Seasonal variation of atmospheric pollutants transport in central
Chile: dynamics and consequences

Abstract. Central Chile faces atmospheric pollution issues all year long, in relation with elevated concentrations of fine particulate matter during the cold months and tropospheric ozone during the warm season. In addition to public health issues, environmental problems regarding vegetation growth and water supply, as well as meteorological feedback are at stake. Sharp spatial gradients in regional emissions along with a complex geographical situation make for variable and heterogeneous dynamics in the localization and long-range transport of pollutants, with seasonal differences. Based on chemistry-transport modeling with WRF-CHIMERE, this work studies for one winter period and one summer period: (i) the contribution of emissions from the Santiago Metropolitan Area to air pollution in central Chile, (ii) the reciprocal contribution of regional pollutants transported into the Santiago basin. The underlying 3-dimensional advection patterns are investigated. We find that on average for the winter period 5 μg m−3 to 10 μg m−3 of fine particulate matter in Santiago come from regional transport, corresponding to 13 % to 15 % of average concentrations. In turn, emissions from the Metropolitan Area contribute to 5 % to 10 % of fine particulate matter pollution as far as 4° north and 4° south. Wintertime transport occurs mostly close to the surface. In summertime, exported precursors from Santiago, in combination with mountain-valley circulation dynamics, are found to account for most of ozone formation in the adjacent Andes cordillera and to create a persistent plume of ozone of more than 50 ppb, extending along 80 km horizontally and 1.5 km vertically, and located several hundred meters above ground, slightly north of Santiago. This work constitutes the first description of such an ozone bubble formation mechanism. Emissions of precursors from the capital city also affect daily maxima of surface ozone hundreds of kilometers away. In parallel, cutting emissions of precursors in the Santiago basin results in an increase of surface ozone mixing ratios in its western area.


dominates the signal over the simulation domain. In the continuation, given their relevance for the associated seasons, PM 2.5 in wintertime and O 3 and its precursors in summertime will be considered as the variables of interest, although PM 2.5 in summertime and O 3 in wintertime can be occasionally discussed.

Simulation validation
Surface meteorology and pollutants concentrations are validated using data from the automated air quality and meteorology 100 monitoring network of Chile, known as Sistema de Información Nacional de Calidad del Aire (https://sinca.mma.gob.cl/index. php/, last access October 1 2020). Different stations across central Chile are considered depending on data availability for the simulated periods (stations locations can be found in Figure 1a). Meteorological vertical profiles in downtown Santiago, for a few days in July 2015, were provided by the Chilean meteorological office, Dirección Meteorológica de Chile.
Simulation scores for surface and vertical profile meteorology are gathered in Tables A2 and A3. Biases on daily mean 105 temperature range between -1.23 • C and 0.31 • C in wintertime except for the mountainous location of Los Andes where the bias reaches -3.33 • C. In summertime the bias is between 0.07 • C to 0.67 • C. For both periods, correlations on surface temperature vary between 0.7 and 0.89, except for Viña del Mar where it drops to 0.25 in wintertime and 0.18 in summertime, which can be explained by the location of this station near the ocean. The corresponding grid point in the model straddles ocean and land hence featuring a strong gradient, and as a result may not be representative of this coastal city. This remark applies 110 to all meteorological variables at coastal sites. The model is a little too dry with average biases on surface relative humidity of -15% to -20% but shows fair correlations around 0.8 to 0.9. Surface winds are fairly reproduced, with limited biases and wind gusts well captured, although the correlations can be low for some locations. Figures A1 and A2 compare observed and simulated surface wind distributions for four sites. Summertime shows a good reproduction of wind regimes for all locations, expect for a small positive bias on speeds. In wintertime, winds are more variable and follow less clear patterns, so that the 115 model performance is not as good, especially for the coastal locations of La Serena and Santo Domingo. A positive bias on speed is also observed. Most features of the vertical profiles of temperature, relative humidity and winds are well reproduced, with very good correlations and limited biases for four winter days in downtown Santiago (Tab. A3). On the whole, these statistics give confidence in the ability of the model to produce realistic transport events for both seasons. Figure 2 shows a scatter plot of measured and modeled daily mean concentrations of PM 2.5 in wintertime and daily maximum 120 mixing ratio of O 3 in summertime. For PM 2.5 the simulation performs better for sites far inland with correlations of 0.77 and 0.69 for Rancagua and Independencia, respectively. Correlations for the two coastal sites considered (Viña del Mar and La Serena) are more moderate with 0.34 and 0.25, respectively. The same issue of straddling grid points explained previously leads to these degraded statistics: while observations at these sites show a chaotic time series, the model produces a smoother diurnal cycle due to the grid point being partially over the ocean. However for those two sites, biases remain small, while the model 125 systematically underestimates concentrations in Rancagua, and slightly overestimates PM 2.5 in Santiago (Independencia). For O 3 the picture is similar with a good reproduction of daily peaks in summertime. Despite the relatively coarse resolution of the simulation and strong spatial heterogeneity in precursors emissions, limited biases of a few ppb are obtained on O 3 peaks in summertime, down to only 1 ppb at the most O 3 -polluted site of Las Condes (northeastern Santiago). The diurnal cycle of O 3 is also well reproduced with hourly correlations (not shown here) of 0.67 for Viña del Mar and 0.8 to 0.9 for the three other 130 sites. In parallel, summertime NO x mixing ratios within the Santiago area (not shown here) are well captured by the model with mean biases between 0.05 and 1.23 ppb for three stations in Santiago (Las Condes, Puente Alto -southeastern Santiagoand Independencia), associated with decent hourly correlations between 0.43 and 0.59.
