol concentrations in the 2 to 6 km column during the expedition suggest that air influenced by biomass buming occupied, on average, 13% of the free troposphere, background air, 42%, stratospherically influenced air, 35%, and other types, 12% [Browell et al., 1994].

A map of the ABLE 3B region is shown in
We use a 0-D model to calculate the concentrations of radicals and other secondary species at chemical steady state in the free troposphere. The calculations are constrained with the ensemble of measurements taken aboard the aircraft including temperature, dew point, pressure, UV radiation fluxes (zenith and nadir), and concentrations of NO, PAN, HNO3, 03, CO, acetone, C•_? alkanes, C2-3 alkenes, benzene, and toluene. Measurements of NO are more reliable than those of NO2 owing to possible interference in the NO2 measurement (see appendix). The calculations are conducted for 3-min averaging intervals in the aircraft observations, representing the time resolution of the NO measurement. There are 165 intervals in the free troposphere (2.5-6.2 km) where concurrent measurements are available for all the above input variables except acetone. The data are sparse for acetone; missing data are filled based on the correlation between acetone and CO (Figure 2) (see also Singh et al. [ 1994b]).
The 0-D model is inadequate in the continental boundary layer (CBL) (below 2.5 km), where the concentrations of radicals are sensitive to the abundance of biogenic isoprene and its oxidation products [Jacob and Wofsy, 1990]. The lifetime of isoprene is only a few hours, while the carbonyls produced successively in the isoprene oxidation chain have lifetimes of a few hours to a few days. One cannot assume that the isoprene oxidation products are in local chemical steady state with the isoprene concentrations measured aboard the aircraft. We use therefore a time-dependent, 1-D model for the CBL, following Trainer et al. [1987,1991] and Wofsy [1988, 1990], and apply this model to simulate observations over the Schefferville tower site on August 7 when detailed measurements from the tower and from an aircraft spiral over the tower are available. The tower site is 0.5 km above sea level. The CBL in our 1-D model extends to 2.25 km above ground level, representing the aftmoon maximum of mixed layer depth as measured locally from rawinsondes [Fitzjarrald and Moore, 1994]. Photochemical calculations are conducted at seven grid points (0.01, 0.1, 0.2, 0.5, 1.0, 1.5, and 2•0 km above the canopy). Vertical transport is simulated with an eddy diffusion parameterization based on local measurements of the mixed layer depth zi and of the fluxes of momentum and sensible heat [Lamb et al., 1975]. The eddy diffusion coefficient between zi and 2.25 km is adjusted to reproduce the observed vertical profiles of O3 mixing ratio, resulting in a ventilation lifetime of 4.5 days for the CBL.  Air mass types are defined following Talbot et al. [ 1994] as 1, Regional background; 2, biomass burning influence; 3, tropical outflow; and 4, stratospheric influence. The mean characteristics are computed from aircraft observations for the 165, 3-min intervals used in our photochemical modeling calculations. Volume mixing ratios are in parts per trillion (ppt) except for CO, O3 in parts per billion. Unlisted species were generally near or below their detection limits; 10 ppt propene, 5 ppt toluene, and 2 ppt for >C4 alkanes.

in the Free Troposphere
Mean production and loss rates of NO,, in the free troposphere during daytime are shown in Figure 3 for individual air mass types. The reaction rates calculated for each 3-min interval were first averaged according to time of day (2-hour bins, with morning/afternoon folding to overcome a lack of measurements in the early morning) and then averaged over daytime hours (0600-1800 LT). Nighttime chemistry is assumed negligible in the free troposphere for reasons discussed above. We find that sources and sinks of NO,, are in close balance in background air (   Long-range transport of midlatitude pollution was found to be of secondary importance. Nitrogen oxides emitted from biomass burning are efficiently converted to PAN in the fire plumes, because of the abundance of reactive hydrocarbons [Jacob et al., 1992], and additional PAN is formed on the regional scale following photolysis of pyrogenic acetone.  PAN (Figure 3). The resulting net source for NO,` in Figure 3 is 10 ppt h -1, but the mean concentration of NO,` in these air masses was only 40-50 parts per trillion (hereafter, ppt). Reconciling the model NO,` budgets with the observed NO,` concentrations would require rapid dilution of the air masses with the regional background. However, an underestimate of PAN formation is also possible, because many primary and secondary species, particulaxly oxygenated hydrocarbons, may be present in type 2 air masses but not included in the model. Such species could decompose to yield CH3CO 3 radical and promote PAN formation.

