Global Budget and Distribution of Peroxyacetyl Nitrate (PAN) for Present and Preindustrial Scenarios. International Journal of Earth & Environmental

A global 3-D chemistry and transport model, STOCHEM integrated with a detailed VOC oxidation scheme (CRI v2-R5) has been employed to study the important NO x reservoir compound, peroxyacetyl nitrate (PAN). Globally, PAN is produced entirely by the reaction of acetyl peroxy radicals (CH 3 CO 3 ) with NO 2 and up to 2.0 ppb of PAN is found over the polluted regions of North America during June-July-August for the present scenario. The imbalances between model and measurement data are noted, with STOCHEM-CRI overestimating PAN mixing ratios relative to the measurement data by +17 and +80 pptv for the lower and upper troposphere, respectively. The inclusion of additional HO x recycling mechanisms (e.g. related to isoprene oxidation) in STOCHEM-CRI causes a decrease in PAN in a present scenario by as much as 40% over sink regions and reduces the model-measurement disagreement by 90% for the lower troposphere and 40% for the upper troposphere. The lower NO x emissions and CH 3 CO 3 formation upon including HO x recycling in a preindustrial scenario led to a decrease in PAN formation by as much as 40%. The decrease in PAN formation results in less nitrogen being transported to remote regions which in turn leads to the greatest percentage change in O 3 concentration (9% decrease) in the equatorial regions.

PAN has no known direct emission sources and is therefore an excellent indicator of the photochemical processing of an air mass which plays a significant role in the atmospheric transport of reactive nitrogen on a regional and global scale. Quantification of the spatial distribution of PAN and its global budget can be a useful tool in determining the oxidative reactions involved in the formation of O 3 as well as other secondary air pollutants (e.g. nitric acid). In spite of its importance in the chemical processing of the troposphere, there are numerous measurements of PAN from different urban and remote areas [11,, but these measurements are sparse both in space and time. The measurements of the temporal variability of PAN are necessary to evaluate the global budget of O 3 , NO y , HO x , and the associated recycling of NO x . Thus, we present the global burden and the global distribution of PAN from the STOCHEM-CRI global chemistry transport model. We compare STOCHEM model results with a wide range of observations of PAN from the flight data set compiled by Emmons et al. [54], Horowitz et al. [55], and Fischer et al. [56] and twenty individual field campaign data set. The effects of additional HO x recycling mechanisms involving isomerisation of isoprene-derived peroxy radicals [57] and propagating channels for the reactions of HO 2 with acyl peroxy radicals, RCO 3 [58] in both present and preindustrial scenarios are shown in the study.

Model description
STOCHEM is a global 3-dimensional tropospheric chemistry transport model that adopts a Lagrangian approach splitting the troposphere into 50,000 constant mass air parcels which are advected by meteorological data from the UKMO Hadley Centre global general circulation model called the Unified Model (UM). The Lagrangian cells are based on a grid resolution of 1.25° longitude, 0.8333° latitude, and 12 unevenly spaced (with respect to altitude) vertical levels between the surface and an upper boundary of 100 hPa [59][60]. The resulting simulated concentrations are mapped onto an Eulerian

Special Issue: Climatic Changes Causes & Consequences
grid resolution of 5° by 5° with 9 equally vertically spaced pressure levels, each 100 hPa thick. A detailed description of the dispersion processes including the vertical coordinate, advection scheme used in STOCHEM can be found in Collins et al. [59] with updates described by Derwent et al. [61].
The STOCHEM-CRI model utilises the Common Representative Intermediates version 2 and reduction 5 chemical mechanism. The CRI v2-R5 was built using a series of five-day box model simulations on each species, on a compound-by-compound basis. The performance of the chemistry of these simulations was optimised using the Master Chemical Mechanism (MCM) with O 3 production being the primary criterion [62]. Simulations completed over a range of 32 VOC/NO x ratios by CRI v2-R5 have shown compelling agreement with MCM v3.1 for ozone and other radical and closed-shell species (including PAN), hence establishing it as an appropriate reference mechanism for the use in global chemistry transport models [63][64]. The details of the CRI v2-R5 mechanism is given by Jenkin et al. [65], Watson et al. [64], and Utembe et al. [66] with updates highlighted in Utembe et al. [60]. The emissions data employed in the base case STOCHEM model were adapted from the Precursor of O 3 and their Effects in the Troposphere (POET) inventory [67] for the year 1998. More details about the global emission data used in the STOCHEM can be found in Khan et al. [68].
The additional HO x recycling mechanisms involving isomerisation of isoprene-derived peroxy radicals [57,[69][70] and the propagating channels for the reactions of acyl peroxy radicals with HO 2 [58] have been introduced into the STOCHEM-CRI. In isoprene induced HO x recycling mechanism, the intermolecular rearrangements of isoprene derived alkyl peroxy radicals were sufficiently rapid to compete with biomolecular routes leading to two additional oxidation pathways of the alkyl peroxy radicals at low NO x levels. The two additional pathways involve 1,5-H-shift resulting in OH, HCHO and methyl vinyl ketone (MVK) or methacrolein (MACR) formation; and 1,6-H-shift leading to the formation of HO 2 and a hydroperoxyaldehyde (HPALD), which is removed predominantly by rapid photolysis to generate additional OH and other radical products.
Two experiments were conducted as the base case reference run referred to as 'Base' described in Utembe et al. [60] and a model including the isoprene HO x recycling and radical propagation in a simulation referred to as 'ISOP' . Two further experiments (referred to as 'B1800' and 'ISOP1800') were performed for a preindustrial scenario described in Khan et al. [71]. All simulations were conducted with meteorology from 1998 for a period of 24 months with the first 12 allowing the model to spin up. Analysis is performed on the subsequent 12 months of data.