The lack of available data for NO x at other locations in central Chile, and more generally VOC data for all locations makes the simulation validation impossible for these precursors. However, the good reproduction of both O 3 and NO x advocates in 135 favor of the production of adequate levels of VOC, at least in the Santiago area. In addition, although such a work is yet to be done for Chile, comparisons of HTAP emissions of VOC with more refined national inventories in South America show that despite a spatial distribution and speciation profile with margin for improvement, total amounts of emitted VOC as provided by HTAP are well in the range of higher resolution inventories based on local observations for Argentina (Puliafito et al., 2017) and São Paulo (Dominutti et al., 2020). In parallel, emissions of NO x and PM 10 from HTAP for the region of Santiago show 140 similar biases and discrepancies as other mega-cities in South America such as Buenos Aires and Rio de Janeiro , so that we can assume this similarity also applies for VOC. By extension, if HTAP for Santiago is comparable to HTAP for Argentinian and Brazilian mega-cities, where VOC total emitted amounts are adequately provided by HTAP, VOC emissions for Santiago likely have the appropriate magnitude as well, despite some discrepancies. Besides, the aforementioned discrepancies are critical when it comes to more detailed approaches for policy making but for the purpose of the present work, 145 having the proper total amount is sufficient as we apply our own downscaling methodology, do not discuss very high-resolution processes, and rely mostly on sensitivity analysis. 6 https://doi.org/10.5194/acp-2020-1249 Preprint. Discussion started: 21 January 2021 c Author(s) 2021. CC BY 4.0 License.
In conclusion, PM 2.5 in wintertime and O 3 and its precursors in summertime, the key pollutants for their respective seasons, as well as meteorological conditions, are fairly reproduced by the model for a selection of sites throughout central Chile, which gives confidence in the model outputs that are described and analyzed in the following sections.