Air masses sensibly influenced by biomass burning in
Tropical air masses originating from the Pacific Ocean (Table  1,

NOx in the Continental Boundary Layer
We now turn to an analysis of the origin of NO,` in the continental boundary layer (CBL) over the boreal woodland at Schefferville using the 1-D model for August 7 described in section 2. We address the following questions. Calvert, 1990], but reactions with HO2 may be rapid [Atkinson, 1990]. The reaction products axe assumed to be organic peroxides (ROOH) which may photolyze, react with OH, or be removed by deposition. The latter two sinks would represent real loss of radicals from the atmosphere.

(1) Does decomposition of PAN subsiding from the free troposphere account for NO,` in the CBL? (2) How does vegetative emission of isoprene affect NO,` and PAN? (3) How sensitive is the NO,` budget in the CBL to the RO2 + HO2 reactions? The last question is motivated by the lack of kinetic data for the reactions of organic peroxy radicals (RO2) arising from photochemical oxidation of isoprene. Reactions of these peroxy radicals with each other are probably slow [Madrordch and
To address the above questions, we present here results from three simulations (1) a standard run including isoprene emission, RO2 + HO2 reactions with a rate constant k = 3.4x10 -]3 e 8øørr cm 3 molecule -1 s -1 taken from Atkinson [ 1990], and ROOH photolysis at a rate twice that for CH3OOH; (2) a sensitivity run with zero isoprene emission; and (3) another sensitivity run with isoprene emission but without the RO2 + HO2 reactions. Our definition of RO2 here does not include the CH30 2 radicals, for which kinetic measurements of the reaction with HO2 axe available. Figure 5 shows aircraft measurements of ambient temperature, absolute humidity, and mixing ratios of CO over the tower site on the afternoon of August 7. The air mass below 2 km originated from the Hudson Bay region 5 days prior to aircraft measurements and was not modified by rain or combustion emissions during the transit period [Shipham et al., 1994]. The composition of that air mass is typical of the regional background (Table 1). Tropical influence from the Pacific is apparent above 2 km and is manifested in Figure 5 by the low mixing ratios of CO (<80 ppb). This tropical influence was transitory [Shipham et al., 1994]; therefore we assume that the boundary layer had been in contact with a free troposphere of background composition in the few days before being overriden by the tropical air. This assumption dictates our choice of upper boundary conditions (section 2). Figure 6 shows the comparisons of model mixing ratios of isoprene, 03, NO, NO2, and PAN with observations. The model simulates NO, NO2, and PAN within the measurement uncertainties. The vertical distribution of isoprene is controlled by turbulent mixing and by OH oxidation in the boundary layer (the isoprene emission flux was adjusted to match the observed isoprene concentxations.) Concentrations of 03 are controlled mainly by transport from the free troposphere and deposition. The net photochemical production of 03 in the CBL is small, about 20% of the flux from the free troposphere.
The sensitivity simulation without isoprene emission yields 03, NO, and NO2 within the range of measurements. However, the PAN mixing ratio in the CBL falls below the measurement by a factor of more than 2. In the standard simulation, oxidation of isoprene yields high concentrations of the CH3CO 3 radical. The sensitivity simulation without RO2 + HO2 reactions yields mixing ratios of NO a factor of 2 lower than the standard simulation and significantly lower than the measurements. The afternoon mixing ratio of total peroxy radicals (ZRO2) in this simulation is higher than 100 ppt (Figure 7). Figure 8 shows the simulated NOx and PAN budgets in the boundary layer for the standard simulation. The formation of HNO3 is nearly balanced by decomposition of PAN subsiding from the free troposphere; net exchange of NO• between the CBL and the free troposphere is small. The production of HNO3 occurs mostly in the daytime. At night, isoprene reacts rapidly with NO3, producing isoprene nitrate radicals [Atkinson et al., 1988;Dlugokencky and Howard, 1989] which release NO2 upon further reactions after sunrise [Paulson and Seinfeld, 1992]; this effectively prevents the nighttime formation of N2Os and hence the loss of NO• to HNO3 via N2Os hydrolysis. Thus isoprene suppresses formarion of HNO3 by depleting OH during daytime (Figure 7) and by reacting with NO3 at night. As a result, the lifetime of NO• in the CBL is considerably longer with than without isoprene, 2.9 versus 1.2 days. Loss of NO• in the former case could be more rapid than computed here if isoprene nitrate radicals react on aerosols to yield HNO3.