The global burden of PAN for the present scenario and the HO x recycling impact on its burden
In the model, PAN is produced entirely from the reaction of acetyl peroxy radicals (CH 3 CO 3 ) with NO 2 . The formation of CH 3 CO 3 from a suite of VOC oxidation processes are summarized in Appendix A. The greater input of the organic materials and the greater complexity of the chemistry scheme in the CRI v2-R5 mechanism of the STOCHEM model [60], leads to a global burden of 398 Gg(N) PAN for the present scenario. The globally averaged tropospheric lifetimes of PAN is found to be 9.3 hours, the value being close to those reported by Singh [9], and Roberts et al. [42]. The HO x recycling mechanism employed in the STOCHEM-CRI gives consistent results with the box model simulations in Archibald et al. [69] and with the parameterized representations of Lelieveld et al. [72] and Pugh et al. [73]. The reduced total PAN global burden (346 Gg(N)) in the study are driven by the decreased production of CH 3 CO 3 (e.g. 11% less CH 3 CO 3 production in the ISOP scenario compared with the base case scenario) leading to the global decrease of PAN by 13% for the present scenario which given that the base simulation generally over-predicts PAN brings the model into closer agreement with measured values. However, the concentrations of CH 3 CO 3 molecules is a balance between faster production via VOC oxidation and faster removal via reaction with HO 2 , NO, NO 3 and partitioning into PAN species [74]. The inclusion of HO x recycling mechanisms leads to a slight decrease in the tropospheric lifetime of PAN (9.0 hours).

Surface and zonal distribution of PAN for the present scenario
The annual mean surface and zonal distribution of PAN from STOCHEM-CRI and their comparison with STOCHEM-ISOP for the present scenario are shown in Figure 1 and  (Figure 1a and Figure 1b). The increased biomass burning activity and formation from the degradation chemistry of both isoprene and monoterpenes in STOCHEM-CRI [60] lead to more PAN formation over forested regions. The more complex mechanism in STOCHEM-CRI shows the importance of PAN especially from isoprene and continued work in improving measurement techniques, increasing the amount of measurements and improving mechanistic understanding is essential. The inclusion of HO x recycling leads to an increase of PAN over source regions by as much as 10% and a decrease of PAN over sink regions by as much as 40% (Figure 1c and Figure  1d).