Results
The semi-permanent South Pacific High, centered around (30 • S, 110 • W), along with the elevated Andes cordillera, are two large-scale drivers of the surface wind systems in central Chile. The persistent high induces high velocity southwesterlies blowing along the coast during both daytime and nighttime (Fig. 3). These also penetrate deep into land as far as the Andes in summertime during the day, before being blocked by the mountains. On the other side of the cordillera, less intensive easterlies 155 coming from Argentina encounter the foothills. Also, the presence of the Andes leads to the development of mountain-valley circulation patterns (e.g. Whiteman, 2000) when the differential heating between narrow valleys and wide plains at the onset and offset of the day lead to the creation of upslope westerlies during daytime (as seen in Fig. 3b), and a reversal at nighttime with downslope easterlies (Fig. 3d). Although this pattern can be perturbed by clouds or synoptic-scale transient phenomena such as coastal lows, it represents the typical surface wind diurnal cycle for basins along the cordillera. From these mean wind 160 fields, the dominant advection pathways of pollutants can be inferred. Polluted air masses are on average blown towards the cordillera and the north during daytime in summertime, and have more complex dynamics in wintertime but also transport northward in general. Deeper and more turbulent planetary boundary layer heights during daytime, as observed in summertime ( Fig. 3a) also enable the vertical export of pollutants up into the free troposphere (FT) where they can be advected farther, while wintertime shallower boundary layers (Fig. 3b) imply more stagnation of air masses. Consistently with the mean wind fields and emission rates of pollutants in central Chile discussed previously, anthropogenic emissions in the Metropolitan Area of Santiago significantly influence surface atmospheric composition over a large region, over land, the Andes and the Pacific ocean. The following results are based on the analysis of sensitivity to emissions from 170 Santiago described in Section 2.1: the difference between the simulation with (baseline case) and without (contribution case) emissions from Santiago yields their contribution to atmospheric composition over the domain. Figures 4a and 4b show the average wintertime PM 2.5 plume (absolute and relative, respectively) attributable to emissions from the capital area. The direct western vicinity of the Santiago basin receives, on average 5 µg m −3 to 15 µg m −3 coming from the capital, a few tens of kilometers from the source, corresponding to more than 30% of the signal simulated in the baseline scenario for this area. At 175 the scale of hundreds of kilometers, the export drops to a few µg m −3 , corresponding to 5% to 20%. It is worth noting that the relative contribution of emissions from Santiago remains greater than 5% on a area as large as more than 8 • meridionally and 3 • zonally, hence stressing the significant impact of the capital on atmospheric composition for the whole region (Fig. 4b). In particular, the southern part of the plume is transported over the Andes down to Argentina with a large spread of several degrees of longitude, whereas the northern part extends mostly along the coast in a narrower manner and transports as far as 180 the boundary of the simulation domain.
More specifically, urban areas along the north-south axis of Santiago (Curicó, San Fernando, Rancagua and Los Andes in Fig. 4a) receive 1 µg m −3 to 2 µg m −3 from Santiago on a hourly basis on average, corresponding to 4% to 8% of the baseline concentrations ( Fig. 4c). Sporadically, up to more than 20 µg m −3 in Rancagua and 9 µg m −3 in San Fernando, Curicó and Los Andes can be attributed to emissions from Santiago. These significant contributions likely lead to alert thresholds crossing for 185 some hours in these cities.
On the eastern side of the Santiago basin is the Andes cordillera. We examine the contribution of Santiago emissions in a village (San Gabriel) and a summit (Maipo volcano) along the Maipo canyon, southeast of Santiago. We find that for the village 34% (1 µg m −3 ) of PM 2.5 , on average, is transported from the metropolitan area. This is consistent with the mountain-valley circulation patterns aforementioned, leading to the intrusion of urban air masses deep into the canyon (Lapere et al., 2021). We 190 acknowledge that the estimate of 34% is probably larger than reality since the HTAP inventory does not capture properly local emissions in the village of San Gabriel, which likely dominate the signal, especially with wood burning for residential heating being largely used in such villages in wintertime. In the Maipo volcano area (summit at 5264 m a.s.l.), a small contribution in absolute value is found (less than 0.5 µg m −3 ) although it can reach up to more than 2 µg m −3 occasionally, but this corresponds to 20% of the signal there on average. This area is covered in snow during wintertime, so that despite the small magnitude of 195 the import of PM 2.5 it can lead to significant radiative effects when deposited, especially given the large fraction of black carbon in PM 2.5 emitted in Santiago (around 15% at Independencia -not shown here).
The Viña del Mar-Valparaíso area is the second largest populated region of Chile, located on the ocean coast west of Santiago.
In wintertime, it is downwind of Santiago, which leads to an average import of particulate matter from the capital city of 3 µg m −3 (18%) and sporadically up to 18 µg m −3 . Again, air quality in this urban area is worsened by export from Santiago, 200 by a significant share. Further north, at the location of La Serena, which also suffers from bad air quality in wintertime, the contribution of Santiago emissions is more moderate but still remains significant in absolute value although its contribution is only 1%.

Summertime O 3
In summertime, except for PM 2.5 emitted by biomass burning events (not considered here), O 3 is the pollutant raising concern.
205 Combined significant emissions of NO x and VOC are required, in the presence of sunlight, to generate high mixing ratios of O 3 . However, a lot of non-linearities are involved in the tropospheric O 3 cycle, so that the sensitivity to its precursors emissions is not straightforward. For instance, imbalances in the ratio of NO x and VOC decrease O 3 formation. Given the crucial role of photolysis in O 3 formation, it features a strong diurnal cycle, with levels coming back to low values at night. Thus, the important variable to determine whether O 3 pollution is high is its daily maximum mixing ratio, on which we will mostly focus 210 hereafter. Figure 5a shows the average of maximum hourly O 3 mixing ratio at ground-level observed each day in the baseline case (SB). In the baseline scenario, O 3 is mostly found at harmful levels near the area of Santiago, and mainly on its eastern side.