Summary
The origin of NOx in the summertime subarctic troposphere over eastern Canada was studied by modeling aircraft and ground observations from the ABLE 3B expedition. It is found that decomposition of PAN can account fully for the observed NO• concentrations in the free troposphere below 6 km except in occasional air masses subsiding from the upper troposphere. There is evidence that other organic nitrates are present in these air masses, and their decomposition may provide significant sources for NO•. We speculate that HNO3 may react with CH20 in concentrated sulfuric acid aerosols to produce HOCH2ONO2 and CH2(ONO2)2 and that these nitrates would photolyze on a timescale of weeks to release NO•.
Decomposition of PAN subsiding from aloft appears to provide the primary source of NO• in the continental boundary layer (CBL) over eastern Canada woodlands. The NOx budget in the CBL is strongly influenced by isoprene emission from vegetation. On the one hand, isoprene increases the lifetime of NO• in the CBL by scavenging OH in the daytime and NO3 at night (the isoprene nitrate radicals produced at night are assumed to return   [Madronich, 1987;. Hence a typical measurement error for the Y-RO2 concentration would be at least 60%. Much larger errors for Y-RO2 are expected when measured mixing ratios of NO fall below 10 ppt and NO 2 below 30 ppt, respectively. Therefore it is not instructive to compare model YRO2 with the implied Y-RO2 at 3min averaging intervals. We reduce the uncertainty in the implied Y-RO2 by averaging over a large number of intervals. Figure A1 shows the implied Y. RO2 mixing ratios as a function of NO averaged over all intervals for which data are available for constraining the model. Values of ZRO2 increase with decreasing NO and exceed 200 ppt for NO less than 5 ppt. Concentrations of ZRO2 of a few 100 ppt would result in rapid 03 production (of the order of 1 ppb h -1 ), which seems inconsistent with the concentrations of 03 typically observed in the free troposphere. Further, this high level of ZRO 2 could not be maintained by known sources of odd hydrogen radicals. On the other hand, possible unknown errors in the NO measurements have been estimated to be at or below 3.5 ppt [Sandholm et al., 1994], so the implied Y-RO2 at a few ppt NO may not be reliable. Figure A2 compares simulated and implied mixing ratios of Y. RO 2 for the subset of data with NO above 10 ppt (i.e., x5 measurement noise). The implied YRO2 mixing ratios show large variances and appear to increase with altitude. The model underpredicts the implied Y. RO 2 by a factor of 2-3 above 4 km; the discrepancy is less at lower altitudes.
The discrepancy could conceivably reflect a large missing source of odd hydrogen radicals in the model at high altitudes. Alternatively, it is possible that interferences in NO 2 measurements, increasing with altitude, may be responsible. In particular, HNO4 could heterogeneously decompose in the sampling tubing [Ridley et al., 1988;Sandholm et al., 1992Sandholm et al., , 1994]. Our model predicts significant levels of HNOn in the free troposphere, with HNOn/NO2 concentration ratios increasing from 0.2 at 3 km to 0.9 at 6 km, on the average. To illustrate the potential effect of an HNOn interference, we recalculated l•RO2 mixing ratios using adjusted NO2 (which equals observed NO2 minus simulated HNOn). The agreement between simulated and implied mixing ratios of l•RO2 is somewhat improved ( Figure A2). Concentrations of PAN also increased with altitude and were many times larger than those of NO2 [Singh et al., 1994a]. If a few percent of PAN decomposed in the sampling tubing, the interference on the NO2 measurement would be significant. For the inslxument configuration used in ABLE 3B the wall reaction efficiency would need to be approximately 1 x 10 -4 for a 50% conversion efficiency of HNO 4 --> NO2 + HO2 and 1 x 10 -5 for a 5% conversion efficiency of PAN.
Measurements of NO are more reliable and were therefore chosen as constraint in the 0-D model calculations.
Since the disagreement between implied and modeled l•RO2 concentrations is less at lower altitudes, we also use the NO2 measurements as model constraints in the boundary layer calculations.