Model-measurement comparison
The data compilation of Emmons et al. [54], Horowitz et al. [55], and Fischer et al. [56] containing measurements of PAN from a selection of aircraft campaigns (see Appendix B for more details) produced a set of data representing a broad distribution of regions throughout the troposphere which was used to compare with the modelled PAN in order to evaluate model performance ( Figure 3). The CRI mechanism in STOCHEM over-predicts the upper tropospheric PAN substantially for most of the locations relative to the measured data. Similar results were found in the global modelling results from the MATCH v3 model showed by von Kuhlmann et al. [75] in where the model over-predicted PAN in the remote troposphere and more often at higher altitudes. In general, PAN concentrations increase with altitude in the troposphere because of the increased thermochemical stability in the cold upper troposphere. The differences between average modelled and observed PAN concentrations for all locations referred to by mean biases are found to be +17 and +80 pptv for the lower troposphere (~0-4 km) and upper troposphere (~6-12 km), respectively.
The increases in PAN in our study are driven by an increase of the CH 3 CO 3 (which is the precursor radical) due to the increased complexity of the emitted VOC speciation represented in STOCHEM-CRI. Among all of the campaigns, the regions dominated by vegetation and biomass burning emissions (e.g. TRACE-A) have higher PAN levels and the model PAN in the bottom five levels (up to 5 km) fit Page 3 of 10 reasonably well with the measured mean. The over-prediction of PAN in the upper troposphere could be due to the higher abundances of CH 3 CO 3 in the STOCHEM-CRI model, which are ubiquitously distributed throughout the troposphere because of its formation during the oxidation of most NMVOCs [76]. PEM-Tropics A was a campaign to investigate the effect of biomass burning on the NO y species in the remote Pacific during September and October.   The HO x recycling caused an increase in OH, reducing the lifetimes of NMVOCs resulting in a decrease in CH 3 CO 3 concentrations to produce PAN at the free troposphere, which has improved agreement with measurements for most of the locations. The mean biases from measurements are reduced by 90% and 40% for the lower troposphere and upper troposphere, respectively upon addition of HO x recycling in STOCHEM-CRI.
The rate coefficient of the temperature dependent PAN formation used in the STOCHEM-CRI model (k 0 =2.7×10 -28 (T/300)-7.1 cm 6 molecule -2 s -1 and k ∞ =12.1×10 -12 (T/300) -0.9 cm 3 molecule -1 s -1 ) were subsequently perturbed upwards and downwards by considering the uncertainties of the rate coefficients. Therefore, two more simulations were performed with two sets of different PAN formation rates deviating k 0 by ± 50% and k ∞ by ± 15% from the unperturbed run with a high rate coefficient referred as "STOCHEM-HIGHK" and with low rate coefficient referred as "STOCHEM-LOWK". The STOCHEM-HIGHK run shows the average increase of PAN mixing ratios in the lower and upper troposphere by 6% and 4%, respectively. In the STOCHEM-LOWK run, the overestimate of the average model PAN for all campaign sites is reduced by 5% and 7% at the lower and upper troposphere, respectively, but the decrease of PAN mixing ratios with altitude above ~6 km is not large enough to decrease the discrepancy of the model-measurement PAN data especially in the upper troposphere.

Page 5 of 10
Twenty surface measurement campaigns have been compiled to form a dataset of PAN (more details in Appendix C) which have also been used to evaluate the model performance. Figure 4 shows the comparison of the seasonal cycle of model PAN mixing ratios with the surface measurements for a series of locations. The modelled PAN does not exhibit a clear seasonal correlation for most of the stations (Figure 4). The seasonality of PAN is found to be sensitive to temperature (through the thermal decomposition rate of PAN), OH mixing ratios and sunlight. However, PAN is long-lived with respect to reaction with OH or photolysis and its lifetime in the boundary layer is controlled by thermal decomposition and by transport and decomposition. During summer, the enhancement in the PAN formation rate would be expected because of the increased formation of CH 3 CO 3 during the oxidation of VOCs by OH. However, the accompanying increased temperature is also responsible for the rapid increase in the rate of thermal decomposition of PAN. These combined effects lead to larger PAN abundances either during spring or during summer, depending on the location and meteorological condition.
PAN abundances at NH stations are at a maximum during April-May-June due to higher photochemical production in the presence of VOC and sunlight when they have a longer thermal lifetime [77][78] and at a minimum during November-December-January because of the lower photochemical production rate from CH 3 CO 3 and NO 2 . The medium to long-range transportation of polluted air masses from the European boundary layer at European stations (e.g. Jungfraujoch, Lindau, Harwell, Munich, and Athens) resulted in a maximum model PAN in spring, which is consistent with the measurement data. The abundances of both model and measured PAN at Asian stations (e.g. Seoul, Beijing, Lanzhou, and Rishiri Island) during July-August are found to be lower than expected because of the significant thermal decomposition of PAN during these months. The model produces less PAN over East Asia (e.g. Beijing, Lanzhou, and Seoul) because of using limited anthropogenic aromatic emissions in STOCHEM-CRI. Fischer et al. [56] performed a simulation with inclusion of emissions of aromatic species and found that the aromatics can account for 30% of the PAN in the Asian outflow region. The overestimation of model data compared with measurement data is found for most of the US stations (e.g. North Carolina, Frijoles, Nashville, Georgia, and Mount Bachelor Observatory) and also Greenland could be due to the overprediction of model NO x emissions and CH 3 CO 3 formation.
Both the model and surface PAN observations indicate that the springtime maximum is pronounced at Zeppelin Mountain near Ny-Ålesund, Svalbard which is due to long range transport of pollutants from northern mid-latitudes sources [12,36]. During summer, the maximum thermal decomposition of PAN at Zeppelin Mountain occurs [36] and the transport of midlatitude air into high northern latitudes is likely [79] which resulted in the summer minimum. During fall, the transport and temperature favours higher PAN levels at Zeppelin Mountain, but during winter, the photochemical production of PAN is reduced due to the lack of sunlight in midlatitudes [36].