Again, mountain-valley circulation accounts for this observation: afternoon westerlies blow O 3 precursors, present in large amounts in urban air masses, towards the Andes. NO x lifetime is a few hours at most, depending on the reactivity of VOC and 215 NO 2 density (e.g. Laughner and Cohen, 2019), while most VOC have an atmospheric lifetime of several days (e.g. Monod et al., 2001), so that on the way, NO x are more consumed than VOC. Consequently, while urban air is mostly a NO x -rich environment, precursors ratios become more balanced along with the export so as to create more favorable conditions for O 3 formation once reaching less urbanized areas. Such a mechanism is observed for Paris and its suburbs for instance (e.g. Menut et al., 2000). Export by easterlies occur less frequently and mostly at night when O 3 cannot be created due to lack of sunlight, 220 which is why the rural area west of Santiago shows smaller O 3 maxima. This mechanism explains the large concentrations of O 3 mostly found east of Santiago. Except for the center part of the domain where average daily maxima reach more than 100 ppb, O 3 pollution is less concerning elsewhere in central Chile where they range between 10 ppb and 35 ppb. Figure 5b shows the decrease in O 3 daily maxima induced by eliminating emissions of the Santiago basin. The spatial pattern is again consistent with the mechanism introduced previously: emissions of precursors from Santiago are the main origin for 225 O 3 formation in the cordillera and north of the city, with a reduction of more than 50 ppb of the daily maxima over this area when the capital city no longer emits pollutants.
More specifically, the northern city of Los Andes shows a decrease of its daily maxima by 15 ppb on average (Fig. 5b) while the average mixing ratio drops from 40 pbb to 33 ppb (Fig. 5c). Similarly, at the ski resort site of Valle Nevado, which shows concerning levels of more than 60 ppb of O 3 on average in the baseline case, the mixing ratio drops to an almost constant value  maximum, dots show the average. "ns" indicates that the distribution in the SB and SC scenarios cannot be distinguished at the 90% level based on a t-test. "*" indicates that they are different at the 99% level based on that same t-test.
On the other hand, the western and southern areas adjacent to Santiago are barely sensitive to its emissions. In Figure 5c, the distribution of hourly O 3 mixing ratio at Viña del Mar and Rancagua is nearly the same in both scenarios, except for the maximum at Rancagua which is a few ppb lesser in scenario SC than SB. For those two locations, the difference in O 3 240 distribution with or without Santiago emissions is not significant at the 90% level, while for all other locations, distributions are significantly different at the 99% level. Regarding the regional transport of PM 2.5 in summertime, the plume shows a similar extent to O 3 in summertime, i.e. an area of influence lesser than in wintertime near the surface (not shown here). O 3 is barely produced in wintertime due to a zenith angle of the sun closer to the horizon and shorter duration of days, resulting in less radiative power and barely active 245 photolysis, so that the question of the export of its precursors at that season is less relevant.

Contribution of regional emissions to atmospheric composition in Santiago
Similarly to the previous approach, we can, in a symmetric manner, deduce the contribution of transport from remote sources to atmospheric composition in the Santiago Metropolitan Area. This is achieved by looking at concentrations in the contribution case, that exclusively originate from non-local sources. Hereafter the analysis focuses again on PM 2.5 in wintertime and O 3 in 250 summertime.
3.2.1 Wintertime PM 2.5 Figure 4b indicated that local emissions largely dominate the wintertime PM 2.5 signal for Santiago, with 50% to 100% of the mean surface concentration of PM 2.5 originating locally within the white rectangle. Nevertheless, the contribution of transport is also significant although heterogeneous in Santiago. Figure 6a shows that in the baseline scenario, the northern and western 255 parts of Santiago feature higher levels of PM 2.5 , with average concentrations ranging between 30 µg m −3 and 100 µg m −3 for the whole metropolis. In the contribution scenario (Fig. 6b), this pattern is partly recovered, with a smoother gradient from west to east. The underlying conclusion is twofold. First, the transport of pollutants in Santiago mostly comes from the west, which is consistent with the presence of the cordillera in the east not featuring many sources of pollution, and the dominant westerly daytime wind direction in wintertime. Second, the districts of Santiago facing the worst air quality are also the ones 260 where the transport of pollutants is larger.
However, the differing patterns between Fig. 6a and 6b also shows that local sources in these districts are also stronger. If emissions were similar, the observed gradient in WB would be closer to that in WC. Consistently with this observed westward gradient, we define 5 zones of interest, comprising 6 grid points each along a meridional axis (rectangles and dots in Fig. 6b).
This arrangement ensures that most of the city is covered while maintaining the west-east variability. For each zone we look 265 at the distribution of PM 2.5 , averaged over the 6 grid points, in scenarios WB and WB-WC (Fig. 6c) and WC (Fig. 6d).
When averaged meridionally, the westward gradient is also obtained in scenario WB, and conserved when the contribution of transport is substracted (WB versus (WB-WC) in Fig. 6c). On average, the contribution of transport to PM 2.5 concentration ranges between 10 µg m −3 for the westernmost area and 5 µg m −3 for the easternmost part, with a monotonic spatial variation ( Fig. 6d). However, this amount always corresponds to 13% to 15% of the WB concentrations. This number is well in line 270 with Barraza et al. (2017) that found 9% of PM 2.5 in Santiago coming from coastal sources, for the period 2011-2012. It is worth noting that given the observed westward gradient, we also recover that coastal sources are likely the main contributor to imported PM 2.5 . Again, transport is larger in western Santiago, and can sporadically reach up to 30 µg m −3 , but does not constitute a greater share than in the east.
This averaged picture provides a first clue as to the main origin of PM 2.5 transport of Santiago but the picture can be 275 refined by looking at the joint distribution of hourly wind direction and PM 2.5 concentrations as shown in Fig. 7. At the selected southeastern location, mostly clean air comes from the east, i.e. from the Andes where pollutants sources are scarce (less than 5 µg m −3 for almost every hour), while winds blowing from the southwest can transport concentrations as high as more than 20 µg m −3 for some hours, pointing to the southern cities of Rancagua or San Fernando mentioned previously, or the southwestern urban location of Melipilla (33.6 • S,71.2 • W). At the northeastern site, winds mainly come from the north 280 where only a handful of urban areas are found, hence leading to a transport seldom exceeding 10 µg m −3 . In the center of the metropolitan area, winds are either southwesterlies (from the Rancagua and Melipilla areas) or northwesterlies (from the Viña del Mar-Valparaíso area), with both cases leading to similar amounts of imported PM 2.5 mostly above 10 µg m −3 , i.e. above average. The picture is similar for the northwestern and southwestern sites although wind directions are shifted.
In the northwest, center and northeast, the dominant wind directions also coincide with higher maximum relative contributions 285 of transport over the period along these directions (black diamonds in Fig. 7). Sporadically, significant transport events can also come from less frequently observed directions. The maximum relative contribution obtained for the northwest point when winds are from NNE is 75% for example, while such winds occur less than 1% of the time. For southwest and southeast locations, these maximum relative transport episodes are observed when winds blow from the south, in particular in the southeast where it can reach up to 100%.  In summary, wintertime PM 2.5 concentrations in Santiago are significantly (5 µg m −3 to 10 µg m −3 on average) and always (at least 1 µg m −3 for every hour) affected by transport, with identifiable origins, and although the different districts are not equally affected in absolute value, the relative burden of imported particulate matter is equivalent.