Page 6 of 10
The SH stations (e.g. coastal Antarctica, Neumayer, and Santiago) show an analogous seasonal trend with maximum and minimum mixing ratios found in spring (September-October) and winter seasons (December-January-February). The surface PAN mixing ratios in Antarctica (e.g. coastal Antarctica and Neumayer) is found to be lower than that in the Arctic because the larger distance from other continents makes the transport of PAN or its precursors to Antarctica less effective.

Preindustrial scenario and implication of HO x recycling
The relative changes of surface mixing ratios of PAN (up to 0.6 ppb) from preindustrial to present day scenarios are found in the NH in between 30°N and 60°N due to greater anthropogenic sources of NO x and VOCs in the present day scenario (Figures 5a and 5c). The percentage change results show that the mean surface PAN mixing ratios have increased since preindustrial times by 400% in the most of the NH surface (maximum changes of up to 1500% near the equator remote region) and by 100% in the SH. Mixing ratios in the upper troposphere have increased by 400-600% largely in between 20°N to 35°N. The percentage changes decline with altitude and are smaller away from the strong anthropogenic source region (Figures 5b and 5d). The increased PAN in the present day scenario contribute significantly to the total reactive nitrogen (NO y ) budget in the troposphere, especially in the NH. The global burden of PAN has increased from 220 Gg(N) in CRI1800 to 398 Gg(N) in the base simulation, which is an increase of 81%.
The importance of the HO x recycling mechanisms with reference to oxygenated peroxy radicals is caused by an increase in HO x levels   (Figure 6a). The increased OH from isoprene recycling has a greater effect than the decreased OH production from the oxygenated peroxy radicals with an increment of OH global burden by 5%. A decrease in NO x for ISOP-1800 relative to B-1800 is found over most of the areas with the largest percentage changes (up to 25% decrease) over the remote oceanic regions (Figure 6b), only two areas (e.g. Amazon and South-East Asia) in where the annual mean NO x are increased by up to 2% because of the increased biomass burning and vegetation NO x emissions from the Amazon and anthropogenic NO x emissions from South-East Asia. The isoprene concentrations in South-East Asia are lower than that in Amazon [69], but the large change in NO x for South-East Asia has led to a much more significant Page 7 of 10 effect of the recycling chemistry in this region. This highlights the important balance between isoprene emissions and NO x levels. HO x recycling occurs at low NO x levels and thus removing the NO x from anthropogenic sources in the preindustrial scenario makes the recycling more efficient. The inclusion of HO x recycling leads to a decrease in PAN (Figure 6d) and an increase in HNO 3 formation over emission regions (Figure 6c). HNO 3 is rapidly deposited over land and its removal leads to reduced global levels of gaseous oxidised nitrogen species (e.g. PAN) which leads to less oxidised nitrogen being transported to remote regions and thus the largest percentage changes in NO x are observed over the remote oceanic regions (Figure 6b). The changes in O 3 are driven by the redistribution of NO x which is found to be largest in remote oceanic regions (Figure 6e). The greatest percentage change (up to 9% decrease) in O 3 concentration is found at the surface in equatorial regions upon including HO x recycling in the preindustrial scenario.

Conclusion
The complexity of the VOC degradation in STOCHEM-CRI resulted in an increase of tropospheric PAN which act as nitrogen reservoir species before releasing and transporting reactive nitrogen (NO x ) away from the pollution centers. The increase in the level of PAN with altitude is consistent with reduced thermal decomposition in the upper troposphere. The overall comparison between measurements and simulations show that STOCHEM-CRI overpredicts the concentrations of PAN in the troposphere. The global burden of PAN for the present scenario is found to be 398 Gg(N) and the inclusion of HO x recycling reduces PAN by as much as 13%, bringing the model into closer agreement with measured values. The globally averaged tropospheric lifetimes of PAN is found to be 9.3 hours. The abundance of PAN have been seen in the continental atmosphere, with mixing ratios up to 2.0 ppb over the polluted regions of North America during J-J-A and up to 1.2 ppb over the Amazon rainforest during D-J-F. The inclusion of HO x recycling leads to a decrease of PAN by up to 40% over the oceans in the surface layers and by up to 10% in the low to middle tropospheric layer of equatorial regions (during J-J-A) and SH regions (during D-J-F) in a present scenario. The maximum percentage changes (up to 1500%) from preindustrial to present scenarios for PAN are found near the equator remote region and the inclusion of HO x recycling in the preindustrial scenario can decrease O 3 concentrations by up to 9%.