Summertime O 3
Based on the same approach, we find that the summertime transport of NO x within the Santiago basin never exceeds 0.5 ppb, 295 while average values in the SB scenario are between 5 ppb and 40 ppb, with a similar westward gradient as observed for PM 2.5 (not shown here). At the 90% level, NO x transport is thus not significant. Similarly, the transport of VOC is homogeneous over the whole basin at 2 ppb on average, while between 20 ppb and 60 ppb in the SB scenario, which is again not significant at the 90% level. Santiago is thus not affected by the transport of O 3 precursors. area of Santiago is more affected by O 3 pollution compared to the western area ( Fig. 8a and 8c), consistently with observations and the literature (e.g. Menares et al., 2020), due to a more balanced VOC/NO x ratio than in the western area. At Independencia (center Santiago), the VOC/NO x ratio at emission is between 1:1 and 2:1 on average. Contrarily, at Las Condes (eastern Santiago) the VOC/NO x ratio at emission is around 6:1 on average, in the baseline case (not shown here). A typical O 3 formation ridge line of the VOC/NO x concentration ratio in urban areas is around 6:1 to 8:1 (e.g. National Research Council, 305 1991; Sillman, 1999), so that Independencia features a VOC-limited regime, while Las Condes features a balanced regime favorable to O 3 formation, hence the larger amounts found at the latter location. Figure 8b shows the consequences, on O 3 surface mixing ratio, of eliminating emissions within Santiago. Given the configuration described previously, in the baseline case, the VOC-limited districts of western Santiago feature mixing ratios well below the background level due to the titration of O 3 by excess quantities of NO x , while the eastern districts feature mixing ratios above 310 the atmospheric background level due to excess O 3 formation under a favorable regime. As a result of shutting off emissions, given that there is no import of precursors as evidenced above, the whole area is set to the background O 3 level of around 30 ppb described in Section 3.1.2, since there is no influence of anthropogenic pollutants anymore. Therefore, this corresponds to an increase (decrease, respectively) of O 3 in western (eastern, respectively) Santiago, thus explaining the dipole obtained in Figure 8b. Such an evolution is also clear in the evolution of the distribution of hourly O 3 mixing ratios across the city 315 between scenario SB and SC (Fig. 8c). While in scenario SB the distribution is shifted towards larger mixing ratios when going eastward, in scenario SC all distributions are equal (significant at the 99% level). The leveling of mixing ratios, with no gradient across the metropolitan area in scenario SC constitutes an additional evidence that O 3 in Santiago is not affected by long-range transport, otherwise heterogeneous patterns similar to what is observed in Figure 6b would be obtained.

Advection processes 320
As discussed around Figure 3 and observed in Figure 4 and Figure 5, advection patterns of pollutants differ between wintertime and summertime. So far, the analyses focused on surface fields, but processes along the vertical drive these differences. Figure 9 shows an average latitude/altitude transect, along central Chile, of winds, afternoon mixing layer height and pollutants concentrations for the corresponding season, in the baseline scenario and the Santiago isolated contribution case.
In wintertime (Fig. 9b and 9c), the boundary layer (solid white line) is shallow, and average winds in the FT are strong, 325 consistently with the observed semi-permanent inversion layer in the region. As a result, the PM 2.5 emitted in large amounts mostly remain trapped within this shallow mixing layer. Injection of polluted air masses into the lower FT can also occur, through mountain venting, as described in Lapere et al. (2021), explaining why residual concentrations of 1 to 5 µg m −3 are observed higher up in the baseline and contribution scenarios at the latitude of Santiago (Fig. 9b and 9c). Nevertheless, the long-range export of PM 2.5 from Santiago observed in Figure 4 is mainly driven by advection within the boundary layer, 330 close to the ground, by weak winds, as evidenced by Figure 9c. Except for a shallow residual layer located above the average afternoon boundary layer, pollutants emitted from Santiago remain within it along the transect from Santiago to Viña del Mar.
The pattern changes when reaching the seashore however, with significant wind shears lifting the PM 2.5 layer above the mixing layer (rightmost part in Fig. 9c) thus explaining the wide northward extent of Santiago contribution, due to transport higher up, by intense southerlies.

335
Interestingly, the summertime transect of O 3 shows a sharp maximum near the latitude of Santiago, several hundred meters above the ground. Figure 9d shows the formation of an O 3 bubble of more than 50 ppb on average, i.e. 15 ppb above the 35 ppb background mixing ratio observed in scenario SB (dominant yellow/green levels in altitude in Figure 9d), around latitude 33 • S i.e. slightly north of Santiago, above the planetary boundary layer, extending between 1.5 km and 3 km altitudes. This additional O 3 plume is mostly attributable to emissions of precursors in Santiago, that account for more than 15 ppb of O 3 on average 340 at the location of the bubble (Fig. 9e), i.e. the background level exceedance. Thus, despite a relatively limited area where precursors from Santiago affect O 3 formation near the ground (Fig. 5b), their impact on the vertical is more dramatic. The process underlying the formation of this significant O 3 bubble clearly departing from the background, is discussed hereafter. It is also worth noting that even though export of O 3 close to the surface is limited (Fig. 5b), it is more widespread higher up, with a residual layer originating from Santiago emissions of a few ppb extending 2 km vertically and transporting northward 345 in the FT along 2 • of latitude (Fig. 9e). Given the proximity of the Andes cordillera to the Santiago basin, the formation of the aforementioned O 3 bubble finds its origin in the mountain-valley circulation and the associated mountain venting mechanism. Daytime upslope winds, strong in summertime, lift polluted air masses from the atmospheric boundary layer (ABL) over Santiago into the lower FT, possibly above another location depending on the FT winds direction. McKendry and Lundgren (2000) and Lu and Turco (1996) find 350 that this process is a net sink for boundary layer O 3 in British Columbia and the Los Angeles basin, respectively. Henne et al. (2005) find that the effect of venting in an Alpine environment on FT O 3 concentrations strongly depends on initial mixing ratios within the vented ABL, with either net production if ABL mixing ratios of O 3 are high (urban valleys), or net loss if they are low (remote valleys). However, our study case falls into none of the aforementioned. Such a bubble of O 3 , detached from the ground, with a mixing ratio much higher than at ground level is not found in the literature to our knowledge. More 355 moderate injections, with O 3 mixing ratios lower to similar to surface levels are usually observed. In our case, we find that the venting of precursors from the Santiago polluted ABL leads to the net production of large quantities of O 3 in the FT, larger than at the surface. Schematically, there is a larger export of VOC than NO x in the FT (Fig. A3a and A3b) which makes for a balanced chemical regime (below 0.5 is NO x -limited, above 0.5 is NO x -rich) at around 0.5 at 2 km altitude, while the regime is close to 1 near the surface hence unfavorable to O 3 production, along the whole transect, due to dominant urban NO x emissions 360 (Fig. A3c). Also, the export of Peroxyacetyl nitrate (PAN), which is a NO x carrier, into the FT, contributes to enhanced O 3 formation (Fig. A3d). Figure 10 further sheds light on the dynamics of this mechanism. Early in the day, at 11:00 UTC (Chile is UTC-3 in summertime), both NO x and NMVOC (non-methane VOC) are highly concentrated near the ground (morning peak of emissions from traffic), and start ascending the Andean foothills, but no O 3 is formed in the region given that sunlight is not intense yet 365 (leftmost column in Fig. 10). A few hours later, at 12:00 LT (15:00 UTC), the precursors have a similar distribution as in early morning, but the photolysis starts taking place and O 3 is created at the top of the precursors plume, well above the ground.
Two factors explain that O 3 is not formed near the source of precursors. First, given the urban environment of the basin, NO x are emitted in large quantities so that the VOC/NO x ratio is adverse to neutral, and O 3 is formed in small amounts (see the related discussion in Sect. 3.2.2). But NO x have a lifetime much shorter than most VOC (e.g. Monod et al., 2001;Laughner 370 and Cohen, 2019), and when the polluted air parcels are lifted up, the ratio becomes more balanced as NO x is consumed closer to the ground. Due to these asymmetric lifetimes and hence export, at some point on the vertical the ratio becomes favorable ( Fig. A3c) and O 3 is formed in large quantities. The second factor comes from the increase of photolysis rates with altitude.
Several hundred meters above the ground, near or above the mixing layer, photolysis rates of NO 2 are much faster than at ground level (e.g. Pfister et al., 2000) hence favoring the formation of O 3 , all other things being equal. In our simulation, on 375 average, the photolysis of NO 2 is 20% faster 1000 m above ground than at the surface. Also, Figure 9d shows that at the point where the O 3 plume is denser, winds are weak to null on average, so that precursors stagnate, again allowing for more O 3 creation.
At 14:00 LT (17:00 UTC), the vertical export of precursors is even more important due to the maximum development of the deep mixing layer and the full intensity of upslope winds, with two main consequences. First, NO x levels in Santiago decrease In addition to being a pollutant, tropospheric O 3 is a strong greenhouse gas, estimated to contribute between 0.2 to 0.6 W m −2 to present climate global radiative forcing (Myhre et al., 2013). The process described above may then imply radiative effects, either directly through greenhouse effect, indirectly through its impact on moisture, clouds and atmospheric circulation, or 390 through its interaction with the cycles of other greenhouse gases (Mohnen et al., 1993). Estimating this impact is however not in the scope of the present work. Nevertheless, it shows that although the venting of pollutants can be beneficial from the urban air pollution perspective, longer-term effects on climate are a corollary worth investigating.

Discussion
Only one particular month per season, for one particular year, are analyzed here. Whether the results presented can be 395 extrapolated with a climatological relevance is not straightforward. Climate variability modes such as ENSO, for instance, can lead to inter-annual changes in circulation and meteorology at the scale of central Chile. However, although our quantitative conclusions may not exactly hold for other years, the underlying processes described remain valid, particularly when it comes to mountain-valley circulation which is radiatively driven. Variations in primary pollutants and precursors emissions, wintertime precipitation and cloud cover may affect our results, but we do not expect new processes to take place or evidenced processes 400 to stop, nor magnitudes to change entirely. Santiago and Los Andes) in early afternoon, measured between 1.5 km and 2.5 km above ground. Such secondary layers higher up are consistent with our findings. If a vertical profile is taken north of Santiago in Fig 9d, a bell shaped profile is recovered, of maximum intensity near 2 km above ground, much like Figure 9 in Seguel et al. (2013). Thus, our simulation results agree very well with the measurements conducted in Seguel et al. (2013), despite being for a different year, hence strengthening our conclusions on the existence of the newly evidenced O 3 bubble and its seasonal persistence. Similarly to our findings, 420 Seguel et al. (2013) also concluded that the residual layer is coming from pollutants venting from Santiago. However their measurement-based approach did not allow to evidence (i) the exact formation mechanism of this bubble, (ii) its persistent character even through nighttime (measurements presented are only for daytime), (iii) its horizontal extent (measurements are discrete in space). Our modeling approach based on sensitivity analysis confirms the primary role of Santiago emissions and gives a clearer and continuous 3-dimensional picture of the phenomenon, while agreeing with the findings of these previous 425 measurements.
We estimate that 14% of PM 2.5 in Santiago come from long-range transport in wintertime. Although summertime PM 2.5 is not discussed within the framework of this paper, it is available in the simulations, and we find its transported contribution to be greater, at 22%, for that period. As a result, we expect the average contribution of long-range transport to PM 2.5 in Santiago for a whole year to be somewhere between these two numbers, around 18%. It is worth noting that for the year 2019, the 430 exceedance of the PM 2.5 Chilean standard was 18% for Santiago according to data from the SINCA network. Although for our study year 2015 this exceedance was higher (close to 50%), our findings suggest that if air quality improvement policies are conducted in all urban areas across central Chile, Santiago might be able to meet the national standards more easily due to the large contribution of transported PM 2.5 .

435
Based on chemistry-transport modeling with WRF-CHIMERE, the present work investigates the transport of atmospheric pollutants in central Chile for one winter month and one summer month. Our findings show that emissions of the Santiago Metropolitan Area greatly affect atmospheric composition in its vicinity and farther, with a contribution of a few µg m −3 to PM 2.5 in wintertime corresponding to 5% to 10% of surface concentrations as far as 4 • north and 4 • south. This transport is mostly driven by surface winds within the boundary layer above land, and takes place in the free troposphere over the ocean if emissions of precursors were to be decreased in the city, its western districts would see their O 3 mixing ratios increase on average, despite daily peaks largely dropping, while the eastern area would improve for every quantile. This phenomenon is linked to heterogeneous emissions within the city, which make for currently higher (lower, respectively) than background levels in the east (the west, respectively). When all emissions are cut, the whole area is brought to background, hence the respective variations. The vertical export of precursors above Santiago in summertime, in relation with unperturbed mountain-450 valley circulation and venting, creates a persistent O 3 bubble of more than 50 ppb on average, around 1000 m above the ground, slightly north of the city. Daytime upslope winds, the heterogeneous lifetimes of precursors, and increasing vertical profiles of photolysis rate account for this formation, which impact should be looked at in greater detail in terms of surface air quality improvement (or worsening), and photo-chemical and greenhouse effects. Dudhia  Table A2. Simulation scores for daily average low-level meteorology for wintertime and summertime 2015. T2 is the 2 m air temperature ( • C), RH the surface relative humidity, U10 10 m zonal wind and V10 10 m meridional wind speed (m s −1 ). MB is the mean bias, NRMSE the normalized root mean square error and R the Pearson correlation coefficient.  Table A3. Simulation scores for meteorological vertical profiles for four days in July 2015 at noon local time at Quinta Normal station in Santiago. The same notations as in Table A2 apply.